Deep Learning-based Classification of Faba Bean Leaf Diseases using DenseNet169

V
Varsha Bodade1,*
S
Shubhangi Sanjay Kadu2
K
Kausar Maner3
P
E
Er. Rihana Jamadar5
A
Amol Sankpal5
1Department of Electronics and Communication (Advance Communication Technology), Shah and Anchor Kutchhi Engineering College, Chembur, Mumbai-400 088, Maharashtra, India.
2Department of Humanities and Applied Sciences, SIES Graduate School of Technology (Autonomous) Nerul, Navi Mumbai-400 706, Maharashtra, India.
3Department of Artificial Intelligence and Data Science, Shah and Anchor Kutchhi Engineering College, Chembur, Mumbai-400 088, Maharashtra, India.
4Department of Computer Engineering, Lokmanya Tilak College of Engineering, Navi Mumbai-400 709, Maharashtra, India.
5Department of Electronics and Telecommunication Engineering, M.H. Saboo Siddik College of Engineering, Mumbai-400 708, Maharashtra, India.

Background: Faba bean is a nutrient-rich legume that plays a vital role in global food security and soil fertility. However, foliar diseases such as chocolate spot, rust and gall significantly impact its yield and quality. Early and accurate disease identification is essential to reduce these losses, particularly under field conditions where manual diagnosis is time-consuming and prone to error.

Methods: This study introduces a deep learning-based model using DenseNet169 for the multi-class classification of faba bean leaf diseases. The model was trained and validated on a dataset of 8,021 expert-labeled images captured in natural field environments. Image augmentation and preprocessing were applied to improve learning. The classification was performed across four categories.

Result: The proposed model achieved a classification accuracy of 95.45%. It showed strong classification matrices across all classes. Comparative analysis with existing CNN models confirmed its competitive performance. The DenseNet169 architecture demonstrated robustness and practical applicability for disease diagnosis in legume cultivation.

Agriculture plays a central role in economic growth and directly influences a country’s gross domestic product. Crop performance, especially in terms of yield and quality, is strongly impacted by environmental conditions and farming practices. Legumes are vital in this context because they serve as an affordable source of plant-based protein and contribute to soil fertility through nitrogen fixation (Akchaya et al., 2025; Mukherjee et al., 2025).
       
Among legumes, faba bean (Vicia faba L.) is widely grown in many parts of the world. It is a staple in regions like North Africa, West Asia and parts of Europe. In 2023, the global production of faba bean was estimated at over 5.8 million tonnes, according to FAOSTAT (FAO, 2024). In India, legumes are important to the dietary needs of a largely vegetarian population. Pulses contribute over 20% of total protein intake for Indian consumers (ICAR, 2023). Although faba bean is not among the top cultivated pulses in India, its cultivation is gradually expanding in northern states such as Uttar Pradesh, Bihar and Himachal Pradesh due to its climate adaptability and soil enrichment properties (Singh et al., 2013; Dhull et al., 2021; Maalouf et al., 2021). India is also one of the largest global producers and consumers of pulses, with a total pulse production of 27.5 million tonnes in 2022-2023 (Ministry of Agriculture and Farmers Welfare, 2023).
       
However, legume production is threatened by a range of foliar diseases. In faba bean, key foliar diseases comprise chocolate spot (Botrytis fabae) and rust (Uromyces viciae-fabae), alongside gall disease and other fungal and viral infections that significantly impair plant health and yield stability. These diseases can lead to substantial yield losses under favorable environmental conditions (Tiwari et al., 2023; Sbai et al., 2024). Early detection of such diseases is important for managing outbreaks and minimizing losses. Traditional disease diagnosis methods rely on manual inspection, which is time-consuming and often inaccurate without expert knowledge (Zeleke and Legesse, 2023; Desfita et al., 2025). Conventional management of crop diseases often relies on chemical pesticides, but their excessive or inappropriate use can contribute to pesticide pollution and create adverse social and ecological consequences (Sriram and Kumari, 2026). Therefore, sustainable and targeted approaches to disease management are increasingly important (Arora et al., 2025; Souza et al., 2026).
       
Precision and smart farming technologies can improve agricultural resource management by enabling more efficient use of water, fertilizers and other farm inputs. These approaches also support better monitoring, planning and decision-making, thereby contributing to improved agricultural productivity and sustainable farming practices (Geetanjali and Singh, 2025).
       
Recent advances in artificial intelligence, particularly deep learning, offer effective solutions to this problem. Convolutional Neural Networks (CNNs) have shown superior performance in plant disease detection tasks under field conditions (Leite et al., 2024; Ibrahim et al., 2025). Deep learning-based CNN architectures have demonstrated considerable potential for automated agricultural disease detection and classification, particularly under real-field conditions where variations in disease symptoms and image backgrounds can affect model performance (Nirgude and Rathi, 2021). CNNs automatically extract and learn features from image data, allowing for rapid and scalable disease classification. From 2020 to 2025, numerous studies have demonstrated the accuracy of CNN models such as ResNet, EfficientNet and DenseNet for plant leaf disease classification, with accuracy levels often exceeding 90% (Atila et al., 2020; Alam et al., 2024; Dheeraj and Chand, 2024; Upadhyay et al., 2025; Baiju et al., 2025). Salau et al., (2023) proposed an end-to-end convolutional neural network (CNN) model for detecting faba bean diseases. The study evaluates the model’s performance using both raw and preprocessed image datasets. The raw image-trained CNN achieved 92.1% accuracy, while the preprocessed dataset model reached 98.14% accuracy. Jeong and Na (2024) developed a deep CNN model to identify major faba bean leaf diseases, including Chocolate Spot, Rust, Gall and Healthy categories. Their model, composed of 8 convolutional layers, achieved 99.37% training accuracy and 89.69% validation accuracy after 75 epochs. Mostafa et al., (2025) developed a sequential CNN model for detecting leaf diseases in faba beans. The model was trained and tested in a Jupyter Notebook environment using a dataset collected with expert assistance. Preprocessing techniques were applied to improve model performance. The CNN achieved an overall accuracy of 98.92%.
       
While these studies demonstrate the feasibility of deep learning for faba bean disease detection, there remains a need for studies that integrate field-acquired image datasets with systematic benchmarking of advanced transfer-learning architectures. In particular, DenseNet169, known for its dense connectivity pattern that promotes feature reuse and reduces vanishing gradient problems, has shown strong performance in general computer vision tasks and in other crop disease applications, yet its evaluation on documented, real-world faba bean field datasets remains limited.
               
To address this need, the present study evaluates the DenseNet169 architecture on an expert-annotated dataset of 8,021 faba bean leaf images captured under natural field conditions. The study emphasizes transparent dataset documentation, standardized preprocessing and systematic model evaluation. The primary objective is to classify healthy and diseased leaves to support early diagnosis and enable scalable decision-support systems for smart legume agriculture.
Tools and environment
 
All experiments were conducted on a Windows 10 system with an Intel® Core™ i5-11320H CPU @ 3.20 GHz and 16 GB RAM. Python 3.11, TensorFlow 2.x and Jupyter Notebook were used for implementation. The following libraries were used: TensorFlow, NumPy, Pandas, Seaborn, Matplotlib and Scikit-learn. To ensure reproducibility, a fixed random seed (seed = 42) was set for NumPy, TensorFlow and Python’s random module. All experiments were conducted under identical initialization settings.
 
Dataset description
 
This study focuses on the classification of faba bean leaf diseases using a deep learning approach. The dataset consists of 8,021 RGB images collected from open-field cultivation plots in Uttar Pradesh, India, during the Rabi growing season (2022-2023). Images were collected from multiple independent agricultural plots within the same agro-climatic region to capture natural variability in soil, crop management practices and background complexity. Image acquisition was performed using a 12-megapixel smartphone camera under natural daylight conditions without artificial illumination. Photographs were taken at an approximate distance of 30-50 cm from the leaf surface, under varying illumination intensities and background conditions, without controlled laboratory settings. All images were annotated with the assistance of agricultural experts specializing in plant pathology. The dataset includes 8,021 RGB images divided into four categories: healthy (2,019 images), rust-infected (2,000 images), gall-infected (2,000 images) and chocolate spot-infected (2,002 images). Fig 1 provides representative images of these four classes. All images were resized to 224 × 224 pixels before processing. A stratified split was applied to divide the data into training, validation and test sets. The split ratio was 60% training, 20% validation and 20% testing (4,812 / 1,604 / 1,605 images). The split was performed at the image level after random shuffling using the fixed seed.

Fig 1: Representative RGB images of faba bean leaves from each class: (a) Healthy, (b) Rust-infected, (c) Faba bean gall-infected and (d) Chocolate spot-infected.


 
Preprocessing and data augmentation
 
A preprocessing step was applied before sending the images to the model. The preprocess_input function from TensorFlow’s DenseNet module was used. It normalized the pixel values in a way that was compatible with the pretrained DenseNet model. To enhance model robustness and reduce overfitting, data augmentation was applied exclusively to the training dataset. The augmentation pipeline consisted of random horizontal flipping with a probability of 0.5, random rotation within ±10 degrees and random zoom within the range of 0.9-1.1 (±10%). These transformations were implemented using TensorFlow’s Sequential preprocessing layers and applied dynamically in real time during model training.
 
Model architecture
 
The deep learning model developed for this study was based on the DenseNet169 architecture, a type of convolutional neural network (CNN) that enables feature reuse and efficient gradient flow by connecting each layer to every other layer in a feed-forward fashion (Fig 2). DenseNet169 includes several major components.

Fig 2. Model architecture of DenseNet-169.


 
Input layer
 
This layer accepts images of size 224 × 224 × 3. Each image is an RGB image with three color channels. The input tensor can be represented as:
 
 

Initial convolution and pooling layers
 
A 7 × 7 convolution is applied with a stride of 2, followed by batch normalization, ReLU activation and a 3 × 3 max pooling operation. The output of this block can be written as:
 
Y1 = MaxPool{ReLU[BN(W1 * X + b1)]}
 
Where,
W1 and b1 = Weights and biases of the first convolution layer.
 
Dense blocks and transition layers
 
DenseNet169 contains four dense blocks with 6, 12, 32 and 32 convolutional layers, respectively. Inside each dense block, each layer obtains input from all preceding layers:
 
x1 = H1{[x0,x1,...,x_(l-1)]}
 
Here,
xl =  Output of the lth layer.
Hl = A composite function of batch normalization, ReLU, 1 × 1 convolution and 3 × 3 convolution.
       
Between dense blocks, transition layers reduce the feature map size via 1 × 1 convolutions and 2 × 2 average pooling.
 
Global average pooling (GAP) layer
 
The output of the final dense block is fed into a global average pooling layer to reduce spatial dimensions:
 
 

Here,
fi, j = Feature map at position (i, j) and GAP averages across all positions.
 
Fully connected (Dense) layer
 
The final classification layer is a fully connected layer with 4 neurons, corresponding to the four disease classes. It uses a softmax activation to output the probability distribution over the classes:
 
ŷ = Softmax (W · z + b)
 
ŷ = Vector of class probabilities.
W = Weight matrix.
b = Bias vector.
 
Output layer
 
The model predicts the class label with the highest probability.
 
Predicted class= argmax(ŷ).
       
The base DenseNet169 model was initialized with ImageNet weights and frozen during training to retain previously learned features. Only the newly added classification head was trained. This transfer learning approach reduced training time and prevented overfitting.
 
Model compilation and training
 
The model was compiled using the Adam optimizer with a learning rate of 5×10-5. The categorical cross-entropy loss function was used since the task involved multi-class classification. The main metric used during training and evaluation was accuracy. Model training was performed for 50 epochs with a batch size of 32. The base DenseNet169 model was initialized with ImageNet pretrained weights. All convolutional layers were frozen during training and only the newly added classification head was trainable. Two callbacks were used to control the training process. The ModelCheckpoint callback saved the best model based on validation accuracy. The EarlyStopping callback stopped training if the model showed no improvement for 5 consecutive epochs. Both callbacks helped in preventing overfitting and saving training time.
 
Model evaluation
 
The model was evaluated on the test set using several metrics. These included overall accuracy, confusion matrix, classification report, Matthews Correlation Coefficient (MCC), AUC-ROC and Precision-Recall curves. The accuracy metric was used to measure the percentage of correct predictions. The confusion matrix showed the number of true positives, true negatives, false positives and false negatives for each class. MCC was also computed to give a balanced measure of prediction quality. It is especially useful for multi-class classification. MCC is calculated using the formula:
 
 
       
ROC curves were plotted for each class using the One-vs-Rest strategy. The area under the curve (AUC) was computed for each class. A higher AUC indicates better model performance. Precision-recall (PR) curves were also plotted. They help in analyzing performance when classes are imbalanced. The average precision (AP) score was calculated for each class. To interpret model confidence, the top predictions with the highest softmax scores were identified.
Training and validation performance
 
The training of the DenseNet169 model was conducted for a maximum of 50 epochs with early stopping based on validation accuracy. Initially, during the first epoch, the training accuracy was low, around 32.48% and the validation accuracy reached 54.43%, indicating that the model had begun learning some useful features. The training loss was 1.4304 and the validation loss was 1.1117, showing the model’s initial difficulty in class separation.
       
As the training progressed, the model improved steadily. By epoch 25, the training accuracy had reached 95.05% with a reduced training loss of 0.2013, while the validation accuracy increased to 95.14%. However, no significant improvement in validation accuracy was observed after that point, suggesting that the model had begun to plateau. The best model checkpoint was obtained at epoch 39, where the validation accuracy reached 96.45%. At this stage, the training accuracy was 95.85%, with a training loss of 0.1504, while the validation loss decreased to 0.1175. The small difference between training and validation performance indicates stable convergence without noticeable overfitting. As no further improvement was observed, early stopping was triggered to preserve the best-performing model. Table 1 presents the training log of the DenseNet169 model near the convergence stage.

Table 1: Training log near convergence of DenseNet169.


 
Test performance evaluation
 
The final evaluation was conducted on the unseen test set, where the model achieved an overall accuracy of 95.45% and a test loss of 0.2054. This high accuracy indicates that the model generalized well and performed reliably across all four classes. As shown in Fig 3, training and validation accuracy increased steadily and both losses decreased consistently, with minor fluctuations near convergence. The gap between the training and validation curves was minimal, suggesting that the model did not overfit.

Fig 3: Epoch-wise training and validation accuracy and loss curves.


 
Class-wise confusion matrix analysis
 
The confusion matrix results of the DenseNet169 model are presented in Fig 4a, while the normalized confusion matrix is shown in Fig 4b. These figures provide detailed insights into the class-wise classification performance of the model on the test dataset. As illustrated in Fig 4a, the Chocolate spot class achieved 354 correct predictions out of 401 samples, while 22 samples were misclassified as Gall, 3 as Healthy and 22 as Rust. For the Gall class, 384 out of 400 samples were correctly classified, with 11 samples misclassified as Chocolate spot, 3 as Healthy and 2 as Rust. The Healthy class showed the highest classification performance, with 400 correctly predicted samples out of 404, while 1 sample was misclassified as Chocolate spot, 2 as Gall and 1 as Rust. Similarly, the Rust class achieved 394 correct predictions out of 400 samples, with 4 samples misclassified as Chocolate spot and 2 as Gall. To provide a clearer interpretation of the class-wise performance, a normalized confusion matrix was generated as shown in Fig 4b, where the classification results are represented as percentages. The normalized matrix indicates that the Healthy class achieved the highest recall (99.01%), followed by Rust (98.50%), Gall (96.00%) and Chocolate spot (88.28%). Overall, the strong concentration of values along the diagonal elements in both matrices indicates that the DenseNet169 model achieved high classification accuracy across all disease categories.

Fig 4: Confusion matrix analysis of the DenseNet169 model: (a) raw confusion matrix showing the number of samples predicted for each class and (b) normalized confusion matrix showing percentage-based classification performance across the four disease categories.



Classification report and metrics
 
The detailed classification report is presented in Table 2. The Chocolate spot class achieved a precision of 95.68%. The recall was 88.28%, indicating that some images were not identified correctly.

Table 2: Classification report metrics for each class.


       
This resulted in an F1-score of 91.83%. The Gall class showed consistent performance with 93.66% precision and 96.00% recall. The F1-score of 94.81% suggests strong balance between sensitivity and specificity. For the Healthy class, the model performed almost perfectly. It achieved 98.52% precision, 99.01% recall and an F1-score of 98.77%. These metrics reflect the model’s ability to identify healthy leaves correctly. The Rust class also had strong results, with 94.03% precision and 98.50% recall, producing an F1-score of 96.21%. The overall accuracy of the model was 95.45% on the test dataset. The macro and weighted averages for precision, recall and F1-score were all around 95.4-95.5%, indicating consistent performance across all classes. The Matthews Correlation Coefficient (MCC) was calculated as 0.9397, showing a high level of balanced classification.
 
ROC curve analysis
 
The ROC curves for each class are displayed in Fig 5. The Chocolate spot class achieved an AUC of 0.994, indicating very high separability from other classes. The Gall class recorded an AUC of 0.997, also showing excellent performance. The healthy class reached the highest possible AUC of 1.000, confirming perfect class distinction. The Rust class followed closely with an AUC of 0.998. These results support the robustness of the model in distinguishing between all four categories.

Fig 5: ROC curves for all classes.


 
Confident prediction visualization
 
The most confident predictions made by the model are presented in Fig 6. These samples represent the top three predictions for each class based on the highest confidence scores. The Chocolate spot images were predicted with confidence values of 0.9996–0.9998, showing clear dark lesions on green leaf surfaces. The Gall class images were all predicted with full confidence (1.0000), with visible swelling and distortion of leaf structures. The Healthy class predictions also showed high confidence (0.9993-0.9996) and displayed smooth, undamaged leaf texture and uniform green coloration. Finally, Rust predictions were made with high confidence (0.9999-1.0000) and showed characteristic orange to brown pustules.

Fig 6: Top three most confident predictions from each class.


       
These confident predictions visually confirm the model’s capability to identify key features with a high degree of certainty. The confidence scores demonstrate that the model is not only accurate but also reliable in recognizing disease symptoms.
       
Table 3 summarizes representative failure cases observed during model evaluation. Most misclassifications occur between visually similar diseases such as Chocolate spot, Rust and Gall. The confusion is primarily caused by overlapping lesion patterns, variations in lesion density and differences in lighting conditions in field images.

Table 3: Summary of representative failure cases observed in the model predictions.


 
Precision-recall curve analysis
 
The precision-recall curves shown in Fig 7 complement the ROC analysis by highlighting how well the model performs under class imbalance. The Chocolate spot class achieved an average precision (AP) of 0.983, while the Gall class recorded 0.992. The Healthy class once again showed optimal performance with an AP of 1.000 and the Rust class followed with 0.995. The curves confirm that the model maintains a strong balance between precision and recall across all disease categories.

Fig 7: Precision-Recall curves for all four classes.


       
A comparative overview of recent CNN-based approaches for legume disease detection is presented in Table 4. These studies report varying performance levels depending on factors such as the CNN architecture, crop type, dataset size, image acquisition conditions and preprocessing strategies. Since these studies were conducted using different datasets and experimental protocols, the reported results should be interpreted with caution and do not represent a direct or strictly comparable benchmark. For instance, DenseNet121 was used by Abed et al., (2021) for detecting diseases in bean leaves, achieving 98.31% accuracy using manually collected images. Similarly, VGG16 was applied by Sahu et al., (2021) to a dataset of 1,296 bean leaf images, resulting in 95.31% accuracy. In another study, Joshi et al., (2021) proposed the VirLeafNet-1 model for Vigna mungo disease detection and reported 91.23% accuracy using manually collected data. More recent research has explored deeper architectures. Singh et al., (2023) utilized EfficientNetB6 with a dataset of 1,295 bean leaf images, achieving 91.74% accuracy, while Serttas and Deniz (2023) applied ResNet50 on a similar dataset size and reported 98.33% accuracy. In contrast, Kursun et al., (2023) used DenseNet201 on a small dataset of 60 dry bean images, achieving 100% accuracy, although the limited dataset size may restrict the generalizability of the results. For faba bean disease detection, Salau et al., (2023) proposed an end-to-end CNN model using both raw and preprocessed images and achieved 98.14% accuracy. Jeong and Na (2024) developed a custom CNN architecture with eight convolutional layers and reported 91.00% accuracy. More recently, Mostafa et al., (2025) trained a sequential CNN on expert-annotated and preprocessed data, achieving 98.92% accuracy.

Table 4: Overview of CNN-based approaches reported in previous studies for legume disease detection.


       
In the present study, the DenseNet169 model was applied to a large-scale dataset of 8,021 expert-labeled faba bean leaf images collected under natural field conditions, achieving an overall classification accuracy of 95.45%. Although direct comparison with previous studies is challenging due to differences in datasets and experimental settings, the results demonstrate that the proposed approach provides reliable and robust performance for practical field-based disease detection scenarios.
This study presented a CNN-based classification model using DenseNet169 for detecting major faba bean leaf diseases. The model achieved 95.45% accuracy on a large, field-based dataset. Results confirm the effectiveness of transfer learning in plant disease detection. Limitations include reliance on image data only, which may not capture environmental factors. The dataset, though large, was collected from limited regions, which may affect generalization. Also, real-time deployment was not tested in this study. Future work will focus on expanding the dataset across regions and seasons. Integrating weather data and soil conditions may improve model accuracy. Real-time mobile or drone-based implementations can help validate the model’s field performance. Combining CNNs with hybrid models could further boost detection and classification efficiency.
Disclaimers
 
The views and conclusions expressed in this article are solely those of the authors and do not necessarily represent the views of their affiliated institutions. The authors are responsible for the accuracy and completeness of the information provided, but do not accept any liability for any direct or indirect losses resulting from the use of this content.
 
Funding details
 
This research received no external funding.
 
Authors’ contributions
 
All authors contributed toward data analysis, drafting and revising the paper and agreed to be responsible for all aspects of this work.
 
Data availability
 
The dataset analyzed in this study consists of 8,021 expert-labeled faba bean leaf images collected under natural field conditions. Due to institutional and field data collection policies, the dataset is not publicly available. However, the data supporting the findings of this study may be made available by the corresponding author for academic and non-commercial research purposes upon reasonable request.
 
Availability of data and materials
 
Not applicable.
 
Use of artificial intelligence
 
Not applicable.
 
Declarations
 
Authors declare that all works are original and this manuscript has not been published in any other journal.
The authors declare that they have no conflict of interest.

  1. Abed, S.H., Al-Waisy, A.S., Mohammed, H.J. and Al-Fahdawi, S. (2021). A modern deep learning framework in robot vision for automated bean leaves diseases detection. International Journal of Intelligent Robotics and Applications. 5(2): 235-251.

  2. Akchaya, K., Parasuraman, P., Pandian, K., Vijayakumar, S., Thirukumaran, K., Mustaffa, M. R.a.F., Rajpoot, S.K. and Choudhary, A.K. (2025). Boosting resource use efficiency, soil fertility, food security, ecosystem services and climate resilience with legume intercropping: A review. Frontiers in Sustainable Food Systems. 9. https:// doi.org/10.3389/fsufs.2025.1527256.

  3. Alam, T.S., Jowthi, C.B. and Pathak, A. (2024). Comparing pre- trained models for efficient leaf disease detection: A study on custom CNN. Journal of Electrical Systems and Information Technology. 11(1). https://doi.org/10.118 6/s43067-024-00137-1.

  4. Arora V., Kanwal A., Sharma I. and Singh R. (2025). Pesticides pollution in agroecosystems and their social and ecological  consequences. Bulletin of Pure and Applied Sciences- Botany. 44B(1): 60-64. https://bpasjournals.com/botany/ index.php/journal/article/view/141/133.

  5. Atila, Ü., Uçar, M., Akyol, K. and Uçar, E. (2020). Plant leaf disease classification using EfficientNet deep learning model. Ecological Informatics. 61: 101182. https://doi.org/10.10 16/j.ecoinf.2020.101182.

  6. Baiju, B.V., Kirupanithi, N., Srinivasan, S., Kapoor, A., Mathivanan, S.K. and Shah, M.A. (2025). Robust CRW crops leaf disease detection and classification in agriculture using hybrid deep learning models. Plant Methods. 21(1). https://doi.org/10.1186/s13007-025-01332-5

  7. Desfita, S., Sari, W., Wahyuni, D., Putri, F., Pramono, A., Pato, U., Pratiwi, D., Grzelczyk, J. and Budryn, G. (2025). Synergistic effects of multi-strain probiotic and prebiotic combinations on immune recovery in aging populations. International Journal of Probiotics and Prebiotics. 20(1): 10-18. https://doi.org/10.37290/ijpp2641-7197.20:10-18.

  8. Dheeraj, A. and Chand, S. (2024). LWDN: lightweight DenseNet model for plant disease diagnosis. Journal of Plant Diseases and Protection. 131(3): 1043-1059. https:// doi.org/10.1007/s41348-024-00915-z.

  9. Dhull, S.B., Kidwai, M.K., Noor, R., Chawla, P. and Rose, P.K. (2021). A review of nutritional profile and processing of faba bean (Vicia faba L.). Legume Science. 4(3). https:// doi.org/10.1002/leg3.129.

  10. FAO. (2024). FAOSTAT Crop Data: Faba Bean Production. Food and Agriculture Organization of the United Nations. https:// www.fao.org/faostat.

  11. Geetanjali and Singh, A.R. (2025). Precision and smart farming: An approach to efficient resource management in agriculture. Journal of Artificial Intelligence in Agriculture and Applied Research. 1(1): 25-33. https://agrpjournals.com/ journals/journal/index.php/jaiaar/article/view/6/5.

  12. Ibrahim, H., Ahmed, A.M., Bouallegue, B., Khattab, M.M., El-Fattah, M.A. and El-Mawla, N.A. (2025). Deep convolutional neural networks for plant disease detection: A mobile application approach (Agri Bot). International Journal of Intelligent Systems. (1). https://doi.org/10.1155/int/ 7644407.

  13. ICAR. (2023). Indian Council of Agricultural Research Annual Report 2022-23. https://icar.org.in

  14. Jeong, H.Y. and Na, I.S. (2024). Efficient faba bean leaf disease identification through smart detection using deep convolutional neural networks. Legume Research - An International Journal. doi: 10.18805/LRF-798.

  15. Joshi, R.C., Kaushik, M., Dutta, M.K., Srivastava, A. and Choudhary, N. (2021). VirLeafNet: Automatic analysis and viral disease diagnosis using deep-learning in Vigna mungo plant. Ecological Informatics. 61: 101197.

  16. Kursun, R., Bastas, K.K. and Koklu, M. (2023). Segmentation of dry bean (Phaseolus vulgaris L.) leaf disease images with U-Net and classification using deep learning algorithms. European Food Research and Technology.  249(10): 1- 16.

  17. Leite, D., Brito, A. and Faccioli, G. (2024). Advancements and outlooks in utilizing Convolutional Neural Networks for plant disease severity assessment: A comprehensive review. Smart Agricultural Technology. 9: 100573. https://doi.org/10.1016/j.atech.2024.100573

  18. Maalouf, F., Ahmed, S. and Bishaw, Z. (2021). Faba Bean. In: Elsevier eBooks. (pp. 105-131). https://doi.org/10.1016/ b978-0-12-821450-3.00008-1.

  19. Ministry of Agriculture and Farmers Welfare. (2023). Annual Report 2022-2023. Government of India. https://agricoop.nic.in

  20. Mostafa, A., Alnuaim, A. and AlZubi, A.A. (2025). Utilizing convolutional neural networks for accurate detection of leaf diseases in fava beans. Legume Research: An International Journal. 48(3): 494-502. doi: 10.18805/LRF-823.

  21. Mukherjee, B., Naskar, M.K., Chakraborty, S., Dutta, S. and Mandal, T.K. (2025). Building Food Security and Climate Resilience: The Promise of Climate Smart Pulse Production Technologies. In: Sustainability Sciences in Asia and Africa (pp. 17- 43). https://doi.org/10.1007/978-981-96-7699-6_2.

  22. Nirgude V. and Rathi S. (2021). A robust deep learning approach to enhance the accuracy of pomegranate fruit disease detection under real field condition. Journal of Experimental Biology and Agricultural Sciences. 9(6): 863-870. https:// doi.org/10.18006/2021.9(6).863.870.

  23. Sahu, P., Chug, A., Singh, A.P., Singh, D. and Singh, R.P. (2021). Deep learning models for beans crop diseases: Classification and visualization techniques. International Journal of Modern Agriculture. 10(1): 796-812.

  24. Salau, A.O., Abeje, B.T., Faisal, A.N. and Asfaw, T.T. (2023). Faba Bean Disease Detection using Deep Learning Techniques. In 2023 International Conference on Cyber Management and Engineering (CyMaEn) (pp. 344-349). IEEE. https:// doi.org/10.1109/CyMaEn57228.2023.10051088.

  25. Sbai, H., Hajib, A., Msairi, S., Amalich, S., Bouyahya, A., Lee, L., Goh, K.W., Tabyaoui, M. and Harhar, H. (2024). Fungal infections of legume crops: Challenges and management approaches. Journal of Agriculture and Food Research. 18: 101447. https://doi.org/10.1016/j.jafr.2024.101447.

  26. Serttas, S. and Deniz, E. (2023). Disease detection in bean leaves using deep learning. Communications Faculty of Sciences University of Ankara Series A2-A3 Physical Sciences and Engineering. 65(2): 115-129.

  27. Singh, A.K., Bharati, R.C., Ch, N., Manibhushan, R. and Pedpati, A. (2013). An assessment of faba bean (Vicia faba L.) current status and future prospect. African Journal of Agricultural Research. 8(50): 6634-6641. https://doi.org/ 10.5897/ajar2013.7335.

  28. Singh, V., Chug, A. and Singh, A.P. (2023). Classification of beans leaf diseases using fine tuned CNN model. Procedia Computer Science. 218: 348-356.

  29. Souza, K.F., Gonçalves, L.C.O. and Espindola, F.S. (2026). A broad overview of the multifaceted royal jelly and its applicability in health and performance. Current Topics in Nutraceutical Research. 24(1): 4-14. https://doi.org/10.37290/ctnr.v24i1.8.

  30. Sriram, P. and Kumari, Y.S. (2026). Game theory in urban birds: A Nash equilibrium analysis using crows and pigeons. Bio- Science Research Bulletin. 42(1): 1-6. https://bpas journals.com/life-sciences/index.php/journal/article/ view/150/140.

  31. Tiwari, N., Barpete, S., Kumar, T., Saxena, D.R. and Maalouf, F. (2023). Disease resistance: An essential for better adaptability and production of faba bean in India (Vicia faba L.). In: Diseases in Legume Crops (Chapter 8). [U.C. Jha, H. Nayyar, K.D. Sharma, E.J.B. von Wettberg, P. Singh and K. H. Siddique (Eds.)], Springer. https:// doi.org/10.1007/978-981-99-3358-7_8.

  32. Upadhyay, A., Chandel, N.S., Singh, K.P., Chakraborty, S.K., Nandede, B.M., Kumar, M., Subeesh, A., Upendar, K., Salem, A. and Elbeltagi, A. (2025). Deep learning and computer vision in plant disease detection: A comprehensive review of techniques, models and trends in precision agriculture. Artificial Intelligence Review. 58(3). https:// doi.org/10.1007/s10462-024-11100-x.

  33. Zeleke, E.A. and Legesse, N.H. (2023). Management of chocolate spot of faba bean (Botrytis fabae) through fungicides and host resistance in North Shewa, Central Highland of Ethiopia. Indian Phytopathology. 76(2): 531-540. https://doi.org/10.1007/s42360-023-00636-7.
 

Deep Learning-based Classification of Faba Bean Leaf Diseases using DenseNet169

V
Varsha Bodade1,*
S
Shubhangi Sanjay Kadu2
K
Kausar Maner3
P
E
Er. Rihana Jamadar5
A
Amol Sankpal5
1Department of Electronics and Communication (Advance Communication Technology), Shah and Anchor Kutchhi Engineering College, Chembur, Mumbai-400 088, Maharashtra, India.
2Department of Humanities and Applied Sciences, SIES Graduate School of Technology (Autonomous) Nerul, Navi Mumbai-400 706, Maharashtra, India.
3Department of Artificial Intelligence and Data Science, Shah and Anchor Kutchhi Engineering College, Chembur, Mumbai-400 088, Maharashtra, India.
4Department of Computer Engineering, Lokmanya Tilak College of Engineering, Navi Mumbai-400 709, Maharashtra, India.
5Department of Electronics and Telecommunication Engineering, M.H. Saboo Siddik College of Engineering, Mumbai-400 708, Maharashtra, India.

Background: Faba bean is a nutrient-rich legume that plays a vital role in global food security and soil fertility. However, foliar diseases such as chocolate spot, rust and gall significantly impact its yield and quality. Early and accurate disease identification is essential to reduce these losses, particularly under field conditions where manual diagnosis is time-consuming and prone to error.

Methods: This study introduces a deep learning-based model using DenseNet169 for the multi-class classification of faba bean leaf diseases. The model was trained and validated on a dataset of 8,021 expert-labeled images captured in natural field environments. Image augmentation and preprocessing were applied to improve learning. The classification was performed across four categories.

Result: The proposed model achieved a classification accuracy of 95.45%. It showed strong classification matrices across all classes. Comparative analysis with existing CNN models confirmed its competitive performance. The DenseNet169 architecture demonstrated robustness and practical applicability for disease diagnosis in legume cultivation.

Agriculture plays a central role in economic growth and directly influences a country’s gross domestic product. Crop performance, especially in terms of yield and quality, is strongly impacted by environmental conditions and farming practices. Legumes are vital in this context because they serve as an affordable source of plant-based protein and contribute to soil fertility through nitrogen fixation (Akchaya et al., 2025; Mukherjee et al., 2025).
       
Among legumes, faba bean (Vicia faba L.) is widely grown in many parts of the world. It is a staple in regions like North Africa, West Asia and parts of Europe. In 2023, the global production of faba bean was estimated at over 5.8 million tonnes, according to FAOSTAT (FAO, 2024). In India, legumes are important to the dietary needs of a largely vegetarian population. Pulses contribute over 20% of total protein intake for Indian consumers (ICAR, 2023). Although faba bean is not among the top cultivated pulses in India, its cultivation is gradually expanding in northern states such as Uttar Pradesh, Bihar and Himachal Pradesh due to its climate adaptability and soil enrichment properties (Singh et al., 2013; Dhull et al., 2021; Maalouf et al., 2021). India is also one of the largest global producers and consumers of pulses, with a total pulse production of 27.5 million tonnes in 2022-2023 (Ministry of Agriculture and Farmers Welfare, 2023).
       
However, legume production is threatened by a range of foliar diseases. In faba bean, key foliar diseases comprise chocolate spot (Botrytis fabae) and rust (Uromyces viciae-fabae), alongside gall disease and other fungal and viral infections that significantly impair plant health and yield stability. These diseases can lead to substantial yield losses under favorable environmental conditions (Tiwari et al., 2023; Sbai et al., 2024). Early detection of such diseases is important for managing outbreaks and minimizing losses. Traditional disease diagnosis methods rely on manual inspection, which is time-consuming and often inaccurate without expert knowledge (Zeleke and Legesse, 2023; Desfita et al., 2025). Conventional management of crop diseases often relies on chemical pesticides, but their excessive or inappropriate use can contribute to pesticide pollution and create adverse social and ecological consequences (Sriram and Kumari, 2026). Therefore, sustainable and targeted approaches to disease management are increasingly important (Arora et al., 2025; Souza et al., 2026).
       
Precision and smart farming technologies can improve agricultural resource management by enabling more efficient use of water, fertilizers and other farm inputs. These approaches also support better monitoring, planning and decision-making, thereby contributing to improved agricultural productivity and sustainable farming practices (Geetanjali and Singh, 2025).
       
Recent advances in artificial intelligence, particularly deep learning, offer effective solutions to this problem. Convolutional Neural Networks (CNNs) have shown superior performance in plant disease detection tasks under field conditions (Leite et al., 2024; Ibrahim et al., 2025). Deep learning-based CNN architectures have demonstrated considerable potential for automated agricultural disease detection and classification, particularly under real-field conditions where variations in disease symptoms and image backgrounds can affect model performance (Nirgude and Rathi, 2021). CNNs automatically extract and learn features from image data, allowing for rapid and scalable disease classification. From 2020 to 2025, numerous studies have demonstrated the accuracy of CNN models such as ResNet, EfficientNet and DenseNet for plant leaf disease classification, with accuracy levels often exceeding 90% (Atila et al., 2020; Alam et al., 2024; Dheeraj and Chand, 2024; Upadhyay et al., 2025; Baiju et al., 2025). Salau et al., (2023) proposed an end-to-end convolutional neural network (CNN) model for detecting faba bean diseases. The study evaluates the model’s performance using both raw and preprocessed image datasets. The raw image-trained CNN achieved 92.1% accuracy, while the preprocessed dataset model reached 98.14% accuracy. Jeong and Na (2024) developed a deep CNN model to identify major faba bean leaf diseases, including Chocolate Spot, Rust, Gall and Healthy categories. Their model, composed of 8 convolutional layers, achieved 99.37% training accuracy and 89.69% validation accuracy after 75 epochs. Mostafa et al., (2025) developed a sequential CNN model for detecting leaf diseases in faba beans. The model was trained and tested in a Jupyter Notebook environment using a dataset collected with expert assistance. Preprocessing techniques were applied to improve model performance. The CNN achieved an overall accuracy of 98.92%.
       
While these studies demonstrate the feasibility of deep learning for faba bean disease detection, there remains a need for studies that integrate field-acquired image datasets with systematic benchmarking of advanced transfer-learning architectures. In particular, DenseNet169, known for its dense connectivity pattern that promotes feature reuse and reduces vanishing gradient problems, has shown strong performance in general computer vision tasks and in other crop disease applications, yet its evaluation on documented, real-world faba bean field datasets remains limited.
               
To address this need, the present study evaluates the DenseNet169 architecture on an expert-annotated dataset of 8,021 faba bean leaf images captured under natural field conditions. The study emphasizes transparent dataset documentation, standardized preprocessing and systematic model evaluation. The primary objective is to classify healthy and diseased leaves to support early diagnosis and enable scalable decision-support systems for smart legume agriculture.
Tools and environment
 
All experiments were conducted on a Windows 10 system with an Intel® Core™ i5-11320H CPU @ 3.20 GHz and 16 GB RAM. Python 3.11, TensorFlow 2.x and Jupyter Notebook were used for implementation. The following libraries were used: TensorFlow, NumPy, Pandas, Seaborn, Matplotlib and Scikit-learn. To ensure reproducibility, a fixed random seed (seed = 42) was set for NumPy, TensorFlow and Python’s random module. All experiments were conducted under identical initialization settings.
 
Dataset description
 
This study focuses on the classification of faba bean leaf diseases using a deep learning approach. The dataset consists of 8,021 RGB images collected from open-field cultivation plots in Uttar Pradesh, India, during the Rabi growing season (2022-2023). Images were collected from multiple independent agricultural plots within the same agro-climatic region to capture natural variability in soil, crop management practices and background complexity. Image acquisition was performed using a 12-megapixel smartphone camera under natural daylight conditions without artificial illumination. Photographs were taken at an approximate distance of 30-50 cm from the leaf surface, under varying illumination intensities and background conditions, without controlled laboratory settings. All images were annotated with the assistance of agricultural experts specializing in plant pathology. The dataset includes 8,021 RGB images divided into four categories: healthy (2,019 images), rust-infected (2,000 images), gall-infected (2,000 images) and chocolate spot-infected (2,002 images). Fig 1 provides representative images of these four classes. All images were resized to 224 × 224 pixels before processing. A stratified split was applied to divide the data into training, validation and test sets. The split ratio was 60% training, 20% validation and 20% testing (4,812 / 1,604 / 1,605 images). The split was performed at the image level after random shuffling using the fixed seed.

Fig 1: Representative RGB images of faba bean leaves from each class: (a) Healthy, (b) Rust-infected, (c) Faba bean gall-infected and (d) Chocolate spot-infected.


 
Preprocessing and data augmentation
 
A preprocessing step was applied before sending the images to the model. The preprocess_input function from TensorFlow’s DenseNet module was used. It normalized the pixel values in a way that was compatible with the pretrained DenseNet model. To enhance model robustness and reduce overfitting, data augmentation was applied exclusively to the training dataset. The augmentation pipeline consisted of random horizontal flipping with a probability of 0.5, random rotation within ±10 degrees and random zoom within the range of 0.9-1.1 (±10%). These transformations were implemented using TensorFlow’s Sequential preprocessing layers and applied dynamically in real time during model training.
 
Model architecture
 
The deep learning model developed for this study was based on the DenseNet169 architecture, a type of convolutional neural network (CNN) that enables feature reuse and efficient gradient flow by connecting each layer to every other layer in a feed-forward fashion (Fig 2). DenseNet169 includes several major components.

Fig 2. Model architecture of DenseNet-169.


 
Input layer
 
This layer accepts images of size 224 × 224 × 3. Each image is an RGB image with three color channels. The input tensor can be represented as:
 
 

Initial convolution and pooling layers
 
A 7 × 7 convolution is applied with a stride of 2, followed by batch normalization, ReLU activation and a 3 × 3 max pooling operation. The output of this block can be written as:
 
Y1 = MaxPool{ReLU[BN(W1 * X + b1)]}
 
Where,
W1 and b1 = Weights and biases of the first convolution layer.
 
Dense blocks and transition layers
 
DenseNet169 contains four dense blocks with 6, 12, 32 and 32 convolutional layers, respectively. Inside each dense block, each layer obtains input from all preceding layers:
 
x1 = H1{[x0,x1,...,x_(l-1)]}
 
Here,
xl =  Output of the lth layer.
Hl = A composite function of batch normalization, ReLU, 1 × 1 convolution and 3 × 3 convolution.
       
Between dense blocks, transition layers reduce the feature map size via 1 × 1 convolutions and 2 × 2 average pooling.
 
Global average pooling (GAP) layer
 
The output of the final dense block is fed into a global average pooling layer to reduce spatial dimensions:
 
 

Here,
fi, j = Feature map at position (i, j) and GAP averages across all positions.
 
Fully connected (Dense) layer
 
The final classification layer is a fully connected layer with 4 neurons, corresponding to the four disease classes. It uses a softmax activation to output the probability distribution over the classes:
 
ŷ = Softmax (W · z + b)
 
ŷ = Vector of class probabilities.
W = Weight matrix.
b = Bias vector.
 
Output layer
 
The model predicts the class label with the highest probability.
 
Predicted class= argmax(ŷ).
       
The base DenseNet169 model was initialized with ImageNet weights and frozen during training to retain previously learned features. Only the newly added classification head was trained. This transfer learning approach reduced training time and prevented overfitting.
 
Model compilation and training
 
The model was compiled using the Adam optimizer with a learning rate of 5×10-5. The categorical cross-entropy loss function was used since the task involved multi-class classification. The main metric used during training and evaluation was accuracy. Model training was performed for 50 epochs with a batch size of 32. The base DenseNet169 model was initialized with ImageNet pretrained weights. All convolutional layers were frozen during training and only the newly added classification head was trainable. Two callbacks were used to control the training process. The ModelCheckpoint callback saved the best model based on validation accuracy. The EarlyStopping callback stopped training if the model showed no improvement for 5 consecutive epochs. Both callbacks helped in preventing overfitting and saving training time.
 
Model evaluation
 
The model was evaluated on the test set using several metrics. These included overall accuracy, confusion matrix, classification report, Matthews Correlation Coefficient (MCC), AUC-ROC and Precision-Recall curves. The accuracy metric was used to measure the percentage of correct predictions. The confusion matrix showed the number of true positives, true negatives, false positives and false negatives for each class. MCC was also computed to give a balanced measure of prediction quality. It is especially useful for multi-class classification. MCC is calculated using the formula:
 
 
       
ROC curves were plotted for each class using the One-vs-Rest strategy. The area under the curve (AUC) was computed for each class. A higher AUC indicates better model performance. Precision-recall (PR) curves were also plotted. They help in analyzing performance when classes are imbalanced. The average precision (AP) score was calculated for each class. To interpret model confidence, the top predictions with the highest softmax scores were identified.
Training and validation performance
 
The training of the DenseNet169 model was conducted for a maximum of 50 epochs with early stopping based on validation accuracy. Initially, during the first epoch, the training accuracy was low, around 32.48% and the validation accuracy reached 54.43%, indicating that the model had begun learning some useful features. The training loss was 1.4304 and the validation loss was 1.1117, showing the model’s initial difficulty in class separation.
       
As the training progressed, the model improved steadily. By epoch 25, the training accuracy had reached 95.05% with a reduced training loss of 0.2013, while the validation accuracy increased to 95.14%. However, no significant improvement in validation accuracy was observed after that point, suggesting that the model had begun to plateau. The best model checkpoint was obtained at epoch 39, where the validation accuracy reached 96.45%. At this stage, the training accuracy was 95.85%, with a training loss of 0.1504, while the validation loss decreased to 0.1175. The small difference between training and validation performance indicates stable convergence without noticeable overfitting. As no further improvement was observed, early stopping was triggered to preserve the best-performing model. Table 1 presents the training log of the DenseNet169 model near the convergence stage.

Table 1: Training log near convergence of DenseNet169.


 
Test performance evaluation
 
The final evaluation was conducted on the unseen test set, where the model achieved an overall accuracy of 95.45% and a test loss of 0.2054. This high accuracy indicates that the model generalized well and performed reliably across all four classes. As shown in Fig 3, training and validation accuracy increased steadily and both losses decreased consistently, with minor fluctuations near convergence. The gap between the training and validation curves was minimal, suggesting that the model did not overfit.

Fig 3: Epoch-wise training and validation accuracy and loss curves.


 
Class-wise confusion matrix analysis
 
The confusion matrix results of the DenseNet169 model are presented in Fig 4a, while the normalized confusion matrix is shown in Fig 4b. These figures provide detailed insights into the class-wise classification performance of the model on the test dataset. As illustrated in Fig 4a, the Chocolate spot class achieved 354 correct predictions out of 401 samples, while 22 samples were misclassified as Gall, 3 as Healthy and 22 as Rust. For the Gall class, 384 out of 400 samples were correctly classified, with 11 samples misclassified as Chocolate spot, 3 as Healthy and 2 as Rust. The Healthy class showed the highest classification performance, with 400 correctly predicted samples out of 404, while 1 sample was misclassified as Chocolate spot, 2 as Gall and 1 as Rust. Similarly, the Rust class achieved 394 correct predictions out of 400 samples, with 4 samples misclassified as Chocolate spot and 2 as Gall. To provide a clearer interpretation of the class-wise performance, a normalized confusion matrix was generated as shown in Fig 4b, where the classification results are represented as percentages. The normalized matrix indicates that the Healthy class achieved the highest recall (99.01%), followed by Rust (98.50%), Gall (96.00%) and Chocolate spot (88.28%). Overall, the strong concentration of values along the diagonal elements in both matrices indicates that the DenseNet169 model achieved high classification accuracy across all disease categories.

Fig 4: Confusion matrix analysis of the DenseNet169 model: (a) raw confusion matrix showing the number of samples predicted for each class and (b) normalized confusion matrix showing percentage-based classification performance across the four disease categories.



Classification report and metrics
 
The detailed classification report is presented in Table 2. The Chocolate spot class achieved a precision of 95.68%. The recall was 88.28%, indicating that some images were not identified correctly.

Table 2: Classification report metrics for each class.


       
This resulted in an F1-score of 91.83%. The Gall class showed consistent performance with 93.66% precision and 96.00% recall. The F1-score of 94.81% suggests strong balance between sensitivity and specificity. For the Healthy class, the model performed almost perfectly. It achieved 98.52% precision, 99.01% recall and an F1-score of 98.77%. These metrics reflect the model’s ability to identify healthy leaves correctly. The Rust class also had strong results, with 94.03% precision and 98.50% recall, producing an F1-score of 96.21%. The overall accuracy of the model was 95.45% on the test dataset. The macro and weighted averages for precision, recall and F1-score were all around 95.4-95.5%, indicating consistent performance across all classes. The Matthews Correlation Coefficient (MCC) was calculated as 0.9397, showing a high level of balanced classification.
 
ROC curve analysis
 
The ROC curves for each class are displayed in Fig 5. The Chocolate spot class achieved an AUC of 0.994, indicating very high separability from other classes. The Gall class recorded an AUC of 0.997, also showing excellent performance. The healthy class reached the highest possible AUC of 1.000, confirming perfect class distinction. The Rust class followed closely with an AUC of 0.998. These results support the robustness of the model in distinguishing between all four categories.

Fig 5: ROC curves for all classes.


 
Confident prediction visualization
 
The most confident predictions made by the model are presented in Fig 6. These samples represent the top three predictions for each class based on the highest confidence scores. The Chocolate spot images were predicted with confidence values of 0.9996–0.9998, showing clear dark lesions on green leaf surfaces. The Gall class images were all predicted with full confidence (1.0000), with visible swelling and distortion of leaf structures. The Healthy class predictions also showed high confidence (0.9993-0.9996) and displayed smooth, undamaged leaf texture and uniform green coloration. Finally, Rust predictions were made with high confidence (0.9999-1.0000) and showed characteristic orange to brown pustules.

Fig 6: Top three most confident predictions from each class.


       
These confident predictions visually confirm the model’s capability to identify key features with a high degree of certainty. The confidence scores demonstrate that the model is not only accurate but also reliable in recognizing disease symptoms.
       
Table 3 summarizes representative failure cases observed during model evaluation. Most misclassifications occur between visually similar diseases such as Chocolate spot, Rust and Gall. The confusion is primarily caused by overlapping lesion patterns, variations in lesion density and differences in lighting conditions in field images.

Table 3: Summary of representative failure cases observed in the model predictions.


 
Precision-recall curve analysis
 
The precision-recall curves shown in Fig 7 complement the ROC analysis by highlighting how well the model performs under class imbalance. The Chocolate spot class achieved an average precision (AP) of 0.983, while the Gall class recorded 0.992. The Healthy class once again showed optimal performance with an AP of 1.000 and the Rust class followed with 0.995. The curves confirm that the model maintains a strong balance between precision and recall across all disease categories.

Fig 7: Precision-Recall curves for all four classes.


       
A comparative overview of recent CNN-based approaches for legume disease detection is presented in Table 4. These studies report varying performance levels depending on factors such as the CNN architecture, crop type, dataset size, image acquisition conditions and preprocessing strategies. Since these studies were conducted using different datasets and experimental protocols, the reported results should be interpreted with caution and do not represent a direct or strictly comparable benchmark. For instance, DenseNet121 was used by Abed et al., (2021) for detecting diseases in bean leaves, achieving 98.31% accuracy using manually collected images. Similarly, VGG16 was applied by Sahu et al., (2021) to a dataset of 1,296 bean leaf images, resulting in 95.31% accuracy. In another study, Joshi et al., (2021) proposed the VirLeafNet-1 model for Vigna mungo disease detection and reported 91.23% accuracy using manually collected data. More recent research has explored deeper architectures. Singh et al., (2023) utilized EfficientNetB6 with a dataset of 1,295 bean leaf images, achieving 91.74% accuracy, while Serttas and Deniz (2023) applied ResNet50 on a similar dataset size and reported 98.33% accuracy. In contrast, Kursun et al., (2023) used DenseNet201 on a small dataset of 60 dry bean images, achieving 100% accuracy, although the limited dataset size may restrict the generalizability of the results. For faba bean disease detection, Salau et al., (2023) proposed an end-to-end CNN model using both raw and preprocessed images and achieved 98.14% accuracy. Jeong and Na (2024) developed a custom CNN architecture with eight convolutional layers and reported 91.00% accuracy. More recently, Mostafa et al., (2025) trained a sequential CNN on expert-annotated and preprocessed data, achieving 98.92% accuracy.

Table 4: Overview of CNN-based approaches reported in previous studies for legume disease detection.


       
In the present study, the DenseNet169 model was applied to a large-scale dataset of 8,021 expert-labeled faba bean leaf images collected under natural field conditions, achieving an overall classification accuracy of 95.45%. Although direct comparison with previous studies is challenging due to differences in datasets and experimental settings, the results demonstrate that the proposed approach provides reliable and robust performance for practical field-based disease detection scenarios.
This study presented a CNN-based classification model using DenseNet169 for detecting major faba bean leaf diseases. The model achieved 95.45% accuracy on a large, field-based dataset. Results confirm the effectiveness of transfer learning in plant disease detection. Limitations include reliance on image data only, which may not capture environmental factors. The dataset, though large, was collected from limited regions, which may affect generalization. Also, real-time deployment was not tested in this study. Future work will focus on expanding the dataset across regions and seasons. Integrating weather data and soil conditions may improve model accuracy. Real-time mobile or drone-based implementations can help validate the model’s field performance. Combining CNNs with hybrid models could further boost detection and classification efficiency.
Disclaimers
 
The views and conclusions expressed in this article are solely those of the authors and do not necessarily represent the views of their affiliated institutions. The authors are responsible for the accuracy and completeness of the information provided, but do not accept any liability for any direct or indirect losses resulting from the use of this content.
 
Funding details
 
This research received no external funding.
 
Authors’ contributions
 
All authors contributed toward data analysis, drafting and revising the paper and agreed to be responsible for all aspects of this work.
 
Data availability
 
The dataset analyzed in this study consists of 8,021 expert-labeled faba bean leaf images collected under natural field conditions. Due to institutional and field data collection policies, the dataset is not publicly available. However, the data supporting the findings of this study may be made available by the corresponding author for academic and non-commercial research purposes upon reasonable request.
 
Availability of data and materials
 
Not applicable.
 
Use of artificial intelligence
 
Not applicable.
 
Declarations
 
Authors declare that all works are original and this manuscript has not been published in any other journal.
The authors declare that they have no conflict of interest.

  1. Abed, S.H., Al-Waisy, A.S., Mohammed, H.J. and Al-Fahdawi, S. (2021). A modern deep learning framework in robot vision for automated bean leaves diseases detection. International Journal of Intelligent Robotics and Applications. 5(2): 235-251.

  2. Akchaya, K., Parasuraman, P., Pandian, K., Vijayakumar, S., Thirukumaran, K., Mustaffa, M. R.a.F., Rajpoot, S.K. and Choudhary, A.K. (2025). Boosting resource use efficiency, soil fertility, food security, ecosystem services and climate resilience with legume intercropping: A review. Frontiers in Sustainable Food Systems. 9. https:// doi.org/10.3389/fsufs.2025.1527256.

  3. Alam, T.S., Jowthi, C.B. and Pathak, A. (2024). Comparing pre- trained models for efficient leaf disease detection: A study on custom CNN. Journal of Electrical Systems and Information Technology. 11(1). https://doi.org/10.118 6/s43067-024-00137-1.

  4. Arora V., Kanwal A., Sharma I. and Singh R. (2025). Pesticides pollution in agroecosystems and their social and ecological  consequences. Bulletin of Pure and Applied Sciences- Botany. 44B(1): 60-64. https://bpasjournals.com/botany/ index.php/journal/article/view/141/133.

  5. Atila, Ü., Uçar, M., Akyol, K. and Uçar, E. (2020). Plant leaf disease classification using EfficientNet deep learning model. Ecological Informatics. 61: 101182. https://doi.org/10.10 16/j.ecoinf.2020.101182.

  6. Baiju, B.V., Kirupanithi, N., Srinivasan, S., Kapoor, A., Mathivanan, S.K. and Shah, M.A. (2025). Robust CRW crops leaf disease detection and classification in agriculture using hybrid deep learning models. Plant Methods. 21(1). https://doi.org/10.1186/s13007-025-01332-5

  7. Desfita, S., Sari, W., Wahyuni, D., Putri, F., Pramono, A., Pato, U., Pratiwi, D., Grzelczyk, J. and Budryn, G. (2025). Synergistic effects of multi-strain probiotic and prebiotic combinations on immune recovery in aging populations. International Journal of Probiotics and Prebiotics. 20(1): 10-18. https://doi.org/10.37290/ijpp2641-7197.20:10-18.

  8. Dheeraj, A. and Chand, S. (2024). LWDN: lightweight DenseNet model for plant disease diagnosis. Journal of Plant Diseases and Protection. 131(3): 1043-1059. https:// doi.org/10.1007/s41348-024-00915-z.

  9. Dhull, S.B., Kidwai, M.K., Noor, R., Chawla, P. and Rose, P.K. (2021). A review of nutritional profile and processing of faba bean (Vicia faba L.). Legume Science. 4(3). https:// doi.org/10.1002/leg3.129.

  10. FAO. (2024). FAOSTAT Crop Data: Faba Bean Production. Food and Agriculture Organization of the United Nations. https:// www.fao.org/faostat.

  11. Geetanjali and Singh, A.R. (2025). Precision and smart farming: An approach to efficient resource management in agriculture. Journal of Artificial Intelligence in Agriculture and Applied Research. 1(1): 25-33. https://agrpjournals.com/ journals/journal/index.php/jaiaar/article/view/6/5.

  12. Ibrahim, H., Ahmed, A.M., Bouallegue, B., Khattab, M.M., El-Fattah, M.A. and El-Mawla, N.A. (2025). Deep convolutional neural networks for plant disease detection: A mobile application approach (Agri Bot). International Journal of Intelligent Systems. (1). https://doi.org/10.1155/int/ 7644407.

  13. ICAR. (2023). Indian Council of Agricultural Research Annual Report 2022-23. https://icar.org.in

  14. Jeong, H.Y. and Na, I.S. (2024). Efficient faba bean leaf disease identification through smart detection using deep convolutional neural networks. Legume Research - An International Journal. doi: 10.18805/LRF-798.

  15. Joshi, R.C., Kaushik, M., Dutta, M.K., Srivastava, A. and Choudhary, N. (2021). VirLeafNet: Automatic analysis and viral disease diagnosis using deep-learning in Vigna mungo plant. Ecological Informatics. 61: 101197.

  16. Kursun, R., Bastas, K.K. and Koklu, M. (2023). Segmentation of dry bean (Phaseolus vulgaris L.) leaf disease images with U-Net and classification using deep learning algorithms. European Food Research and Technology.  249(10): 1- 16.

  17. Leite, D., Brito, A. and Faccioli, G. (2024). Advancements and outlooks in utilizing Convolutional Neural Networks for plant disease severity assessment: A comprehensive review. Smart Agricultural Technology. 9: 100573. https://doi.org/10.1016/j.atech.2024.100573

  18. Maalouf, F., Ahmed, S. and Bishaw, Z. (2021). Faba Bean. In: Elsevier eBooks. (pp. 105-131). https://doi.org/10.1016/ b978-0-12-821450-3.00008-1.

  19. Ministry of Agriculture and Farmers Welfare. (2023). Annual Report 2022-2023. Government of India. https://agricoop.nic.in

  20. Mostafa, A., Alnuaim, A. and AlZubi, A.A. (2025). Utilizing convolutional neural networks for accurate detection of leaf diseases in fava beans. Legume Research: An International Journal. 48(3): 494-502. doi: 10.18805/LRF-823.

  21. Mukherjee, B., Naskar, M.K., Chakraborty, S., Dutta, S. and Mandal, T.K. (2025). Building Food Security and Climate Resilience: The Promise of Climate Smart Pulse Production Technologies. In: Sustainability Sciences in Asia and Africa (pp. 17- 43). https://doi.org/10.1007/978-981-96-7699-6_2.

  22. Nirgude V. and Rathi S. (2021). A robust deep learning approach to enhance the accuracy of pomegranate fruit disease detection under real field condition. Journal of Experimental Biology and Agricultural Sciences. 9(6): 863-870. https:// doi.org/10.18006/2021.9(6).863.870.

  23. Sahu, P., Chug, A., Singh, A.P., Singh, D. and Singh, R.P. (2021). Deep learning models for beans crop diseases: Classification and visualization techniques. International Journal of Modern Agriculture. 10(1): 796-812.

  24. Salau, A.O., Abeje, B.T., Faisal, A.N. and Asfaw, T.T. (2023). Faba Bean Disease Detection using Deep Learning Techniques. In 2023 International Conference on Cyber Management and Engineering (CyMaEn) (pp. 344-349). IEEE. https:// doi.org/10.1109/CyMaEn57228.2023.10051088.

  25. Sbai, H., Hajib, A., Msairi, S., Amalich, S., Bouyahya, A., Lee, L., Goh, K.W., Tabyaoui, M. and Harhar, H. (2024). Fungal infections of legume crops: Challenges and management approaches. Journal of Agriculture and Food Research. 18: 101447. https://doi.org/10.1016/j.jafr.2024.101447.

  26. Serttas, S. and Deniz, E. (2023). Disease detection in bean leaves using deep learning. Communications Faculty of Sciences University of Ankara Series A2-A3 Physical Sciences and Engineering. 65(2): 115-129.

  27. Singh, A.K., Bharati, R.C., Ch, N., Manibhushan, R. and Pedpati, A. (2013). An assessment of faba bean (Vicia faba L.) current status and future prospect. African Journal of Agricultural Research. 8(50): 6634-6641. https://doi.org/ 10.5897/ajar2013.7335.

  28. Singh, V., Chug, A. and Singh, A.P. (2023). Classification of beans leaf diseases using fine tuned CNN model. Procedia Computer Science. 218: 348-356.

  29. Souza, K.F., Gonçalves, L.C.O. and Espindola, F.S. (2026). A broad overview of the multifaceted royal jelly and its applicability in health and performance. Current Topics in Nutraceutical Research. 24(1): 4-14. https://doi.org/10.37290/ctnr.v24i1.8.

  30. Sriram, P. and Kumari, Y.S. (2026). Game theory in urban birds: A Nash equilibrium analysis using crows and pigeons. Bio- Science Research Bulletin. 42(1): 1-6. https://bpas journals.com/life-sciences/index.php/journal/article/ view/150/140.

  31. Tiwari, N., Barpete, S., Kumar, T., Saxena, D.R. and Maalouf, F. (2023). Disease resistance: An essential for better adaptability and production of faba bean in India (Vicia faba L.). In: Diseases in Legume Crops (Chapter 8). [U.C. Jha, H. Nayyar, K.D. Sharma, E.J.B. von Wettberg, P. Singh and K. H. Siddique (Eds.)], Springer. https:// doi.org/10.1007/978-981-99-3358-7_8.

  32. Upadhyay, A., Chandel, N.S., Singh, K.P., Chakraborty, S.K., Nandede, B.M., Kumar, M., Subeesh, A., Upendar, K., Salem, A. and Elbeltagi, A. (2025). Deep learning and computer vision in plant disease detection: A comprehensive review of techniques, models and trends in precision agriculture. Artificial Intelligence Review. 58(3). https:// doi.org/10.1007/s10462-024-11100-x.

  33. Zeleke, E.A. and Legesse, N.H. (2023). Management of chocolate spot of faba bean (Botrytis fabae) through fungicides and host resistance in North Shewa, Central Highland of Ethiopia. Indian Phytopathology. 76(2): 531-540. https://doi.org/10.1007/s42360-023-00636-7.
 
In this Article
Published In
Indian Journal of Agricultural Research

Editorial Board

View all (0)