• Submitted24-03-2026|

  • Accepted20-08-2026|

  • First Online 17-11-2026|

  • doi 10.18805/LR-5660

Background: Haricot bean is an important legume crop contributing to food security and income generation in many developing countries. However, its productivity is severely constrained by major foliar diseases such as angular leaf spot, bacterial blight and bean rust. Traditional disease identification relies on manual inspection, which is time-consuming, subjective and prone to error. Therefore, developing an accurate and automated disease detection system is essential for early diagnosis and effective crop management.

Methods: In this study, a modified AlexNet-based convolutional neural network was developed for automated identification of foliar diseases of haricot bean. A total of 3,447 leaf images belonging to four classes-angular leaf spot, bacterial blight, bean rust and healthy leaf-were used. The dataset was divided into training, validation and testing subsets using an 80:10:10 ratio. Data augmentation techniques, including rotation, flipping, zooming and brightness adjustments, were applied to enhance model generalization. Batch normalization and dropout layers were incorporated into the original AlexNet architecture to improve training stability and reduce overfitting. The model was trained for 50 epochs using categorical cross-entropy loss.

Result: The proposed model achieved 96.75% training accuracy and 88.41% validation accuracy. On the independent test set, it attained 99.13% accuracy with high precision, recall and F1-scores across all classes. ROC and precision-recall analyses produced AUC and AP values of 1.0000, demonstrating excellent classification capability. The results confirm that the developed model provides a reliable and efficient approach for automated haricot bean leaf disease detection.

Food security in many developing countries depends heavily on grain legumes, among which haricot bean (Phaseolus vulgaris L.; 2n = 22) holds a prominent position. As a diploid, self-pollinated species belonging to the Fabaceae family, it is cultivated on more than 35 million hectares globally, with an annual production exceeding 30 million tonnes and an average productivity of approximately 800-900 kg/ha (FAO, 2023). It is recognized as one of the most important food and cash crops due to its high nutritional value, adaptability and market demand (Huang et al., 2020; Ahmed et al., 2025). Globally, Africa contributes over one-fifth of total common bean production, ranking third after Asia and the Americas (FAO, 2023). Countries in East Africa, including Ethiopia, Kenya, Uganda and Tanzania, represent major production centers (Pandey and Kumar, 2024; Tigist et al., 2023).
       
In Ethiopia, haricot bean production has expanded rapidly in recent years, reflecting its high market demand and role in food security (CSA, 2022). The crop contributes significantly to national agricultural GDP and smallholder farmer income, with productivity levels ranging between 1.2-1.8 t/ha under improved management conditions (Krishna and Meenakshi, 2022). However, biotic stresses such as angular leaf spot, bacterial blight and bean rust are among the most destructive foliar diseases, causing yield losses ranging from 20% to 80% depending on disease severity, environmental conditions and management practices, resulting in substantial economic losses to farmers (Zamukulu et al., 2023; Etana and Nebiyu, 2023). Traditional disease identification in the field is typically conducted through visual inspection by agronomists or farmers, which is labor-intensive, time-consuming and often results in misclassification due to symptom similarity and human subjectivity (Tavares et al., 2024). In contrast, recent studies have demonstrated that automated image-based approaches using machine vision can significantly improve diagnostic accuracy, consistency and scalability, enabling rapid and real-time disease detection (Cholo et al., 2023; Naik et al., 2023).
       
In recent years, deep learning (DL) methods, particularly convolutional neural networks (CNNs), have emerged as highly effective tools for plant disease classification. A recent systematic review highlights that CNN architectures have achieved consistently high accuracy across diverse plant species and disease types, often exceeding 90% in multi-class classification tasks when trained on sufficient image data (Broughton et al., 2003; Pacal et al., 2024). Studies have also demonstrated that optimised CNN models, such as AlexNet, EfficientNet and MobileNet, can achieve high accuracy and enable real-time deployment on resource-limited devices (Mehta et al., 2025; Mohammad et al., 2026; Ozçelik et al., 2026).
       
Among CNN models, AlexNet remains one of the most influential deep learning architectures used in plant disease detection research due to its balance of performance and computational efficiency (Paek et al., 2026; Souza et al., 2026; Sriram and Kumari, 2026). Chen et al., (2022) and Ramya and Kumar (2023) showed that modified and transfer-learning-based AlexNet models can deliver strong accuracy for tomato disease detection. Mishra et al., (2025) conducted a comparative analysis across CNN, AlexNet, ResNet and EfficientNet, where EfficientNet achieved superior generalization performance. Similarly, Ghosh et al., (2025) introduced a hybrid deep learning architecture combining advanced networks such as ViT and EfficientNetV2, achieving 98% accuracy. In rice disease detection, Akter et al., (2024) reported exceptionally high accuracy (99.99%) using a CNN-based approach. Despite substantial progress, most existing research has focused on common crops such as tomato, maize and potato, with relatively fewer studies concentrating specifically on haricot bean leaf disease classification using deep learning frameworks. Moreover, challenges such as class imbalance, environmental variation and limited annotated datasets often hinder model generalization in real field conditions.
       
This study aims to develop an enhanced AlexNet framework specifically for multi-class classification of haricot bean leaf diseases, with the following objectives:
• Design a modified AlexNet architecture tailored for haricot bean leaf image classification.
• Incorporate data augmentation to improve robustness against environmental variability.
• Apply class-weighted training to address dataset imbalance.
• Evaluate model performance using comprehensive metrics.
Dataset description
 
This study utilized a total of 3,447 haricot bean leaf images categorized into four classes: angular leaf spot, bacterial blight, bean rust and healthy leaves. Angular leaf spot is characterized by small, water-soaked lesions that enlarge into angular brown spots restricted by leaf veins; bacterial blight appears as irregular necrotic patches with yellow halos and water-soaked margins; bean rust is identified by reddish-brown pustules primarily on the underside of leaves; whereas healthy leaves exhibit uniform green coloration without visible lesions or discoloration. The images were collected and organized to represent realistic disease variations commonly observed under field conditions. To ensure reliable model evaluation and to prevent information leakage, the dataset was divided into training, validation and testing subsets following an 80:10:10 ratio. The training set contained 2,066 images, the validation set consisted of 690 images and the testing set included 691 images. This stratified division ensured that each class was proportionally represented across all subsets. Fig 1 illustrates representative samples from each class used in the classification task.

Fig 1: Representative haricot bean leaf classes used in the study.


 
Reproducibility setup
 
To guarantee experimental reproducibility, fixed random seeds were applied across different computational libraries. Random seed initialization ensures that stochastic processes such as dataset shuffling, weight initialization and batch sampling remain consistent across repeated runs. Seeds were set for Python’s built-in random module, NumPy and TensorFlow frameworks. The seed value (42) was consistently applied across all experiments to ensure replicability. Mathematically, reproducibility ensures that the initialization function W0 = f (seed) produces identical initial weights W0 when the same seed value is used. This step enhances experimental reliability and enables fair performance comparison.
 
Image preprocessing
 
All images were resized to 224 × 224 × 3 pixels to match the input requirement of the AlexNet architecture. Additionally, images were checked for noise, distortion and inconsistent illumination to ensure data quality before training. Each image was normalized to scale pixel values into the range [0, 1], computed as:
 

 
Where,
X= The original pixel intensity values. 
       
Normalization reduces numerical instability and accelerates convergence during training by preventing large gradient fluctuations.
 
Data augmentation
 
To improve generalization and minimize overfitting, real-time data augmentation was applied exclusively to the training dataset. Augmentation artificially increases dataset diversity by introducing controlled transformations that simulate real agricultural field variations. The applied transformations included horizontal and vertical flipping, random rotation (20%), random zooming (10%), brightness adjustment (20%) and contrast modification (20%).
       
These transformations can be mathematically expressed as:
 
X′ = T (X)
 
Where,
T= A transformation function applied to the original image X.
       
Augmentation improves robustness against changes in lighting conditions, camera angle, leaf orientation and scale variations encountered in practical scenarios.
 
Class weight calculation
 
Although the dataset was relatively balanced, minor variations in class frequency may bias the learning process toward majority classes. To address this issue, class weights were calculated and incorporated into the loss function. Class weights were computed using:
 

 
Where,
N= Total number of samples.
k= Number of classes.
ni= Number of samples in class i.

The computed class weights are shown in Table 1.

Table 1: Computed class weights.



These weights help balance the contribution of each class during training and reduce classification bias.
 
Proposed AlexNet architecture
 
In this work, a modified AlexNet architecture was implemented using a functional API for four-class bean leaf disease classification. The network receives an input tensor X ∈ ℝ 224 × 224 × 3, representing RGB images resized to 224 × 224 pixels. The first convolutional layer applies 96 learnable filters to extract low-level spatial features such as edges, color gradients and simple textures. In general, the convolution operation can be expressed as:
 

 
Where,
W= Convolutional kernel weights.
bk= Bias term for the kth filter.
Yi,j,k= Output feature map.
       
This first layer produces an output of size 54 × 54 × 96 with 34,944 parameters. A Rectified Linear Unit (ReLU) activation function is applied to introduce non-linearity:
 
f(x) = max (0, x)
 
Batch normalization follows to stabilize the learning process by normalizing intermediate activations. The normalization process is defined as:
 

 
Where
μ and σ2= Batch mean and variance.
γ and β= Learnable scaling and shifting parameters.
       
A max-pooling layer then reduces spatial dimensions to 26 × 26 × 96 defined as:
 

 
Where,
R= The pooling region.
       
The second convolutional layer increases feature depth to 256 channels while preserving spatial dimensions at 26 × 26, containing 614,656 parameters. This layer captures more complex patterns such as vein structures and disease-specific textures. Batch normalization and a second max-pooling layer reduce the output to 12 × 12 × 256. The third convolutional layer produces 384 feature maps of size 12 × 12 × 384 with 885,120 parameters, enabling extraction of higher-level morphological features. After batch normalization, the fourth convolutional layer maintains 384 channels with 1,327,488 parameters, further refining hierarchical representations. The fifth convolutional layer reduces channel depth to 256 with 884,992 parameters, followed by batch normalization and a final max-pooling layer, producing compact feature maps of size 5 × 5 × 256.
       
The resulting tensor is flattened into a vector of length 6,400, which serves as input to the fully connected layers. The first dense layer contains 2,048 neurons and 13,109,248 parameters. The dense transformation is computed as:
 
z = Wx + b
 
Where,
W= Weight matrix.
b= Bias vector.
       
ReLU activation is again applied, followed by dropout regularization to prevent overfitting. Dropout randomly deactivates neurons during training according to:
 
 
Where,
ri ~ Bernoulli (p) and p = The retention probability.
       
The second dense layer contains 1,024 neurons with 2,098,176 parameters and is also followed by dropout. The final output layer consists of 4 neurons corresponding to the four bean leaf classes. A Softmax activation function converts logits into probability distributions:

 
Ensuring that predicted probabilities sum to one. The model is trained using categorical cross-entropy loss, defined as:
 

 
Where,
yk= The true label.
yk= The predicted probability.
       
Overall, the architecture contains 18,964,228 total parameters, of which 18,961,476 are trainable and 2,752 are non-trainable, primarily associated with batch normalization statistics. By integrating deep convolutional feature extraction, normalization, pooling and fully connected classification layers with dropout regularization, the proposed AlexNet-based model effectively learns discriminative features for accurate bean leaf disease classification while maintaining training stability and generalization capability. The model was trained for 50 epochs with a batch size of 32. Fig 2 illustrates the overall architecture of the proposed model.

Fig 2: Modified AlexNet architecture used for haricot bean disease classification.


 
Performance evaluation metrics
 
Model performance was assessed using accuracy, precision, recall, F1-score and confusion matrix analysis. Accuracy is defined as:
     

 
Precision, recall and F1-score are calculated as:
 





 
The confusion matrix provides a class-wise breakdown of prediction performance and helps identify misclassification patterns.
Fig 3 presents the training and validation accuracy and loss curves obtained over 50 epochs. The training accuracy shows a rapid increase during the early epochs, indicating that the model quickly learned meaningful visual features from the haricot bean leaf images. As training progressed, the accuracy continued to improve gradually and stabilized at 96.75% by the final epoch. Similarly, the validation accuracy followed an upward trend and reached 88.41%, demonstrating that the model maintained good generalization performance on unseen validation data. The loss curves further support this observation. Training loss decreased steadily throughout the training process, reflecting effective optimization of the categorical cross-entropy objective function. Validation loss also showed a general downward trend, although minor fluctuations were observed across epochs. These variations are expected in stochastic gradient-based learning due to mini-batch updates and real-time data augmentation. Importantly, there was no significant divergence between training and validation curves, suggesting that overfitting was effectively controlled. The incorporation of dropout layers, batch normalization and augmentation techniques contributed significantly to maintaining stable convergence.

Fig 3: Training and validation accuracy and loss curves.


       
Although a moderate gap exists between training and validation accuracy, the difference remains within an acceptable range for deep learning models trained on real-world agricultural datasets. This indicates that the model learned robust and discriminative features such as lesion patterns, rust formations and discoloration characteristics associated with different disease classes. Evaluation on the independent test set further confirms the effectiveness of the proposed model. The network achieved a test accuracy of 99.13% with a test loss of 1.7738, demonstrating strong predictive capability on unseen haricot bean leaf images. The high test performance indicates excellent generalization ability and suggests that the trained AlexNet-based architecture successfully captured disease-specific visual patterns. Overall, the results validate the suitability of the proposed approach for accurate and reliable automated bean leaf disease classification in practical agricultural applications.
       
Fig 4 presents the confusion matrix obtained from the test dataset for the four-class haricot bean leaf disease classification task. The matrix provides a detailed breakdown of correct and incorrect predictions made by the model across the classes. The results indicate that the model achieved near-perfect classification performance for most categories. Specifically, 173 samples of angular leaf spot, 172 samples of bacterial blight and 173 samples of bean rust were correctly classified, with no misclassifications observed for these three classes. This demonstrates that the model effectively captured distinctive disease-related visual characteristics such as lesion boundaries, discoloration patterns and rust pustules. For the healthy leaf category, 167 samples were correctly classified, while 6 samples were misclassified as bean rust. This minor misclassification may be attributed to the presence of very early or mild rust symptoms, which are visually indistinguishable from healthy leaf surfaces, as well as variations in illumination, background noise and image acquisition conditions. Nevertheless, the overall classification performance remains exceptionally high. The dominance of diagonal elements and the minimal off-diagonal values confirm excellent class separability and strong generalization capability.

Fig 4: Confusion matrix of the proposed AlexNet model.


       
Table 2 presents the detailed classification report of the proposed AlexNet-based model evaluated on the test dataset. The results show outstanding performance across all four categories. Both angular leaf spot and bacterial blight achieved perfect precision, recall and F1-score values of 1.0000, indicating that all samples in these classes were correctly identified without any false positives or false negatives. For the bean rust class, the model achieved a precision of 0.9665, recall of 1.0000 and an F1-score of 0.9830. The perfect recall indicates that all actual bean rust samples were correctly detected, although a small number of samples from other classes were predicted as bean rust, slightly lowering the precision. Similarly, the healthy leaf class achieved a precision of 1.0000, recall of 0.9653 and an F1-score of 0.9824. The slightly lower recall value corresponds to the small number of healthy samples misclassified as bean rust, as observed in the confusion matrix. The overall test accuracy reached 99.13%, with both macro and weighted averages of precision, recall and F1-score equal to 0.9913. The macro average indicates balanced performance across all classes, while the weighted average confirms that class distribution did not bias the results.

Table 2: Classification report of the proposed model.


       
Fig 5 illustrates the receiver operating characteristic (ROC) curves and precision-recall (PR) curves for all four classes. All four classes achieved an area under the curve (AUC) value of 1.0000, indicating perfect separability between positive and negative samples for each category. The ROC curves closely follow the top-left boundary of the graph, which reflects an optimal classifier with maximum sensitivity and specificity. Similarly, the precision-recall curves demonstrate exceptional classification performance. The average precision (AP) for each class is reported as 1.0000, indicating near-perfect precision across all recall levels. The PR curves remain close to the upper-right corner of the graph, further validating the model’s strong predictive capability, especially in correctly identifying positive disease cases.

Fig 5: ROC and precision-recall curves of the proposed model.


       
Fig 6 presents representative test samples along with their predicted labels and confidence scores generated by the proposed AlexNet-based model. The first sample corresponds to a healthy leaf, correctly classified as healthy leaf with a confidence score of 0.9991, demonstrating the model’s ability to accurately recognize normal leaf texture and color distribution. The second image represents bean rust, which was correctly predicted with a perfect confidence of 1.0000, highlighting the model’s strong capability in detecting rust-specific visual characteristics such as yellowing and lesion patterns. The third example shows a bacterial blight leaf correctly classified with a confidence of 0.9991, indicating that the model effectively captures irregular necrotic lesions and edge discoloration associated with this disease. Similarly, the fourth image of angular leaf spot was accurately predicted with a confidence score of 0.9997, reflecting precise detection of angular-shaped dark lesions and scattered spotting patterns. The consistently high confidence values across all examples demonstrate the robustness and reliability of the trained model. These qualitative results further support the quantitative findings from the confusion matrix, classification report and ROC analysis, confirming the effectiveness of the proposed approach for automated haricot bean leaf disease classification.

Fig 6: Sample predictions of the proposed AlexNet model.


       
Previous studies have applied AlexNet and other deep learning architectures for plant disease classification across different crops, demonstrating varying levels of performance depending on dataset size, crop type and model design. Chen et al., (2022) utilized a modified AlexNet on a large-scale tomato dataset consisting of 22,930 images across 10 classes and achieved an accuracy of 98%, highlighting the effectiveness of CNN-based models for multi-class classification tasks. Similarly, Kewat (2023) applied AlexNet to an ImageNet-based plant disease and pest dataset and reported 96.50% accuracy, confirming its suitability for agricultural image analysis.
       
In contrast, studies using smaller datasets reported relatively lower performance. Sholihin et al., (2024) implemented a modified AlexNet for cassava classification using 1,400 images and achieved 87% accuracy, indicating the influence of dataset size and variability on model performance. Fang et al., (2025) applied AlexNet for bean leaf disease classification (angular leaf spot, rust and healthy) and obtained 95.4% training accuracy but only 82.81% overall accuracy, suggesting challenges in distinguishing visually similar classes under real conditions.Furthermore, comparative architecture studies highlight the advantage of advanced models. Mishra et al., (2025) evaluated CNN, AlexNet, ResNet and EfficientNet on potato (2,152 images) and mango (4,000 images) datasets, where EfficientNet achieved the highest accuracy of 97.8%, outperforming other architectures in terms of generalization.
       
Compared to these existing studies, the present work demonstrates superior performance. The proposed modified AlexNet model, enhanced with batch normalization and dropout, achieved 99.13% test accuracy on a dataset of 3,447 haricot bean images. Despite using a moderate-sized dataset, the model shows improved robustness and generalization capability, outperforming several previously reported approaches. This confirms the effectiveness and competitiveness of the proposed architecture for plant disease classification tasks.
This study presented a modified AlexNet-based convolutional neural network for automated haricot bean leaf disease classification. The model demonstrated high accuracy, strong generalization capability and excellent discriminative performance across four classes. The integration of batch normalization, dropout and data augmentation effectively minimized overfitting and improved robustness. Despite the promising results, the study has some limitations. The dataset was relatively limited in size and collected under controlled conditions, which may not fully represent complex field environments. Additionally, only a single deep learning architecture was extensively evaluated. Future work will focus on expanding the dataset with diverse field images, incorporating real-time mobile-based deployment and comparing the model with advanced architectures such as EfficientNet and Vision Transformers. Integrating explainable AI techniques may further enhance model transparency and practical adoption in smart agriculture systems.
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 the aspects of this work.
 
Data availability
 
The data analysed/generated in the present study will be made available from corresponding authors 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.
Authors declare that they have no conflict of interest.

  1. Ahmed, E.H., Teka, T.A., Tolera, K.D., Gemede, H.F. and Hussen, K.E. (2025). Technological properties of improved haricot bean (Phaseolus vulgaris L.) varieties cultivated in Ethiopia. Applied Food Research. 5(2): 101107. https://doi.org/10.1016/j.afres.2025.101107.

  2. Akter, S., Sumon, R.I., Ali, H. and Kim, H. (2024). Utilizing convolutional neural networks for the effective classification of rice leaf diseases through a deep learning approach. Electronics. 13(20): 4095. https://doi.org/10.3390/electronics13204095.

  3. Broughton, W. J., Hernández, G., Blair, M., Beebe, S., Gepts, P. and Vanderleyden, J. (2003). Beans (Phaseolus spp.)- Model food legumes. Plant and Soil. 252(1): 55-128. https://doi.org/10.1023/a:1024146710611.

  4. Central Statistical Agency (CSA). (2022). Agricultural Sample Survey Series: Report on Area and Production for Major Crops (Private Holdings, Main Season). Addis Ababa, Ethiopia.

  5. Chen, H., Widodo, A.M., Wisnujati, A., Rahaman, M., Lin, J.C., Chen, L. and Weng, C. (2022). ALEXNET convolutional neural network for disease detection and classification of tomato leaf. Electronics. 11(6): 951. https://doi.org/10.3390/electronics11060951.

  6. Cholo, M., Marisennayya, S., Bojago, E., Leja, D. and Divya, R. (2023). Determinants of adoption and intensity of improved haricot bean (Phaseolus vulgaris L.) varieties: A socio-agronomic study from southern Ethiopia. Journal of Agriculture and Food Research. 13: 100656. https://doi.org/10.1016/j.jafr.2023.100656.

  7. Etana, D. and Nebiyu, A. (2023). Response of common bean (Phaseolus vulgaris L.) to lime and TSP fertilizer under acid soil. Heliyon. 9(4): e15176. https://doi.org/10.1016/j.heliyon.2023.e15176.

  8. Fang, H., Sun, Y., Kim, T., Xu, J. and Deng, Q. (2025). The role of deep learning in enhancing crop sustainability: A study on AlexNet’s application in detecting bean leaf disease. Legume Research-An International Journal. 49(4): 701- 708. doi: 10.18805/LRF-843.

  9. Food and Agriculture Organization of the United Nations (FAO). (2023). Faostat. https://www.fao.org/faostat.

  10. Ghosh, H., Rahat, I.S., Emon, M.M.R., Mashrafi, M.J., Tanzin, M.A. A.A., Mohanty, S.N. and Kant, S. (2025). Advanced neural network architectures for tomato leaf disease diagnosis in precision agriculture. Discover Sustainability. 6(1). https://doi.org/10.1007/s43621-025-01149-1.

  11. Huang, Y., Liu, X., Cao, D., Chen, G., Li, S., Wang, G., Wang, J. and Xu, S. (2020). Cross-species amplification of common bean (Phaseolus vulgaris) EST-SSRs within hyacinth bean, pea and soybean. Legume Research-An International Journal. 44(3): 287-291. doi: 10.18805/LR-574.

  12. Kewat, K. (2023). Plant Disease Classification using Alex Net. Research Square (Research Square). https://doi.org/ 10.21203/rs.3.rs-2612739/v1.

  13. Krishna, K.L. and Meenakshi, J.V. (2022). Agricultural productivity growth and structural transformation in rural India: Some recent evidence. Journal of Quantitative Economics. 20(S1): 277-302. https://doi.org/10.1007/s40953-022-00321-y.

  14. Mehta, A.R., Kumar, P., Prem, G., Aggarwal, S. and Kumar, R. (2025). AI-powered innovations in agriculture: A systematic review on plant disease detection and classification. Indian Journal of Agricultural Research. 59(9): 1321-1330. doi: 10.18805/IJARe.A-6371.

  15. Mishra, U., Pandey, A., G, L. and K, T. (2025). Deep learning-based disease detection in potato and mango leaves: A comparative study of CNN, AlexNet, ResNet and EfficientNet. Scientific Reports. 16(1): 2788. https://doi.org/10.1038/s41598-025-32607-5.

  16. Mohammad, A.A.S., Mohammad, S.I., Vasudevan, A. and Heindric, A. (2026). Intelligent decision support system for crop management using data analytics. Journal of Experimental Biology and Agricultural Sciences. 14(3): 9-14. https://jebas.org/ojs/index.php/jebas/article/view/3796.

  17. Naik, N.K., Sethy, P.K., Behera, S.K. and Amat, R. (2023). A methodical analysis of deep learning techniques for detecting Indian lentils. Journal of Agriculture and Food Research. 15: 100943. https://doi.org/10.1016/j.jafr.2023.100943.

  18. Ozçelik, H., Sağlam, B. and Öcal, A.S. (2026). New natural methods for combating mold fungi in agricultural products: The isparta example. Bulletin of Pure and Applied Sciences- Botany. 45B(2): 1-17. https://bpasjournals.com/botany/index.php/journal/article/view/174/150.

  19. Pacal, I., Kunduracioglu, I., Alma, M.H., Deveci, M., Kadry, S., Nedoma, J., Slany, V. and Martinek, R. (2024). A systematic review of deep learning techniques for plant diseases. Artificial Intelligence Review. 57(11). https://doi.org/10.1007/s10462-024-10944-7.

  20. Paek, J.J., Kim, J., Bae, S., Paek, K.,and Lee, Y. (2026). Chive- fortified fermentation enhances gamma-aminobutyric acid production by Levilactobacillus brevis PL9014. International Journal of Probiotics and Prebiotics. 21(1): 1-6. https://doi.org/10.37290/ijpp.v21i1.12.

  21. Pandey, M. and Kumar, N. (2024). Diversity of traditional grain legumes of Himalayan region of Uttarakhand: A review. Agricultural Reviews. 45(1): 96-102. doi: 10.18805/ag.R-2362.

  22. Ramya, R. and Kumar, P. (2023). High-performance deep transfer learning model with batch normalization based on multiscale feature fusion for tomato plant disease identification and categorization. Environmental Research Communications. 5(12): 125015. https://doi.org/10.1088/2515-7620/ace594.

  23. Sholihin, M., Fudzee, M.F.M., Ismail, M.N., Wati, E.N., Arshad, M.S. and Gusman, T. (2024). Modified alexnet architecture for classification of cassava based on leaf images. JOIV International Journal on Informatics Visualization. 8(3): 1074. https://doi.org/10.62527/joiv.8.3.2966.

  24. 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.

  25. 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://bpasjournals.com/life-sciences/index.php/journal/article/view/150/140.

  26. Tavares, M.S., Silva, C.A. A. C., Regazzo, J.R., De Souza, S.E.J., Da Silva, T.L., Fiorio, P.R. and Baesso, M.M. (2024). Performance of machine learning models in predicting common bean (Phaseolus vulgaris L.) crop nitrogen using NIR spectroscopy. Agronomy. 14(8): 1634. https://doi.org/10.3390/agronomy14081634.

  27. Tigist, S.G., Sibiya, J., Amelework, A. and Keneni, G. (2023). Agromorphological and physiological performance of ethiopian common bean (Phaseolus vulgaris L.) genotypes under different agroecological conditions. Plants. 12(12): 2342. https://doi.org/10.3390/plants12122342.

  28. Zamukulu, P.M., Bagula, E.M., Mondo, J.M., Chuma, G.B., Safina, F.B., Cishesa, T.H., Kavange, A.B., Masumbuko, D.R., Kazadi, J.W., Mushagalusa, G.N. and Lubobo, A.K. (2023). Optimization of plant density and fertilizer application to improve biofortified common bean (Phaseolus vulgaris L.) yield on Nitisols of South-Kivu, Eastern D.R. Congo. Heliyon. 9(6): e17293. https://doi.org/10.1016/j.heliyon.2023.e17293.
  • Submitted24-03-2026|

  • Accepted20-08-2026|

  • First Online 17-11-2026|

  • doi 10.18805/LR-5660

Background: Haricot bean is an important legume crop contributing to food security and income generation in many developing countries. However, its productivity is severely constrained by major foliar diseases such as angular leaf spot, bacterial blight and bean rust. Traditional disease identification relies on manual inspection, which is time-consuming, subjective and prone to error. Therefore, developing an accurate and automated disease detection system is essential for early diagnosis and effective crop management.

Methods: In this study, a modified AlexNet-based convolutional neural network was developed for automated identification of foliar diseases of haricot bean. A total of 3,447 leaf images belonging to four classes-angular leaf spot, bacterial blight, bean rust and healthy leaf-were used. The dataset was divided into training, validation and testing subsets using an 80:10:10 ratio. Data augmentation techniques, including rotation, flipping, zooming and brightness adjustments, were applied to enhance model generalization. Batch normalization and dropout layers were incorporated into the original AlexNet architecture to improve training stability and reduce overfitting. The model was trained for 50 epochs using categorical cross-entropy loss.

Result: The proposed model achieved 96.75% training accuracy and 88.41% validation accuracy. On the independent test set, it attained 99.13% accuracy with high precision, recall and F1-scores across all classes. ROC and precision-recall analyses produced AUC and AP values of 1.0000, demonstrating excellent classification capability. The results confirm that the developed model provides a reliable and efficient approach for automated haricot bean leaf disease detection.

Food security in many developing countries depends heavily on grain legumes, among which haricot bean (Phaseolus vulgaris L.; 2n = 22) holds a prominent position. As a diploid, self-pollinated species belonging to the Fabaceae family, it is cultivated on more than 35 million hectares globally, with an annual production exceeding 30 million tonnes and an average productivity of approximately 800-900 kg/ha (FAO, 2023). It is recognized as one of the most important food and cash crops due to its high nutritional value, adaptability and market demand (Huang et al., 2020; Ahmed et al., 2025). Globally, Africa contributes over one-fifth of total common bean production, ranking third after Asia and the Americas (FAO, 2023). Countries in East Africa, including Ethiopia, Kenya, Uganda and Tanzania, represent major production centers (Pandey and Kumar, 2024; Tigist et al., 2023).
       
In Ethiopia, haricot bean production has expanded rapidly in recent years, reflecting its high market demand and role in food security (CSA, 2022). The crop contributes significantly to national agricultural GDP and smallholder farmer income, with productivity levels ranging between 1.2-1.8 t/ha under improved management conditions (Krishna and Meenakshi, 2022). However, biotic stresses such as angular leaf spot, bacterial blight and bean rust are among the most destructive foliar diseases, causing yield losses ranging from 20% to 80% depending on disease severity, environmental conditions and management practices, resulting in substantial economic losses to farmers (Zamukulu et al., 2023; Etana and Nebiyu, 2023). Traditional disease identification in the field is typically conducted through visual inspection by agronomists or farmers, which is labor-intensive, time-consuming and often results in misclassification due to symptom similarity and human subjectivity (Tavares et al., 2024). In contrast, recent studies have demonstrated that automated image-based approaches using machine vision can significantly improve diagnostic accuracy, consistency and scalability, enabling rapid and real-time disease detection (Cholo et al., 2023; Naik et al., 2023).
       
In recent years, deep learning (DL) methods, particularly convolutional neural networks (CNNs), have emerged as highly effective tools for plant disease classification. A recent systematic review highlights that CNN architectures have achieved consistently high accuracy across diverse plant species and disease types, often exceeding 90% in multi-class classification tasks when trained on sufficient image data (Broughton et al., 2003; Pacal et al., 2024). Studies have also demonstrated that optimised CNN models, such as AlexNet, EfficientNet and MobileNet, can achieve high accuracy and enable real-time deployment on resource-limited devices (Mehta et al., 2025; Mohammad et al., 2026; Ozçelik et al., 2026).
       
Among CNN models, AlexNet remains one of the most influential deep learning architectures used in plant disease detection research due to its balance of performance and computational efficiency (Paek et al., 2026; Souza et al., 2026; Sriram and Kumari, 2026). Chen et al., (2022) and Ramya and Kumar (2023) showed that modified and transfer-learning-based AlexNet models can deliver strong accuracy for tomato disease detection. Mishra et al., (2025) conducted a comparative analysis across CNN, AlexNet, ResNet and EfficientNet, where EfficientNet achieved superior generalization performance. Similarly, Ghosh et al., (2025) introduced a hybrid deep learning architecture combining advanced networks such as ViT and EfficientNetV2, achieving 98% accuracy. In rice disease detection, Akter et al., (2024) reported exceptionally high accuracy (99.99%) using a CNN-based approach. Despite substantial progress, most existing research has focused on common crops such as tomato, maize and potato, with relatively fewer studies concentrating specifically on haricot bean leaf disease classification using deep learning frameworks. Moreover, challenges such as class imbalance, environmental variation and limited annotated datasets often hinder model generalization in real field conditions.
       
This study aims to develop an enhanced AlexNet framework specifically for multi-class classification of haricot bean leaf diseases, with the following objectives:
• Design a modified AlexNet architecture tailored for haricot bean leaf image classification.
• Incorporate data augmentation to improve robustness against environmental variability.
• Apply class-weighted training to address dataset imbalance.
• Evaluate model performance using comprehensive metrics.
Dataset description
 
This study utilized a total of 3,447 haricot bean leaf images categorized into four classes: angular leaf spot, bacterial blight, bean rust and healthy leaves. Angular leaf spot is characterized by small, water-soaked lesions that enlarge into angular brown spots restricted by leaf veins; bacterial blight appears as irregular necrotic patches with yellow halos and water-soaked margins; bean rust is identified by reddish-brown pustules primarily on the underside of leaves; whereas healthy leaves exhibit uniform green coloration without visible lesions or discoloration. The images were collected and organized to represent realistic disease variations commonly observed under field conditions. To ensure reliable model evaluation and to prevent information leakage, the dataset was divided into training, validation and testing subsets following an 80:10:10 ratio. The training set contained 2,066 images, the validation set consisted of 690 images and the testing set included 691 images. This stratified division ensured that each class was proportionally represented across all subsets. Fig 1 illustrates representative samples from each class used in the classification task.

Fig 1: Representative haricot bean leaf classes used in the study.


 
Reproducibility setup
 
To guarantee experimental reproducibility, fixed random seeds were applied across different computational libraries. Random seed initialization ensures that stochastic processes such as dataset shuffling, weight initialization and batch sampling remain consistent across repeated runs. Seeds were set for Python’s built-in random module, NumPy and TensorFlow frameworks. The seed value (42) was consistently applied across all experiments to ensure replicability. Mathematically, reproducibility ensures that the initialization function W0 = f (seed) produces identical initial weights W0 when the same seed value is used. This step enhances experimental reliability and enables fair performance comparison.
 
Image preprocessing
 
All images were resized to 224 × 224 × 3 pixels to match the input requirement of the AlexNet architecture. Additionally, images were checked for noise, distortion and inconsistent illumination to ensure data quality before training. Each image was normalized to scale pixel values into the range [0, 1], computed as:
 

 
Where,
X= The original pixel intensity values. 
       
Normalization reduces numerical instability and accelerates convergence during training by preventing large gradient fluctuations.
 
Data augmentation
 
To improve generalization and minimize overfitting, real-time data augmentation was applied exclusively to the training dataset. Augmentation artificially increases dataset diversity by introducing controlled transformations that simulate real agricultural field variations. The applied transformations included horizontal and vertical flipping, random rotation (20%), random zooming (10%), brightness adjustment (20%) and contrast modification (20%).
       
These transformations can be mathematically expressed as:
 
X′ = T (X)
 
Where,
T= A transformation function applied to the original image X.
       
Augmentation improves robustness against changes in lighting conditions, camera angle, leaf orientation and scale variations encountered in practical scenarios.
 
Class weight calculation
 
Although the dataset was relatively balanced, minor variations in class frequency may bias the learning process toward majority classes. To address this issue, class weights were calculated and incorporated into the loss function. Class weights were computed using:
 

 
Where,
N= Total number of samples.
k= Number of classes.
ni= Number of samples in class i.

The computed class weights are shown in Table 1.

Table 1: Computed class weights.



These weights help balance the contribution of each class during training and reduce classification bias.
 
Proposed AlexNet architecture
 
In this work, a modified AlexNet architecture was implemented using a functional API for four-class bean leaf disease classification. The network receives an input tensor X ∈ ℝ 224 × 224 × 3, representing RGB images resized to 224 × 224 pixels. The first convolutional layer applies 96 learnable filters to extract low-level spatial features such as edges, color gradients and simple textures. In general, the convolution operation can be expressed as:
 

 
Where,
W= Convolutional kernel weights.
bk= Bias term for the kth filter.
Yi,j,k= Output feature map.
       
This first layer produces an output of size 54 × 54 × 96 with 34,944 parameters. A Rectified Linear Unit (ReLU) activation function is applied to introduce non-linearity:
 
f(x) = max (0, x)
 
Batch normalization follows to stabilize the learning process by normalizing intermediate activations. The normalization process is defined as:
 

 
Where
μ and σ2= Batch mean and variance.
γ and β= Learnable scaling and shifting parameters.
       
A max-pooling layer then reduces spatial dimensions to 26 × 26 × 96 defined as:
 

 
Where,
R= The pooling region.
       
The second convolutional layer increases feature depth to 256 channels while preserving spatial dimensions at 26 × 26, containing 614,656 parameters. This layer captures more complex patterns such as vein structures and disease-specific textures. Batch normalization and a second max-pooling layer reduce the output to 12 × 12 × 256. The third convolutional layer produces 384 feature maps of size 12 × 12 × 384 with 885,120 parameters, enabling extraction of higher-level morphological features. After batch normalization, the fourth convolutional layer maintains 384 channels with 1,327,488 parameters, further refining hierarchical representations. The fifth convolutional layer reduces channel depth to 256 with 884,992 parameters, followed by batch normalization and a final max-pooling layer, producing compact feature maps of size 5 × 5 × 256.
       
The resulting tensor is flattened into a vector of length 6,400, which serves as input to the fully connected layers. The first dense layer contains 2,048 neurons and 13,109,248 parameters. The dense transformation is computed as:
 
z = Wx + b
 
Where,
W= Weight matrix.
b= Bias vector.
       
ReLU activation is again applied, followed by dropout regularization to prevent overfitting. Dropout randomly deactivates neurons during training according to:
 
 
Where,
ri ~ Bernoulli (p) and p = The retention probability.
       
The second dense layer contains 1,024 neurons with 2,098,176 parameters and is also followed by dropout. The final output layer consists of 4 neurons corresponding to the four bean leaf classes. A Softmax activation function converts logits into probability distributions:

 
Ensuring that predicted probabilities sum to one. The model is trained using categorical cross-entropy loss, defined as:
 

 
Where,
yk= The true label.
yk= The predicted probability.
       
Overall, the architecture contains 18,964,228 total parameters, of which 18,961,476 are trainable and 2,752 are non-trainable, primarily associated with batch normalization statistics. By integrating deep convolutional feature extraction, normalization, pooling and fully connected classification layers with dropout regularization, the proposed AlexNet-based model effectively learns discriminative features for accurate bean leaf disease classification while maintaining training stability and generalization capability. The model was trained for 50 epochs with a batch size of 32. Fig 2 illustrates the overall architecture of the proposed model.

Fig 2: Modified AlexNet architecture used for haricot bean disease classification.


 
Performance evaluation metrics
 
Model performance was assessed using accuracy, precision, recall, F1-score and confusion matrix analysis. Accuracy is defined as:
     

 
Precision, recall and F1-score are calculated as:
 





 
The confusion matrix provides a class-wise breakdown of prediction performance and helps identify misclassification patterns.
Fig 3 presents the training and validation accuracy and loss curves obtained over 50 epochs. The training accuracy shows a rapid increase during the early epochs, indicating that the model quickly learned meaningful visual features from the haricot bean leaf images. As training progressed, the accuracy continued to improve gradually and stabilized at 96.75% by the final epoch. Similarly, the validation accuracy followed an upward trend and reached 88.41%, demonstrating that the model maintained good generalization performance on unseen validation data. The loss curves further support this observation. Training loss decreased steadily throughout the training process, reflecting effective optimization of the categorical cross-entropy objective function. Validation loss also showed a general downward trend, although minor fluctuations were observed across epochs. These variations are expected in stochastic gradient-based learning due to mini-batch updates and real-time data augmentation. Importantly, there was no significant divergence between training and validation curves, suggesting that overfitting was effectively controlled. The incorporation of dropout layers, batch normalization and augmentation techniques contributed significantly to maintaining stable convergence.

Fig 3: Training and validation accuracy and loss curves.


       
Although a moderate gap exists between training and validation accuracy, the difference remains within an acceptable range for deep learning models trained on real-world agricultural datasets. This indicates that the model learned robust and discriminative features such as lesion patterns, rust formations and discoloration characteristics associated with different disease classes. Evaluation on the independent test set further confirms the effectiveness of the proposed model. The network achieved a test accuracy of 99.13% with a test loss of 1.7738, demonstrating strong predictive capability on unseen haricot bean leaf images. The high test performance indicates excellent generalization ability and suggests that the trained AlexNet-based architecture successfully captured disease-specific visual patterns. Overall, the results validate the suitability of the proposed approach for accurate and reliable automated bean leaf disease classification in practical agricultural applications.
       
Fig 4 presents the confusion matrix obtained from the test dataset for the four-class haricot bean leaf disease classification task. The matrix provides a detailed breakdown of correct and incorrect predictions made by the model across the classes. The results indicate that the model achieved near-perfect classification performance for most categories. Specifically, 173 samples of angular leaf spot, 172 samples of bacterial blight and 173 samples of bean rust were correctly classified, with no misclassifications observed for these three classes. This demonstrates that the model effectively captured distinctive disease-related visual characteristics such as lesion boundaries, discoloration patterns and rust pustules. For the healthy leaf category, 167 samples were correctly classified, while 6 samples were misclassified as bean rust. This minor misclassification may be attributed to the presence of very early or mild rust symptoms, which are visually indistinguishable from healthy leaf surfaces, as well as variations in illumination, background noise and image acquisition conditions. Nevertheless, the overall classification performance remains exceptionally high. The dominance of diagonal elements and the minimal off-diagonal values confirm excellent class separability and strong generalization capability.

Fig 4: Confusion matrix of the proposed AlexNet model.


       
Table 2 presents the detailed classification report of the proposed AlexNet-based model evaluated on the test dataset. The results show outstanding performance across all four categories. Both angular leaf spot and bacterial blight achieved perfect precision, recall and F1-score values of 1.0000, indicating that all samples in these classes were correctly identified without any false positives or false negatives. For the bean rust class, the model achieved a precision of 0.9665, recall of 1.0000 and an F1-score of 0.9830. The perfect recall indicates that all actual bean rust samples were correctly detected, although a small number of samples from other classes were predicted as bean rust, slightly lowering the precision. Similarly, the healthy leaf class achieved a precision of 1.0000, recall of 0.9653 and an F1-score of 0.9824. The slightly lower recall value corresponds to the small number of healthy samples misclassified as bean rust, as observed in the confusion matrix. The overall test accuracy reached 99.13%, with both macro and weighted averages of precision, recall and F1-score equal to 0.9913. The macro average indicates balanced performance across all classes, while the weighted average confirms that class distribution did not bias the results.

Table 2: Classification report of the proposed model.


       
Fig 5 illustrates the receiver operating characteristic (ROC) curves and precision-recall (PR) curves for all four classes. All four classes achieved an area under the curve (AUC) value of 1.0000, indicating perfect separability between positive and negative samples for each category. The ROC curves closely follow the top-left boundary of the graph, which reflects an optimal classifier with maximum sensitivity and specificity. Similarly, the precision-recall curves demonstrate exceptional classification performance. The average precision (AP) for each class is reported as 1.0000, indicating near-perfect precision across all recall levels. The PR curves remain close to the upper-right corner of the graph, further validating the model’s strong predictive capability, especially in correctly identifying positive disease cases.

Fig 5: ROC and precision-recall curves of the proposed model.


       
Fig 6 presents representative test samples along with their predicted labels and confidence scores generated by the proposed AlexNet-based model. The first sample corresponds to a healthy leaf, correctly classified as healthy leaf with a confidence score of 0.9991, demonstrating the model’s ability to accurately recognize normal leaf texture and color distribution. The second image represents bean rust, which was correctly predicted with a perfect confidence of 1.0000, highlighting the model’s strong capability in detecting rust-specific visual characteristics such as yellowing and lesion patterns. The third example shows a bacterial blight leaf correctly classified with a confidence of 0.9991, indicating that the model effectively captures irregular necrotic lesions and edge discoloration associated with this disease. Similarly, the fourth image of angular leaf spot was accurately predicted with a confidence score of 0.9997, reflecting precise detection of angular-shaped dark lesions and scattered spotting patterns. The consistently high confidence values across all examples demonstrate the robustness and reliability of the trained model. These qualitative results further support the quantitative findings from the confusion matrix, classification report and ROC analysis, confirming the effectiveness of the proposed approach for automated haricot bean leaf disease classification.

Fig 6: Sample predictions of the proposed AlexNet model.


       
Previous studies have applied AlexNet and other deep learning architectures for plant disease classification across different crops, demonstrating varying levels of performance depending on dataset size, crop type and model design. Chen et al., (2022) utilized a modified AlexNet on a large-scale tomato dataset consisting of 22,930 images across 10 classes and achieved an accuracy of 98%, highlighting the effectiveness of CNN-based models for multi-class classification tasks. Similarly, Kewat (2023) applied AlexNet to an ImageNet-based plant disease and pest dataset and reported 96.50% accuracy, confirming its suitability for agricultural image analysis.
       
In contrast, studies using smaller datasets reported relatively lower performance. Sholihin et al., (2024) implemented a modified AlexNet for cassava classification using 1,400 images and achieved 87% accuracy, indicating the influence of dataset size and variability on model performance. Fang et al., (2025) applied AlexNet for bean leaf disease classification (angular leaf spot, rust and healthy) and obtained 95.4% training accuracy but only 82.81% overall accuracy, suggesting challenges in distinguishing visually similar classes under real conditions.Furthermore, comparative architecture studies highlight the advantage of advanced models. Mishra et al., (2025) evaluated CNN, AlexNet, ResNet and EfficientNet on potato (2,152 images) and mango (4,000 images) datasets, where EfficientNet achieved the highest accuracy of 97.8%, outperforming other architectures in terms of generalization.
       
Compared to these existing studies, the present work demonstrates superior performance. The proposed modified AlexNet model, enhanced with batch normalization and dropout, achieved 99.13% test accuracy on a dataset of 3,447 haricot bean images. Despite using a moderate-sized dataset, the model shows improved robustness and generalization capability, outperforming several previously reported approaches. This confirms the effectiveness and competitiveness of the proposed architecture for plant disease classification tasks.
This study presented a modified AlexNet-based convolutional neural network for automated haricot bean leaf disease classification. The model demonstrated high accuracy, strong generalization capability and excellent discriminative performance across four classes. The integration of batch normalization, dropout and data augmentation effectively minimized overfitting and improved robustness. Despite the promising results, the study has some limitations. The dataset was relatively limited in size and collected under controlled conditions, which may not fully represent complex field environments. Additionally, only a single deep learning architecture was extensively evaluated. Future work will focus on expanding the dataset with diverse field images, incorporating real-time mobile-based deployment and comparing the model with advanced architectures such as EfficientNet and Vision Transformers. Integrating explainable AI techniques may further enhance model transparency and practical adoption in smart agriculture systems.
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 the aspects of this work.
 
Data availability
 
The data analysed/generated in the present study will be made available from corresponding authors 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.
Authors declare that they have no conflict of interest.

  1. Ahmed, E.H., Teka, T.A., Tolera, K.D., Gemede, H.F. and Hussen, K.E. (2025). Technological properties of improved haricot bean (Phaseolus vulgaris L.) varieties cultivated in Ethiopia. Applied Food Research. 5(2): 101107. https://doi.org/10.1016/j.afres.2025.101107.

  2. Akter, S., Sumon, R.I., Ali, H. and Kim, H. (2024). Utilizing convolutional neural networks for the effective classification of rice leaf diseases through a deep learning approach. Electronics. 13(20): 4095. https://doi.org/10.3390/electronics13204095.

  3. Broughton, W. J., Hernández, G., Blair, M., Beebe, S., Gepts, P. and Vanderleyden, J. (2003). Beans (Phaseolus spp.)- Model food legumes. Plant and Soil. 252(1): 55-128. https://doi.org/10.1023/a:1024146710611.

  4. Central Statistical Agency (CSA). (2022). Agricultural Sample Survey Series: Report on Area and Production for Major Crops (Private Holdings, Main Season). Addis Ababa, Ethiopia.

  5. Chen, H., Widodo, A.M., Wisnujati, A., Rahaman, M., Lin, J.C., Chen, L. and Weng, C. (2022). ALEXNET convolutional neural network for disease detection and classification of tomato leaf. Electronics. 11(6): 951. https://doi.org/10.3390/electronics11060951.

  6. Cholo, M., Marisennayya, S., Bojago, E., Leja, D. and Divya, R. (2023). Determinants of adoption and intensity of improved haricot bean (Phaseolus vulgaris L.) varieties: A socio-agronomic study from southern Ethiopia. Journal of Agriculture and Food Research. 13: 100656. https://doi.org/10.1016/j.jafr.2023.100656.

  7. Etana, D. and Nebiyu, A. (2023). Response of common bean (Phaseolus vulgaris L.) to lime and TSP fertilizer under acid soil. Heliyon. 9(4): e15176. https://doi.org/10.1016/j.heliyon.2023.e15176.

  8. Fang, H., Sun, Y., Kim, T., Xu, J. and Deng, Q. (2025). The role of deep learning in enhancing crop sustainability: A study on AlexNet’s application in detecting bean leaf disease. Legume Research-An International Journal. 49(4): 701- 708. doi: 10.18805/LRF-843.

  9. Food and Agriculture Organization of the United Nations (FAO). (2023). Faostat. https://www.fao.org/faostat.

  10. Ghosh, H., Rahat, I.S., Emon, M.M.R., Mashrafi, M.J., Tanzin, M.A. A.A., Mohanty, S.N. and Kant, S. (2025). Advanced neural network architectures for tomato leaf disease diagnosis in precision agriculture. Discover Sustainability. 6(1). https://doi.org/10.1007/s43621-025-01149-1.

  11. Huang, Y., Liu, X., Cao, D., Chen, G., Li, S., Wang, G., Wang, J. and Xu, S. (2020). Cross-species amplification of common bean (Phaseolus vulgaris) EST-SSRs within hyacinth bean, pea and soybean. Legume Research-An International Journal. 44(3): 287-291. doi: 10.18805/LR-574.

  12. Kewat, K. (2023). Plant Disease Classification using Alex Net. Research Square (Research Square). https://doi.org/ 10.21203/rs.3.rs-2612739/v1.

  13. Krishna, K.L. and Meenakshi, J.V. (2022). Agricultural productivity growth and structural transformation in rural India: Some recent evidence. Journal of Quantitative Economics. 20(S1): 277-302. https://doi.org/10.1007/s40953-022-00321-y.

  14. Mehta, A.R., Kumar, P., Prem, G., Aggarwal, S. and Kumar, R. (2025). AI-powered innovations in agriculture: A systematic review on plant disease detection and classification. Indian Journal of Agricultural Research. 59(9): 1321-1330. doi: 10.18805/IJARe.A-6371.

  15. Mishra, U., Pandey, A., G, L. and K, T. (2025). Deep learning-based disease detection in potato and mango leaves: A comparative study of CNN, AlexNet, ResNet and EfficientNet. Scientific Reports. 16(1): 2788. https://doi.org/10.1038/s41598-025-32607-5.

  16. Mohammad, A.A.S., Mohammad, S.I., Vasudevan, A. and Heindric, A. (2026). Intelligent decision support system for crop management using data analytics. Journal of Experimental Biology and Agricultural Sciences. 14(3): 9-14. https://jebas.org/ojs/index.php/jebas/article/view/3796.

  17. Naik, N.K., Sethy, P.K., Behera, S.K. and Amat, R. (2023). A methodical analysis of deep learning techniques for detecting Indian lentils. Journal of Agriculture and Food Research. 15: 100943. https://doi.org/10.1016/j.jafr.2023.100943.

  18. Ozçelik, H., Sağlam, B. and Öcal, A.S. (2026). New natural methods for combating mold fungi in agricultural products: The isparta example. Bulletin of Pure and Applied Sciences- Botany. 45B(2): 1-17. https://bpasjournals.com/botany/index.php/journal/article/view/174/150.

  19. Pacal, I., Kunduracioglu, I., Alma, M.H., Deveci, M., Kadry, S., Nedoma, J., Slany, V. and Martinek, R. (2024). A systematic review of deep learning techniques for plant diseases. Artificial Intelligence Review. 57(11). https://doi.org/10.1007/s10462-024-10944-7.

  20. Paek, J.J., Kim, J., Bae, S., Paek, K.,and Lee, Y. (2026). Chive- fortified fermentation enhances gamma-aminobutyric acid production by Levilactobacillus brevis PL9014. International Journal of Probiotics and Prebiotics. 21(1): 1-6. https://doi.org/10.37290/ijpp.v21i1.12.

  21. Pandey, M. and Kumar, N. (2024). Diversity of traditional grain legumes of Himalayan region of Uttarakhand: A review. Agricultural Reviews. 45(1): 96-102. doi: 10.18805/ag.R-2362.

  22. Ramya, R. and Kumar, P. (2023). High-performance deep transfer learning model with batch normalization based on multiscale feature fusion for tomato plant disease identification and categorization. Environmental Research Communications. 5(12): 125015. https://doi.org/10.1088/2515-7620/ace594.

  23. Sholihin, M., Fudzee, M.F.M., Ismail, M.N., Wati, E.N., Arshad, M.S. and Gusman, T. (2024). Modified alexnet architecture for classification of cassava based on leaf images. JOIV International Journal on Informatics Visualization. 8(3): 1074. https://doi.org/10.62527/joiv.8.3.2966.

  24. 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.

  25. 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://bpasjournals.com/life-sciences/index.php/journal/article/view/150/140.

  26. Tavares, M.S., Silva, C.A. A. C., Regazzo, J.R., De Souza, S.E.J., Da Silva, T.L., Fiorio, P.R. and Baesso, M.M. (2024). Performance of machine learning models in predicting common bean (Phaseolus vulgaris L.) crop nitrogen using NIR spectroscopy. Agronomy. 14(8): 1634. https://doi.org/10.3390/agronomy14081634.

  27. Tigist, S.G., Sibiya, J., Amelework, A. and Keneni, G. (2023). Agromorphological and physiological performance of ethiopian common bean (Phaseolus vulgaris L.) genotypes under different agroecological conditions. Plants. 12(12): 2342. https://doi.org/10.3390/plants12122342.

  28. Zamukulu, P.M., Bagula, E.M., Mondo, J.M., Chuma, G.B., Safina, F.B., Cishesa, T.H., Kavange, A.B., Masumbuko, D.R., Kazadi, J.W., Mushagalusa, G.N. and Lubobo, A.K. (2023). Optimization of plant density and fertilizer application to improve biofortified common bean (Phaseolus vulgaris L.) yield on Nitisols of South-Kivu, Eastern D.R. Congo. Heliyon. 9(6): e17293. https://doi.org/10.1016/j.heliyon.2023.e17293.
In this Article
Published In
Legume Research

Editorial Board

View all (0)