Current IssueSpecial IssueEditorial BoardOnline First ArticlesArchive
Current IssueSpecial IssueEditorial BoardOnline First ArticlesArchive

Full Research Article

Deep Learning Framework for the Automated Identification and Classification of Mango Disease, Pest and Healthy Conditions

P
Prashant Chaudhary1,*
https://orcid.org/0000-0001-5221-6274
T
Taniya Sengupta Rathore2
https://orcid.org/0000-0002-9987-1871
D
Durgaprasad Navulla3
https://orcid.org/0000-0001-5942-6864
B
Bharati Chandrayan4
https://orcid.org/0000-0002-5691-136X
K
Kunal D. Jadhav5
https://orcid.org/0000-0001-8331-5363
A
Alok S. Shah6
https://orcid.org/0000-0001-5708-3546
1Department of Information Technology, Usha Pravin Gandhi College of Arts, Science and Commerce, SVKM, Mumbai-400 056, Maharashtra, India.
2Department of Biotechnology, School of Science, O.P. Jindal University, Raigarh-49 6109, Chhattisgarh, India.
3KL Business School, Koneru Lakshmaiah Education Foundation, Guntur-521 180, Andhra Pradesh, India.
4Department of Language, Culture and Society, SRM Institute of Science and Technology Delhi NCR Campus, Ghaziabad-201 204, Uttar Pradesh, India.
5Department of Lifelong Learning and Extension, University of Mumbai, Mumbai-400 020, Maharashtra, India.
6Bharati Vidyapeeth (Deemed to be) University, Department of Management Studies, Kharghar, Navi Mumbai-410 210, Maharashtra, India.

ABSTRACT

Background: Mango (Mangifera indica) is a commercially important fruit crop, but its yield and quality are often threatened by diseases and insect pests. Among the most common are fungal diseases such as sooty mould and powdery mildew and insect pests like gall midge. If not detected early, these cause substantial economic losses to farmers. Manual identification is time-consuming and error-prone, creating the need for automated solutions. Deep learning, particularly Convolutional Neural Networks (CNNs), has shown strong potential in plant disease and pest recognition.
Methods: In this study, a sequential CNN-based framework was developed for the automated identification and classification of mango leaf images into four classes: Sooty mould, powdery mildew, gall midge and healthy. The architecture consisted of five convolutional layers with max-pooling for feature extraction, followed by fully connected layers for classification. Model performance was assessed using accuracy, precision, recall, F1-score and confusion matrix analysis.
Result: The model achieved an overall accuracy of 96.0%, with high weighted average precision, recall and F1-scores, indicating reliable performance despite class imbalance. The confusion matrix confirmed the model’s capability to distinguish between disease, pest and healthy conditions with minimal misclassification.

INTRODUCTION

Fruits and grains are essential sources of nutrients that play a vital role in human health. Among these, fruits are particularly important because they supply dietary fibre, carotenoids, phytosterols, folic acid, vitamins, minerals and antioxidants. A growing body of evidence highlights the benefits of medicinal plants and bioactive phytocompounds in preventing and managing diseases (Adeleye et al., 2021). Polyphenols, for example, are known to scavenge free radicals and exert anti-carcinogenic effects, contributing significantly to human well-being (Mustafa et al., 2020). Regular consumption of fruits lowers the risk of cancer and helps prevent obesity, cardiovascular disease and chronic respiratory conditions (Noce et al., 2021). Thus, fruits are considered indispensable for a balanced diet and long-term health.
       
Despite their importance, access to nutrient-rich fruits remains limited in many regions due to global challenges. Population growth, climate change and resource scarcity threaten food security, particularly in developing and underdeveloped nations. The food and agriculture organization reported that approximately 258 million people worldwide suffered from severe food shortages, the highest level in seven years (FAO, 2023). The rising population, projected to increase by 22% in the next five years, will continue to stress agricultural systems, affecting political stability, economic growth and environmental sustainability. This scenario underlines the need for crop protection, yield improvement and sustainable farming strategies.
       
Among tropical fruits, mango (Mangifera indica) holds special significance. Known as the “king of fruits,” mango is cherished for its flavour, aroma and nutritional value (Hussain et al., 2021). Native to India and Southeast Asia, mango has been cultivated for over 4,000 years and is now grown across Asia, Africa, Central America, Australia and parts of Europe. Hundreds of varieties exist globally, although only a small number are commercially produced on a large scale (Rajan and Srivastav, 2021). Mango contributes substantially to both local economies and global trade. However, its productivity and quality are often hampered by pests and diseases that attack leaves, flowers and fruits, leading to substantial economic losses.
       
Several diseases and pests pose serious challenges to mango cultivation. Among the most damaging are gall midge, powdery mildew and sooty mould. Gall midges are small flies that infest mango leaves, flowers and fruits. More than 16 species are reported to attack mango worldwide, particularly in Asia. Common species include Erosomya indica, Erosomya mangiferae, Asynapta mangiferae, Gephyraulus mangiferae and Procontarinia matteiana (https://www.mango.org/wp-content/uploads/2020/08/Mango_Pests_and_ Diseases_ENG.). The larvae feed on plant tissues, producing black spots on leaves and affecting growth. These pests significantly reduce yield, either by lowering fruit quality or quantity (Reddy et al., 2020; Khan et al., 2020). Sooty mould is another widespread problem caused by fungi such as Meliola mangiferae, Capnodium mangiferae, Capnodium ramosum and Tripospermum acerinum. The disease is characterised by the presence of a black, velvety layer on leaves and twigs, resulting from fungal spores adhering to honeydew secretions produced by insects. In severe cases, entire leaf surfaces become blackened, reducing photosynthesis and leading to curling and shrivelling of leaves (Singh et al., 2022). Powdery mildew, caused by Pseudoidium anacardii, is considered one of the most economically significant fungal diseases of mango worldwide. Its symptoms appear initially on young leaves and inflorescences as whitish, powdery growth. If untreated, the infection spreads rapidly, leading to flower and fruit drop (Iqbal et al., 2024).
       
Recent advances in computer vision (CV), machine learning (ML) and deep learning (DL) have opened new opportunities for smart agriculture (Sharma et al., 2022). Machine learning has proven especially valuable in disease detection due to its strong pattern recognition capabilities. By analysing leaf colour, texture and shape, ML algorithms can distinguish between healthy and diseased leaves with high accuracy (Mia et al., 2020). Applications of ML extend beyond agriculture to areas such as biometrics, face recognition, food classification, animal disease detection and medical diagnostics (AlZubi, 2023; Wasik and Pattinson, 2024; Kusuma et al., 2022). In agriculture, ML-based systems can process large datasets, extract hidden features and adapt to complex variations in disease symptoms, making them suitable for real-time applications.
       
Several studies have demonstrated the potential of ML in mango disease detection. Mia et al., (2020) developed a neural network ensemble combined with a support vector machine to classify mango leaf diseases. They used gray-level Co-occurrence matrix (GLCM) features to capture leaf texture, achieving reliable recognition accuracy. Gulavnai and Patil (2019) employed a convolutional neural network (CNN) with a retrained ResNet architecture to detect mango leaf diseases. CNNs are particularly effective because they automatically learn features from raw image data, eliminating the need for manual feature extraction.
       
Deep learning models such as CNNs, ResNet and VGG have gained popularity in recent years for agricultural applications. Their hierarchical structure allows them to capture both low-level and high-level features, improving classification performance. CNNs are also computationally efficient, making them suitable for mobile applications and resource-limited settings. Compared to more complex models, CNNs strike a balance between accuracy and implementation simplicity (Shickel and Rashidi, 2020; Zock et al., 2024).
       
Early detection and effective management of mango diseases are therefore critical. Traditional methods rely on farmers’ experience and laboratory-based testing. While reliable, these methods are time-consuming, costly and impractical for large-scale monitoring. Moreover, accurate identification often requires expert knowledge that may not be accessible to farmers in rural or resource-limited settings. In addition, visual symptoms of diseases can be misleading due to overlapping signs, environmental stress, or nutrient deficiencies. These limitations highlight the urgent need for automated, rapid and accurate approaches for disease identification.
       
The present study contributes to this effort by exploring deep learning-based frameworks for the automated identification and classification of mango diseases and pests, including gall midge, powdery mildew and sooty mould. By leveraging CNN architectures, this study aims to demonstrate the effectiveness of AI-driven tools for real-time disease diagnosis in mango cultivation. The findings are expected to support precision agriculture practices, enhance yield quality and provide a scalable solution adaptable to other crops and regions.

MATERIALS AND METHODS

Data description
 
The present study employed the publicly available MangoLeafBD dataset hosted on the Mendeley repository. The dataset contains a total of 2,494 mango leaf images, categorized into four distinct classes: Gall midge (Pest-infected), healthy, powdery mildew (Fungal infection) and sooty mould (Fungal infection) (Fig 1). The images were collected using a standard smartphone camera under natural lighting conditions in the orchards of Sher-e-Bangla Agricultural University and Jahangirnagar University, Bangladesh (Ali et al., 2022).

Fig 1: Images from the dataset.



Pre-processing of raw data
 
Using image data to create CV and ML models presents a number of challenges because of issues with sufficiency, accuracy and complexity (Charte et al., 2021). For this reason, data preprocessing, that is, cleaning and reshaping the photos into an appropriate format, is essential before creating any machine learning models in order to get the intended results (Cho, 2024).
       
Raw image data required systematic pre-processing to ensure uniformity and improve model performance. All images were resized to 256 x 256 pixels to standardize input dimensions. Data augmentation was applied using the following values: rotation (20°), width and height shifts (0.2), shear (0.2), zoom (0.2) and horizontal flipping. These techniques increased dataset variability and reduced the risk of overfitting. Pixel values were normalized to a [0,1] range to lower computational complexity and support faster convergence. This ensured that the CNN extracted meaningful features without being biased by differences in size, orientation, or illumination.
 
Sequential-CNN architecture
 
A flowchart of CNN is presented in Fig 2. It classifies mango leaf images into four categories (Gall midge, healthy, powdery mildew and sooty mould). The model accepted pre-processed 256 x 256 RGB images as input.

Fig 2: Flow diagram of the machine learning process.


       
The architecture of the CNN model had a sequential design and is presented in Fig 3. It comprised multiple layers, each with a specific role in feature extraction and classification. The input layer received resized 256 x 256 images with three color channels (RGB), ensuring uniform dimensions and pixel scaling. The first convolutional layer applied 32 filters of size 3 x 3 to detect simple, low-level features such as edges, corners and textures, using the ReLU activation function to introduce non-linearity. This was followed by max-pooling layer 1 with a 2 x 2 pooling operation, which reduced feature map size while retaining essential information and lowering computational cost. The second convolutional layer increased the number of filters to 64 (3 x 3) to detect more complex structures, including leaf venation and disease spots, again using ReLU activation. Max-pooling layer 2 further reduced feature map dimensions while preserving critical spatial details. The third convolutional layer, with 128 filters (3 x 3), captured finer disease-specific patterns such as mildew patches and sooty mold textures, followed by max-pooling layer 3 to maintain efficiency and avoid overfitting. The fourth convolutional layer used 128 filters (3 x 3) to refine feature extraction and focus on abstract, high-level visual patterns, with max-pooling layer 4 compressing the extracted features. The fifth and final convolutional layer applied 256 filters (3 x 3) to capture complex and hierarchical representations of pest and disease symptoms and max-pooling layer 5 reduced dimensionality before flattening. The flatten layer transformed pooled feature maps into a one-dimensional vector suitable for fully connected layers. The first dense layer consisted of 64 neurons with ReLU activation to integrate extracted features into meaningful representations for classification. A dropout layer with a rate of 0.5 was applied to randomly deactivate neurons during training, preventing overfitting and improving generalization. The final dense output layer contained 5 neurons representing the four disease categories plus the healthy class, using softmax activation to output class probabilities. This architecture was designed to progressively capture hierarchical features, from basic edges and textures in early layers to complex disease- or pest-specific visual cues in deeper layers, ultimately supporting accurate and automated classification of mango leaf images.

Fig 3: Visualization of the convolutional model proposed in the present study.


 
Training model
 
The CNN model consisted of 2,124,996 trainable parameters, including weights and biases, optimized during training. The dataset was split into training (70%), validation (15%) and testing (15%) subsets to ensure unbiased evaluation. The model was trained for 50 epochs with a batch size of 32. Since the task involved multi-class classification, category cross-entropy loss was used as the objective function. An Adam optimizer with a learning rate of 0.001 was employed to accelerate convergence and avoid local minima. To prevent overfitting, a dropout layer (rate = 0.5) was incorporated between dense layers, which randomly deactivates neurons during training. Training and validation losses were monitored after each epoch to ensure model stability.
 
Metrics to evaluate the performance of the model
 
Four metric variables evaluate the performance of any newly developed model (Sammut and Webb, 2011). They are accuracy, F1 score, Precision and Recall. In addition, a confusion matrix is used.
 
Accuracy
 
Total number of true predictions (i.e. True positive (TP)+ True Negative (TN)) out of all the instances.
 
 
 
Precision
 
Number of true positives of all the true predictions.
 
 
 
Recall
 
Number of true predictions of all the true instances.
 
 
 
F1-score
 
Combines recall and precision. F1 shows how effectively the models make the trade-off between precision and recall.
 
 
 
A confusion matrix shows how accurate a model is in classifying data. An example of a confusion matrix with the three classes A, B and C is displayed in Table 1.

Table 1: Confusion matrix: Example.

RESULTS AND DISCUSSION

The developed CNN model effectively classified images into different disease categories. At final epoch, the model achieved 97.6% validation accuracy and 98.1% training accuracy, indicating strong generalization performance. The training loss was 0.053, while the validation loss was 0.062, showing close agreement and minimal overfitting (Fig 4).

Fig 4: a) Training accuracy and loss, b) Validation accuracy and loss.


       
The actual and predicted classes, along with confidence scores for each category, are presented in Fig 5. The sample predictions show that the model performed with very high accuracy across different classes of mango leaf conditions. In the first case, a leaf affected by sooty mould was correctly identified with a confidence of 97.18%. The second sample, a healthy leaf, was classified as healthy with a confidence of 99.98%, showing the model’s ability to distinguish normal leaves from diseased ones. The third example involved a leaf affected by gall midge, which was perfectly classified with 100% confidence. In the fourth case, a leaf with powdery mildew was also correctly detected, with a confidence score of 99.85%. These results highlight that the model consistently recognized both healthy and diseased leaves with very high certainty. The high prediction confidence across all examples indicates the robustness of the model and its suitability for practical disease identification.

Fig 5: Classification of leaf images by the proposed model.


       
The confusion matrix provides a clear evaluation of the model’s predictions (Fig 6). For gall midge, 54 samples were correctly identified, while 2 were misclassified as sooty mould. No cases were wrongly placed in the healthy or powdery mildew categories. For healthy leaves, 49 were correctly classified, but 2 were wrongly predicted as gall midge and 1 as powdery mildew. For powdery mildew, 54 were classified correctly, with 4 misclassified as sooty mould. No samples were confused with gall midge or healthy. For sooty mould, all 58 samples were classified correctly with no errors in other categories.

Fig 6: Confusion Matrix of diseased and healthy mango leaves.


       
This analysis shows that most predictions matched the actual labels, with very few misclassifications. The errors mainly occurred between classes with somewhat similar visual symptoms, such as powdery mildew and sooty mould. Despite these minor errors, the model demonstrated high reliability in classifying all four categories.
       
The classification report gives a detailed view of the model’s performance (Table 2). For the gall midge class, the model achieved a precision of 0.9643 and a recall of 0.9643, with an F1-score of 0.9643. This indicates balanced and consistent performance. For healthy leaves, precision reached 1.000, showing that all predicted healthy samples were correct. Recall was 0.9423, meaning a few healthy samples were misclassified. The F1-score of 0.9703 reflects this slight drop in recall. For powdery mildew, the model showed strong precision at 0.9818, but recall was 0.931. The F1-score of 0.9558 reflects good overall balance. For sooty mould, the model produced a precision of 0.9062 and a recall of 1.000, which means all sooty mould samples were correctly identified, though a few predictions for this class included errors. The F1-score here was 0.9508.

Table 2: Classification matrices after testing the dataset.


       
The overall accuracy of the model across all classes was 95.98%. The macro average values were 0.9631 for precision, 0.9594 for recall and 0.9603 for F1-score. The weighted averages, which account for class imbalance, were close in value, with precision at 0.9621, recall at 0.9598 and F1-score at 0.9600. These results confirm that the model performs consistently well across all categories, with high reliability in both precision and recall.
       
Fig 7 displays the ROC curves together with the corresponding Area under the curve (AUC). A true positive/false positive ratio for the discrimination cut-off value of the used test is shown by each point on the ROC curve. In essence, it shows how the TPR and FPR trade off at different categorization criteria. The model’s overall performance is measured by overall potential classification thresholds using the Area Under the ROC curve (AUC-ROC) metric (Yang and Berdine, 2017). Regardless of class imbalance, the superior discriminating ability is shown by a higher AUC-ROC value. A high AUC-ROC indicates that the model can successfully discriminate between the minority class and the majority classes in the situation of unbalanced datasets. The Powdery mildew and the Gall midge in the current research had the greatest AUC (0.76 and 0.75), followed by Sooty mould (0.67) and Healthy class (0.62).

Fig 7: ROC(AUC) curve for all diseased classes.


       
The accuracy of the proposed model was found to be competitive when compared with existing studies (Table 3). Rajbongshi et al., (2021) reported 98.0% accuracy using DenseNet201, while Rizvee et al., (2023) achieved 98.55% with LeafNet. Gautam et al., (2023) obtained 98.57% using an ensemble approach. These values are slightly higher than the 96.0% achieved in this study. However, those works relied on larger datasets and included more disease classes. Given the smaller dataset used here, the performance of the present CNN remains notable.

Table 3: Comparative analysis of classification accuracy of the proposed CNN model with existing studies.


       
Several studies have reported results closer to or below the accuracy of this model. Vijay and Pushpalatha (2023) achieved 93.01% with a hybrid EfficientNet model. Pratap and Kumar (2024) reported 94.5% using YOLOv8 with African Buffalo Optimization. Rao et al., (2021) obtained 89% for mango leaf classification using AlexNet in their JIT CROPFIX application. In these comparisons, the current CNN clearly outperformed existing approaches. Kalaivani and Saravanan (2024) achieved a slightly higher accuracy of 98.25% using a segmentation-based ResNet50 framework. However, that method required additional preprocessing, whereas the present model achieved robust results with a simpler workflow.
       
Nalawade et al., (2023) reported 96-97% accuracy by combining thermal and RGB images for anthracnose and powdery mildew classification. Their approach, however, depended on thermal imaging equipment, which limits field-level adoption. In contrast, the current study relied only on RGB images, making it more practical and accessible for large-scale use.
       
In summary, while some ensemble and advanced architectures achieved marginally higher accuracies, the proposed CNN offers competitive results with key advantages. It is simple, requires less computation and incorporates both pest and disease classes. This broader classification scope strengthens its novelty and enhances its potential for real-world precision agriculture.

CONCLUSION

The present study demonstrated that the proposed CNN model can accurately classify mango leaf images into gall midge, powdery mildew, sooty mould and healthy categories with an overall accuracy of 95.98%. The findings highlight the potential of deploying CNN-based tools as practical decision-support systems in precision agriculture. From these results, it is recommended that farmers and extension workers adopt mobile or drone-based image analysis systems powered by CNN models for early detection and management of mango pests and diseases. Such tools can help reduce crop loss, optimize pesticide use and improve yield. Policymakers and agricultural agencies should support the integration of AI-driven plant health monitoring systems into extension services. Future research should expand disease coverage and test the model under varied field conditions to improve generalizability and adoption. This study was limited by the size of the dataset and the number of disease classes. Performance may vary when applied to larger and more diverse field conditions. Future work will include additional diseases, multi-location data and cross-validation techniques. Integrating advanced architectures, ensemble models, or multimodal data such as hyperspectral or drone imagery could further improve accuracy and robustness.

ACKNOWLEDGEMENT

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.
 
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.

CONFLICT OF INTEREST

Authors declare that they have no conflict of interest.

REFERENCES


  1. Adeleye, O.A., Bamiro, O.A., Bakre, L.G., Odeleye, F.O., Adebowale, M.N., Okunye, O.L., Sodeinde, M.A., Adebona, A.C. and Menaa, F. (2021). Medicinal plants with potential inhibitory bioactive compounds against coronaviruses. Advanced Pharmaceutical Bulletin. https://doi.org/10.34172/ apb.2022.003.

  2. Ali, S., Ibrahim, M. Ahmed, S.I., Nadim, M., Mizanur, M.R., Shejunti, M.M. and Jabid, T.  (2022). MangoLeafBD dataset. Mendeley Data. V1, doi: 10.17632/hxsnvwty3r.1.

  3. AlZubi, A.A. (2023). Artificial intelligence and its application in the prediction and diagnosis of animal diseases: A review. Indian Journal of Animal Research. 57(10): 1265-1271. doi: 10.18805/IJAR.BF-1684. 

  4. Charte, D., Charte, F. and Herrera, F. (2021). Reducing data complexity using autoencoders with class-informed loss functions. IEEE Transactions on Pattern Analysis and Machine Intelligence. 44(12): 9549-9560. https://doi.org/10.1109/ TPAMI.2021.3127698.

  5. Cho, O.H. (2024). An evaluation of various machine learning approaches for detecting leaf diseases in agriculture. Legume Research. 47(4): 619-627. doi: 10.18805/LRF- 787.

  6. Food and Agriculture Organisation (FAO). 2023. Global Report on Food Crises: Number of People Facing Acute Food Insecurity Rose to 258 Million in 58 Countries in 2022. https://www.fao.org/newsroom/detail/global-report-on- food-crises-GRFC-2023-GNAFC-fao-wfp-unicef-ifpri/ en#:~:text =The%20 report%20 finds%20that%2 0around, year% 20history%20of%20the%20report. [Accessed on 11/05/2023].

  7. Gautam, V., Ranjan, R.K., Dahiya, P. and Kumar, A. (2023). ESDNN: A novel ensembled stack deep neural network for mango leaf disease classification and detection. Multimedia Tools and Applications. 83(4): 10989-11015. https:// doi.org/10.1007/s11042-023-16012-6.

  8. Gulavnai, S. and Patil, R. (2019). Deep learning for image based mango leaf disease detection. International Journal of Recent Technology and Engineering. 8(3S3): 54-56.

  9. https://www.mango.org/wp-content/uploads/2020/08/Mango_ Pests_and_ Diseases_ENG.

  10. Hussain, S.Z., Naseer, B., Qadri, T., Fatima, T. and Bhat, T.A. (2021). Mango (Mangifera indica)-Morphology, Taxonomy, Composition and Health Benefits. In Fruits Grown in Highland Regions of the Himalayas: Nutritional and Health Benefits. Cham: Springer International Publishing. (pp. 245-255).

  11. Iqbal, N.S., Atiq, N.M., Fayyaz, N.M., Zakria, N.M., Rajput, N.N.A., Aasma, N., Kachelo, N.G. A., Ahmad, N.I., Usman, N.M. and Mehmood, N.A. (2024). Powdery mildew of mango current status, prospective and emerging tools for management. Agricultural Sciences Journal. 6(1): 92- 101. https://doi.org/10.56520/asj.v6i1.365.

  12. Kalaivani, R. and Saravanan, A. (2024). A CONV-EGBDNN model for the classification and detection of mango diseases on diseased mango images utilizing transfer learning. Engineering, Technology and Applied Science Research. 14(3): 14349-14354.

  13. Khan, A.U., Choudhury, M.A.R., Tarapder, S.A., Maukeeb, A.R.M. and Ema, I.J. (2020). Status of mango fruit infestation at home garden in Mymensingh, Bangladesh. Curr. Rese. Agri. Far. 1(4): 35-42. http://dx.doi.org/10.18782/2582- 7146.119. 

  14. Kusuma, K.B.M., Arora, M.,  AlZubi, A.A., Verma, A. and Andrze, S. (2022). Application of blockchain and internet of things (IoT) in the food and beverage industry. Pacific Business Review (International). 15(10): 50-59.

  15. Mia, M.R., Roy, S., Das, S.K. and Rahman, M.A. (2020). Mango leaf disease recognition using neural network and support vector machine. Iran Journal of Computer Science. 3(3): 185-193. https://doi.org/10.1007/s42044-020-00057-z. 

  16. Mustafa, S.K., Oyouni, A.A.W.A., Aljohani, M. and Ahmad, M.A. (2020). Polyphenols more than an antioxidant: Role and scope. Journal of Pure and Applied Microbiology. 14(1): 47-61.

  17. Nalawade, R.R., Sawant, S.D., Joshi, M.S., Ingle, P.M., More, V.G. and Kadam, J.J. (2023). Mango anthracnose and powdery mildew disease detection using convolutional neural network and artificial neural network. Journal of Plant Disease Sciences. 18(1): 11-19. doi: https://doi.org./ 10.48165/jpds.2023.1801.03. 

  18. Noce, A., Romani, A. and Bernini, R. (2021). Dietary intake and chronic disease prevention. Nutrients. 13(4): 1358.

  19. Pratap, V.K. and Kumar, N.S. (2024). Deep learning based mango leaf disease detection for classifying and evaluating mango leaf diseases. Fusion: Practice and Applications. 15(2): 261-61. doi: https://doi.org/10.54216/FPA.150222. 

  20. Rajan, S., Srivastav, M. and Rymbai, H. (2021). Genetic Resources in Mango. In: The Mango Genome. [Kole, C. (Ed.)], Springer. (pp. 63-88). https://doi.org/10.1007/978-3-030- 47829-2_4.

  21. Rajbongshi, A., Khan, T., Pramanik, M.M.R.A., Tanvir, S.M. and Siddiquee, N.R.C. (2021). Recognition of mango leaf disease using convolutional neural network models: A transfer learning approach. Indonesian Journal of Electrical Engineering and Computer Science. 23(3): 1681-1688. doi: 10.11591/ijeecs.v23.i3.pp1681-1688.  

  22. Rao, U.S., Swathi, R., Sanjana, V., Arpitha, L., Chandrasekhar, K. and Naik, P.K. (2021). Deep learning precision farming: Grapes and mango leaf disease detection by transfer learning. Global Transitions Proceedings. 2(2): 535- 544. https://doi.org/10.1016/j.gltp.2021.08.002.

  23. Reddy, P.V.R., Rashmi, M.A., Sreedevi, K. and Singh, S. (2020). Sucking pests of mango. Sucking Pests of Crops. pp. 411-424.

  24. Rizvee, R.A., Orpa, T.H., Ahnaf, A., Kabir, M.A., Rashid, M.R.A., Islam, M.M., Islam, M., Jabid, T. and Ali, M.S. (2023). LeafNet: A proficient convolutional neural network for detecting seven prominent mango leaf diseases. Journal of Agriculture and Food Research. 14: 100787. https://doi.org/10.1016/ j.jafr.2023.100787.

  25. Sammut, C. and Webb, G.I. (Eds.). (2011). Encyclopedia of Machine Learning. Springer Science and Business Media. (pp. 292-293).

  26. Sharma, A., Bijral, R.K., Manhas, J. and Sharma, V. (2022). Mango leaf diseases detection using deep learning. International Journal of Knowledge Based Computer Systems. 10(1): 40-44.

  27. Shickel, B. and Rashidi, P. (2020). Sequential interpretability: Methods, applications and future direction for understanding deep learning models in the context of sequential data. arXiv preprint arXiv. 2004.12524.

  28. Singh, Y., Sinha, S., Malvi, S., Karte, S. and Mimrot, M.K. (2022). Importance disease and pest of mango crop and their management. Advances in Horticulture Crops and Their Challenges. pp 103-128.

  29. Vijay, C.P. and Pushpalatha, K. (2023). Revolutionizing mango leaf disease detection: Leveraging segmentation and hybrid deep learning for enhanced accuracy and sustainability. International Journal of Intelligent Systems and Applications in Engineering. 11(4): 121-131.

  30. Wasik, S. and Pattinson, R.  (2024). Artificial intelligence applications in fish classification and taxonomy: Advancing our understanding of aquatic biodiversity. Fish Taxa. 31: 11-21. 

  31. Yang, S. and Berdine, G. (2017). The receiver operating characteristic (ROC) curve. The Southwest Respiratory and Critical Care Chronicles. 5(19): 34-36.

  32. Zock, M., Chersoni, E., Hsu, Y.Y. and De Deyne, S. (2024). The Workshop on Cognitive Aspects of the Lexicon (CogALex@  LREC-COLING 2024). In the 8th Workshop on Cognitive Aspects of the Lexicon. European Language Resources Association.
Published In
Agricultural Science Digest
Article Metrics
Metrics Icon
75
Views
0
Citations
Reviewed By
Share Article
Download Article
In this Article
    Contact us
    Follow us

    Full Research Article

    Deep Learning Framework for the Automated Identification and Classification of Mango Disease, Pest and Healthy Conditions

    P
    Prashant Chaudhary1,*
    https://orcid.org/0000-0001-5221-6274
    T
    Taniya Sengupta Rathore2
    https://orcid.org/0000-0002-9987-1871
    D
    Durgaprasad Navulla3
    https://orcid.org/0000-0001-5942-6864
    B
    Bharati Chandrayan4
    https://orcid.org/0000-0002-5691-136X
    K
    Kunal D. Jadhav5
    https://orcid.org/0000-0001-8331-5363
    A
    Alok S. Shah6
    https://orcid.org/0000-0001-5708-3546
    1Department of Information Technology, Usha Pravin Gandhi College of Arts, Science and Commerce, SVKM, Mumbai-400 056, Maharashtra, India.
    2Department of Biotechnology, School of Science, O.P. Jindal University, Raigarh-49 6109, Chhattisgarh, India.
    3KL Business School, Koneru Lakshmaiah Education Foundation, Guntur-521 180, Andhra Pradesh, India.
    4Department of Language, Culture and Society, SRM Institute of Science and Technology Delhi NCR Campus, Ghaziabad-201 204, Uttar Pradesh, India.
    5Department of Lifelong Learning and Extension, University of Mumbai, Mumbai-400 020, Maharashtra, India.
    6Bharati Vidyapeeth (Deemed to be) University, Department of Management Studies, Kharghar, Navi Mumbai-410 210, Maharashtra, India.

    ABSTRACT

    Background: Mango (Mangifera indica) is a commercially important fruit crop, but its yield and quality are often threatened by diseases and insect pests. Among the most common are fungal diseases such as sooty mould and powdery mildew and insect pests like gall midge. If not detected early, these cause substantial economic losses to farmers. Manual identification is time-consuming and error-prone, creating the need for automated solutions. Deep learning, particularly Convolutional Neural Networks (CNNs), has shown strong potential in plant disease and pest recognition.
    Methods: In this study, a sequential CNN-based framework was developed for the automated identification and classification of mango leaf images into four classes: Sooty mould, powdery mildew, gall midge and healthy. The architecture consisted of five convolutional layers with max-pooling for feature extraction, followed by fully connected layers for classification. Model performance was assessed using accuracy, precision, recall, F1-score and confusion matrix analysis.
    Result: The model achieved an overall accuracy of 96.0%, with high weighted average precision, recall and F1-scores, indicating reliable performance despite class imbalance. The confusion matrix confirmed the model’s capability to distinguish between disease, pest and healthy conditions with minimal misclassification.

    INTRODUCTION

    Fruits and grains are essential sources of nutrients that play a vital role in human health. Among these, fruits are particularly important because they supply dietary fibre, carotenoids, phytosterols, folic acid, vitamins, minerals and antioxidants. A growing body of evidence highlights the benefits of medicinal plants and bioactive phytocompounds in preventing and managing diseases (Adeleye et al., 2021). Polyphenols, for example, are known to scavenge free radicals and exert anti-carcinogenic effects, contributing significantly to human well-being (Mustafa et al., 2020). Regular consumption of fruits lowers the risk of cancer and helps prevent obesity, cardiovascular disease and chronic respiratory conditions (Noce et al., 2021). Thus, fruits are considered indispensable for a balanced diet and long-term health.
           
    Despite their importance, access to nutrient-rich fruits remains limited in many regions due to global challenges. Population growth, climate change and resource scarcity threaten food security, particularly in developing and underdeveloped nations. The food and agriculture organization reported that approximately 258 million people worldwide suffered from severe food shortages, the highest level in seven years (FAO, 2023). The rising population, projected to increase by 22% in the next five years, will continue to stress agricultural systems, affecting political stability, economic growth and environmental sustainability. This scenario underlines the need for crop protection, yield improvement and sustainable farming strategies.
           
    Among tropical fruits, mango (Mangifera indica) holds special significance. Known as the “king of fruits,” mango is cherished for its flavour, aroma and nutritional value (Hussain et al., 2021). Native to India and Southeast Asia, mango has been cultivated for over 4,000 years and is now grown across Asia, Africa, Central America, Australia and parts of Europe. Hundreds of varieties exist globally, although only a small number are commercially produced on a large scale (Rajan and Srivastav, 2021). Mango contributes substantially to both local economies and global trade. However, its productivity and quality are often hampered by pests and diseases that attack leaves, flowers and fruits, leading to substantial economic losses.
           
    Several diseases and pests pose serious challenges to mango cultivation. Among the most damaging are gall midge, powdery mildew and sooty mould. Gall midges are small flies that infest mango leaves, flowers and fruits. More than 16 species are reported to attack mango worldwide, particularly in Asia. Common species include Erosomya indica, Erosomya mangiferae, Asynapta mangiferae, Gephyraulus mangiferae and Procontarinia matteiana (https://www.mango.org/wp-content/uploads/2020/08/Mango_Pests_and_ Diseases_ENG.). The larvae feed on plant tissues, producing black spots on leaves and affecting growth. These pests significantly reduce yield, either by lowering fruit quality or quantity (Reddy et al., 2020; Khan et al., 2020). Sooty mould is another widespread problem caused by fungi such as Meliola mangiferae, Capnodium mangiferae, Capnodium ramosum and Tripospermum acerinum. The disease is characterised by the presence of a black, velvety layer on leaves and twigs, resulting from fungal spores adhering to honeydew secretions produced by insects. In severe cases, entire leaf surfaces become blackened, reducing photosynthesis and leading to curling and shrivelling of leaves (Singh et al., 2022). Powdery mildew, caused by Pseudoidium anacardii, is considered one of the most economically significant fungal diseases of mango worldwide. Its symptoms appear initially on young leaves and inflorescences as whitish, powdery growth. If untreated, the infection spreads rapidly, leading to flower and fruit drop (Iqbal et al., 2024).
           
    Recent advances in computer vision (CV), machine learning (ML) and deep learning (DL) have opened new opportunities for smart agriculture (Sharma et al., 2022). Machine learning has proven especially valuable in disease detection due to its strong pattern recognition capabilities. By analysing leaf colour, texture and shape, ML algorithms can distinguish between healthy and diseased leaves with high accuracy (Mia et al., 2020). Applications of ML extend beyond agriculture to areas such as biometrics, face recognition, food classification, animal disease detection and medical diagnostics (AlZubi, 2023; Wasik and Pattinson, 2024; Kusuma et al., 2022). In agriculture, ML-based systems can process large datasets, extract hidden features and adapt to complex variations in disease symptoms, making them suitable for real-time applications.
           
    Several studies have demonstrated the potential of ML in mango disease detection. Mia et al., (2020) developed a neural network ensemble combined with a support vector machine to classify mango leaf diseases. They used gray-level Co-occurrence matrix (GLCM) features to capture leaf texture, achieving reliable recognition accuracy. Gulavnai and Patil (2019) employed a convolutional neural network (CNN) with a retrained ResNet architecture to detect mango leaf diseases. CNNs are particularly effective because they automatically learn features from raw image data, eliminating the need for manual feature extraction.
           
    Deep learning models such as CNNs, ResNet and VGG have gained popularity in recent years for agricultural applications. Their hierarchical structure allows them to capture both low-level and high-level features, improving classification performance. CNNs are also computationally efficient, making them suitable for mobile applications and resource-limited settings. Compared to more complex models, CNNs strike a balance between accuracy and implementation simplicity (Shickel and Rashidi, 2020; Zock et al., 2024).
           
    Early detection and effective management of mango diseases are therefore critical. Traditional methods rely on farmers’ experience and laboratory-based testing. While reliable, these methods are time-consuming, costly and impractical for large-scale monitoring. Moreover, accurate identification often requires expert knowledge that may not be accessible to farmers in rural or resource-limited settings. In addition, visual symptoms of diseases can be misleading due to overlapping signs, environmental stress, or nutrient deficiencies. These limitations highlight the urgent need for automated, rapid and accurate approaches for disease identification.
           
    The present study contributes to this effort by exploring deep learning-based frameworks for the automated identification and classification of mango diseases and pests, including gall midge, powdery mildew and sooty mould. By leveraging CNN architectures, this study aims to demonstrate the effectiveness of AI-driven tools for real-time disease diagnosis in mango cultivation. The findings are expected to support precision agriculture practices, enhance yield quality and provide a scalable solution adaptable to other crops and regions.

    MATERIALS AND METHODS

    Data description
     
    The present study employed the publicly available MangoLeafBD dataset hosted on the Mendeley repository. The dataset contains a total of 2,494 mango leaf images, categorized into four distinct classes: Gall midge (Pest-infected), healthy, powdery mildew (Fungal infection) and sooty mould (Fungal infection) (Fig 1). The images were collected using a standard smartphone camera under natural lighting conditions in the orchards of Sher-e-Bangla Agricultural University and Jahangirnagar University, Bangladesh (Ali et al., 2022).

    Fig 1: Images from the dataset.



    Pre-processing of raw data
     
    Using image data to create CV and ML models presents a number of challenges because of issues with sufficiency, accuracy and complexity (Charte et al., 2021). For this reason, data preprocessing, that is, cleaning and reshaping the photos into an appropriate format, is essential before creating any machine learning models in order to get the intended results (Cho, 2024).
           
    Raw image data required systematic pre-processing to ensure uniformity and improve model performance. All images were resized to 256 x 256 pixels to standardize input dimensions. Data augmentation was applied using the following values: rotation (20°), width and height shifts (0.2), shear (0.2), zoom (0.2) and horizontal flipping. These techniques increased dataset variability and reduced the risk of overfitting. Pixel values were normalized to a [0,1] range to lower computational complexity and support faster convergence. This ensured that the CNN extracted meaningful features without being biased by differences in size, orientation, or illumination.
     
    Sequential-CNN architecture
     
    A flowchart of CNN is presented in Fig 2. It classifies mango leaf images into four categories (Gall midge, healthy, powdery mildew and sooty mould). The model accepted pre-processed 256 x 256 RGB images as input.

    Fig 2: Flow diagram of the machine learning process.


           
    The architecture of the CNN model had a sequential design and is presented in Fig 3. It comprised multiple layers, each with a specific role in feature extraction and classification. The input layer received resized 256 x 256 images with three color channels (RGB), ensuring uniform dimensions and pixel scaling. The first convolutional layer applied 32 filters of size 3 x 3 to detect simple, low-level features such as edges, corners and textures, using the ReLU activation function to introduce non-linearity. This was followed by max-pooling layer 1 with a 2 x 2 pooling operation, which reduced feature map size while retaining essential information and lowering computational cost. The second convolutional layer increased the number of filters to 64 (3 x 3) to detect more complex structures, including leaf venation and disease spots, again using ReLU activation. Max-pooling layer 2 further reduced feature map dimensions while preserving critical spatial details. The third convolutional layer, with 128 filters (3 x 3), captured finer disease-specific patterns such as mildew patches and sooty mold textures, followed by max-pooling layer 3 to maintain efficiency and avoid overfitting. The fourth convolutional layer used 128 filters (3 x 3) to refine feature extraction and focus on abstract, high-level visual patterns, with max-pooling layer 4 compressing the extracted features. The fifth and final convolutional layer applied 256 filters (3 x 3) to capture complex and hierarchical representations of pest and disease symptoms and max-pooling layer 5 reduced dimensionality before flattening. The flatten layer transformed pooled feature maps into a one-dimensional vector suitable for fully connected layers. The first dense layer consisted of 64 neurons with ReLU activation to integrate extracted features into meaningful representations for classification. A dropout layer with a rate of 0.5 was applied to randomly deactivate neurons during training, preventing overfitting and improving generalization. The final dense output layer contained 5 neurons representing the four disease categories plus the healthy class, using softmax activation to output class probabilities. This architecture was designed to progressively capture hierarchical features, from basic edges and textures in early layers to complex disease- or pest-specific visual cues in deeper layers, ultimately supporting accurate and automated classification of mango leaf images.

    Fig 3: Visualization of the convolutional model proposed in the present study.


     
    Training model
     
    The CNN model consisted of 2,124,996 trainable parameters, including weights and biases, optimized during training. The dataset was split into training (70%), validation (15%) and testing (15%) subsets to ensure unbiased evaluation. The model was trained for 50 epochs with a batch size of 32. Since the task involved multi-class classification, category cross-entropy loss was used as the objective function. An Adam optimizer with a learning rate of 0.001 was employed to accelerate convergence and avoid local minima. To prevent overfitting, a dropout layer (rate = 0.5) was incorporated between dense layers, which randomly deactivates neurons during training. Training and validation losses were monitored after each epoch to ensure model stability.
     
    Metrics to evaluate the performance of the model
     
    Four metric variables evaluate the performance of any newly developed model (Sammut and Webb, 2011). They are accuracy, F1 score, Precision and Recall. In addition, a confusion matrix is used.
     
    Accuracy
     
    Total number of true predictions (i.e. True positive (TP)+ True Negative (TN)) out of all the instances.
     
     
     
    Precision
     
    Number of true positives of all the true predictions.
     
     
     
    Recall
     
    Number of true predictions of all the true instances.
     
     
     
    F1-score
     
    Combines recall and precision. F1 shows how effectively the models make the trade-off between precision and recall.
     
     
     
    A confusion matrix shows how accurate a model is in classifying data. An example of a confusion matrix with the three classes A, B and C is displayed in Table 1.

    Table 1: Confusion matrix: Example.

    RESULTS AND DISCUSSION

    The developed CNN model effectively classified images into different disease categories. At final epoch, the model achieved 97.6% validation accuracy and 98.1% training accuracy, indicating strong generalization performance. The training loss was 0.053, while the validation loss was 0.062, showing close agreement and minimal overfitting (Fig 4).

    Fig 4: a) Training accuracy and loss, b) Validation accuracy and loss.


           
    The actual and predicted classes, along with confidence scores for each category, are presented in Fig 5. The sample predictions show that the model performed with very high accuracy across different classes of mango leaf conditions. In the first case, a leaf affected by sooty mould was correctly identified with a confidence of 97.18%. The second sample, a healthy leaf, was classified as healthy with a confidence of 99.98%, showing the model’s ability to distinguish normal leaves from diseased ones. The third example involved a leaf affected by gall midge, which was perfectly classified with 100% confidence. In the fourth case, a leaf with powdery mildew was also correctly detected, with a confidence score of 99.85%. These results highlight that the model consistently recognized both healthy and diseased leaves with very high certainty. The high prediction confidence across all examples indicates the robustness of the model and its suitability for practical disease identification.

    Fig 5: Classification of leaf images by the proposed model.


           
    The confusion matrix provides a clear evaluation of the model’s predictions (Fig 6). For gall midge, 54 samples were correctly identified, while 2 were misclassified as sooty mould. No cases were wrongly placed in the healthy or powdery mildew categories. For healthy leaves, 49 were correctly classified, but 2 were wrongly predicted as gall midge and 1 as powdery mildew. For powdery mildew, 54 were classified correctly, with 4 misclassified as sooty mould. No samples were confused with gall midge or healthy. For sooty mould, all 58 samples were classified correctly with no errors in other categories.

    Fig 6: Confusion Matrix of diseased and healthy mango leaves.


           
    This analysis shows that most predictions matched the actual labels, with very few misclassifications. The errors mainly occurred between classes with somewhat similar visual symptoms, such as powdery mildew and sooty mould. Despite these minor errors, the model demonstrated high reliability in classifying all four categories.
           
    The classification report gives a detailed view of the model’s performance (Table 2). For the gall midge class, the model achieved a precision of 0.9643 and a recall of 0.9643, with an F1-score of 0.9643. This indicates balanced and consistent performance. For healthy leaves, precision reached 1.000, showing that all predicted healthy samples were correct. Recall was 0.9423, meaning a few healthy samples were misclassified. The F1-score of 0.9703 reflects this slight drop in recall. For powdery mildew, the model showed strong precision at 0.9818, but recall was 0.931. The F1-score of 0.9558 reflects good overall balance. For sooty mould, the model produced a precision of 0.9062 and a recall of 1.000, which means all sooty mould samples were correctly identified, though a few predictions for this class included errors. The F1-score here was 0.9508.

    Table 2: Classification matrices after testing the dataset.


           
    The overall accuracy of the model across all classes was 95.98%. The macro average values were 0.9631 for precision, 0.9594 for recall and 0.9603 for F1-score. The weighted averages, which account for class imbalance, were close in value, with precision at 0.9621, recall at 0.9598 and F1-score at 0.9600. These results confirm that the model performs consistently well across all categories, with high reliability in both precision and recall.
           
    Fig 7 displays the ROC curves together with the corresponding Area under the curve (AUC). A true positive/false positive ratio for the discrimination cut-off value of the used test is shown by each point on the ROC curve. In essence, it shows how the TPR and FPR trade off at different categorization criteria. The model’s overall performance is measured by overall potential classification thresholds using the Area Under the ROC curve (AUC-ROC) metric (Yang and Berdine, 2017). Regardless of class imbalance, the superior discriminating ability is shown by a higher AUC-ROC value. A high AUC-ROC indicates that the model can successfully discriminate between the minority class and the majority classes in the situation of unbalanced datasets. The Powdery mildew and the Gall midge in the current research had the greatest AUC (0.76 and 0.75), followed by Sooty mould (0.67) and Healthy class (0.62).

    Fig 7: ROC(AUC) curve for all diseased classes.


           
    The accuracy of the proposed model was found to be competitive when compared with existing studies (Table 3). Rajbongshi et al., (2021) reported 98.0% accuracy using DenseNet201, while Rizvee et al., (2023) achieved 98.55% with LeafNet. Gautam et al., (2023) obtained 98.57% using an ensemble approach. These values are slightly higher than the 96.0% achieved in this study. However, those works relied on larger datasets and included more disease classes. Given the smaller dataset used here, the performance of the present CNN remains notable.

    Table 3: Comparative analysis of classification accuracy of the proposed CNN model with existing studies.


           
    Several studies have reported results closer to or below the accuracy of this model. Vijay and Pushpalatha (2023) achieved 93.01% with a hybrid EfficientNet model. Pratap and Kumar (2024) reported 94.5% using YOLOv8 with African Buffalo Optimization. Rao et al., (2021) obtained 89% for mango leaf classification using AlexNet in their JIT CROPFIX application. In these comparisons, the current CNN clearly outperformed existing approaches. Kalaivani and Saravanan (2024) achieved a slightly higher accuracy of 98.25% using a segmentation-based ResNet50 framework. However, that method required additional preprocessing, whereas the present model achieved robust results with a simpler workflow.
           
    Nalawade et al., (2023) reported 96-97% accuracy by combining thermal and RGB images for anthracnose and powdery mildew classification. Their approach, however, depended on thermal imaging equipment, which limits field-level adoption. In contrast, the current study relied only on RGB images, making it more practical and accessible for large-scale use.
           
    In summary, while some ensemble and advanced architectures achieved marginally higher accuracies, the proposed CNN offers competitive results with key advantages. It is simple, requires less computation and incorporates both pest and disease classes. This broader classification scope strengthens its novelty and enhances its potential for real-world precision agriculture.

    CONCLUSION

    The present study demonstrated that the proposed CNN model can accurately classify mango leaf images into gall midge, powdery mildew, sooty mould and healthy categories with an overall accuracy of 95.98%. The findings highlight the potential of deploying CNN-based tools as practical decision-support systems in precision agriculture. From these results, it is recommended that farmers and extension workers adopt mobile or drone-based image analysis systems powered by CNN models for early detection and management of mango pests and diseases. Such tools can help reduce crop loss, optimize pesticide use and improve yield. Policymakers and agricultural agencies should support the integration of AI-driven plant health monitoring systems into extension services. Future research should expand disease coverage and test the model under varied field conditions to improve generalizability and adoption. This study was limited by the size of the dataset and the number of disease classes. Performance may vary when applied to larger and more diverse field conditions. Future work will include additional diseases, multi-location data and cross-validation techniques. Integrating advanced architectures, ensemble models, or multimodal data such as hyperspectral or drone imagery could further improve accuracy and robustness.

    ACKNOWLEDGEMENT

    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.
     
    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.

    CONFLICT OF INTEREST

    Authors declare that they have no conflict of interest.

    REFERENCES


    1. Adeleye, O.A., Bamiro, O.A., Bakre, L.G., Odeleye, F.O., Adebowale, M.N., Okunye, O.L., Sodeinde, M.A., Adebona, A.C. and Menaa, F. (2021). Medicinal plants with potential inhibitory bioactive compounds against coronaviruses. Advanced Pharmaceutical Bulletin. https://doi.org/10.34172/ apb.2022.003.

    2. Ali, S., Ibrahim, M. Ahmed, S.I., Nadim, M., Mizanur, M.R., Shejunti, M.M. and Jabid, T.  (2022). MangoLeafBD dataset. Mendeley Data. V1, doi: 10.17632/hxsnvwty3r.1.

    3. AlZubi, A.A. (2023). Artificial intelligence and its application in the prediction and diagnosis of animal diseases: A review. Indian Journal of Animal Research. 57(10): 1265-1271. doi: 10.18805/IJAR.BF-1684. 

    4. Charte, D., Charte, F. and Herrera, F. (2021). Reducing data complexity using autoencoders with class-informed loss functions. IEEE Transactions on Pattern Analysis and Machine Intelligence. 44(12): 9549-9560. https://doi.org/10.1109/ TPAMI.2021.3127698.

    5. Cho, O.H. (2024). An evaluation of various machine learning approaches for detecting leaf diseases in agriculture. Legume Research. 47(4): 619-627. doi: 10.18805/LRF- 787.

    6. Food and Agriculture Organisation (FAO). 2023. Global Report on Food Crises: Number of People Facing Acute Food Insecurity Rose to 258 Million in 58 Countries in 2022. https://www.fao.org/newsroom/detail/global-report-on- food-crises-GRFC-2023-GNAFC-fao-wfp-unicef-ifpri/ en#:~:text =The%20 report%20 finds%20that%2 0around, year% 20history%20of%20the%20report. [Accessed on 11/05/2023].

    7. Gautam, V., Ranjan, R.K., Dahiya, P. and Kumar, A. (2023). ESDNN: A novel ensembled stack deep neural network for mango leaf disease classification and detection. Multimedia Tools and Applications. 83(4): 10989-11015. https:// doi.org/10.1007/s11042-023-16012-6.

    8. Gulavnai, S. and Patil, R. (2019). Deep learning for image based mango leaf disease detection. International Journal of Recent Technology and Engineering. 8(3S3): 54-56.

    9. https://www.mango.org/wp-content/uploads/2020/08/Mango_ Pests_and_ Diseases_ENG.

    10. Hussain, S.Z., Naseer, B., Qadri, T., Fatima, T. and Bhat, T.A. (2021). Mango (Mangifera indica)-Morphology, Taxonomy, Composition and Health Benefits. In Fruits Grown in Highland Regions of the Himalayas: Nutritional and Health Benefits. Cham: Springer International Publishing. (pp. 245-255).

    11. Iqbal, N.S., Atiq, N.M., Fayyaz, N.M., Zakria, N.M., Rajput, N.N.A., Aasma, N., Kachelo, N.G. A., Ahmad, N.I., Usman, N.M. and Mehmood, N.A. (2024). Powdery mildew of mango current status, prospective and emerging tools for management. Agricultural Sciences Journal. 6(1): 92- 101. https://doi.org/10.56520/asj.v6i1.365.

    12. Kalaivani, R. and Saravanan, A. (2024). A CONV-EGBDNN model for the classification and detection of mango diseases on diseased mango images utilizing transfer learning. Engineering, Technology and Applied Science Research. 14(3): 14349-14354.

    13. Khan, A.U., Choudhury, M.A.R., Tarapder, S.A., Maukeeb, A.R.M. and Ema, I.J. (2020). Status of mango fruit infestation at home garden in Mymensingh, Bangladesh. Curr. Rese. Agri. Far. 1(4): 35-42. http://dx.doi.org/10.18782/2582- 7146.119. 

    14. Kusuma, K.B.M., Arora, M.,  AlZubi, A.A., Verma, A. and Andrze, S. (2022). Application of blockchain and internet of things (IoT) in the food and beverage industry. Pacific Business Review (International). 15(10): 50-59.

    15. Mia, M.R., Roy, S., Das, S.K. and Rahman, M.A. (2020). Mango leaf disease recognition using neural network and support vector machine. Iran Journal of Computer Science. 3(3): 185-193. https://doi.org/10.1007/s42044-020-00057-z. 

    16. Mustafa, S.K., Oyouni, A.A.W.A., Aljohani, M. and Ahmad, M.A. (2020). Polyphenols more than an antioxidant: Role and scope. Journal of Pure and Applied Microbiology. 14(1): 47-61.

    17. Nalawade, R.R., Sawant, S.D., Joshi, M.S., Ingle, P.M., More, V.G. and Kadam, J.J. (2023). Mango anthracnose and powdery mildew disease detection using convolutional neural network and artificial neural network. Journal of Plant Disease Sciences. 18(1): 11-19. doi: https://doi.org./ 10.48165/jpds.2023.1801.03. 

    18. Noce, A., Romani, A. and Bernini, R. (2021). Dietary intake and chronic disease prevention. Nutrients. 13(4): 1358.

    19. Pratap, V.K. and Kumar, N.S. (2024). Deep learning based mango leaf disease detection for classifying and evaluating mango leaf diseases. Fusion: Practice and Applications. 15(2): 261-61. doi: https://doi.org/10.54216/FPA.150222. 

    20. Rajan, S., Srivastav, M. and Rymbai, H. (2021). Genetic Resources in Mango. In: The Mango Genome. [Kole, C. (Ed.)], Springer. (pp. 63-88). https://doi.org/10.1007/978-3-030- 47829-2_4.

    21. Rajbongshi, A., Khan, T., Pramanik, M.M.R.A., Tanvir, S.M. and Siddiquee, N.R.C. (2021). Recognition of mango leaf disease using convolutional neural network models: A transfer learning approach. Indonesian Journal of Electrical Engineering and Computer Science. 23(3): 1681-1688. doi: 10.11591/ijeecs.v23.i3.pp1681-1688.  

    22. Rao, U.S., Swathi, R., Sanjana, V., Arpitha, L., Chandrasekhar, K. and Naik, P.K. (2021). Deep learning precision farming: Grapes and mango leaf disease detection by transfer learning. Global Transitions Proceedings. 2(2): 535- 544. https://doi.org/10.1016/j.gltp.2021.08.002.

    23. Reddy, P.V.R., Rashmi, M.A., Sreedevi, K. and Singh, S. (2020). Sucking pests of mango. Sucking Pests of Crops. pp. 411-424.

    24. Rizvee, R.A., Orpa, T.H., Ahnaf, A., Kabir, M.A., Rashid, M.R.A., Islam, M.M., Islam, M., Jabid, T. and Ali, M.S. (2023). LeafNet: A proficient convolutional neural network for detecting seven prominent mango leaf diseases. Journal of Agriculture and Food Research. 14: 100787. https://doi.org/10.1016/ j.jafr.2023.100787.

    25. Sammut, C. and Webb, G.I. (Eds.). (2011). Encyclopedia of Machine Learning. Springer Science and Business Media. (pp. 292-293).

    26. Sharma, A., Bijral, R.K., Manhas, J. and Sharma, V. (2022). Mango leaf diseases detection using deep learning. International Journal of Knowledge Based Computer Systems. 10(1): 40-44.

    27. Shickel, B. and Rashidi, P. (2020). Sequential interpretability: Methods, applications and future direction for understanding deep learning models in the context of sequential data. arXiv preprint arXiv. 2004.12524.

    28. Singh, Y., Sinha, S., Malvi, S., Karte, S. and Mimrot, M.K. (2022). Importance disease and pest of mango crop and their management. Advances in Horticulture Crops and Their Challenges. pp 103-128.

    29. Vijay, C.P. and Pushpalatha, K. (2023). Revolutionizing mango leaf disease detection: Leveraging segmentation and hybrid deep learning for enhanced accuracy and sustainability. International Journal of Intelligent Systems and Applications in Engineering. 11(4): 121-131.

    30. Wasik, S. and Pattinson, R.  (2024). Artificial intelligence applications in fish classification and taxonomy: Advancing our understanding of aquatic biodiversity. Fish Taxa. 31: 11-21. 

    31. Yang, S. and Berdine, G. (2017). The receiver operating characteristic (ROC) curve. The Southwest Respiratory and Critical Care Chronicles. 5(19): 34-36.

    32. Zock, M., Chersoni, E., Hsu, Y.Y. and De Deyne, S. (2024). The Workshop on Cognitive Aspects of the Lexicon (CogALex@  LREC-COLING 2024). In the 8th Workshop on Cognitive Aspects of the Lexicon. European Language Resources Association.
    In this Article
      Contact us
      Follow us
      Published In
      Agricultural Science Digest

      Editorial Board

      View all (0)