INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
Real-Time Image-Based Recognition of Mango Leaf Diseases Using  
Convolutional Neural Networks  
Priyanka Gonnade, Saloni Zade, Sonal Shende, Tanishka Mahajan, Tanvi Jain, Tejas Agarkar  
Department of Computer Science & Engineering, G H Raisoni College of Engineering & Management,  
Nagpur, Maharashtra, India  
DOI : https://doi.org/10.51583/IJLTEMAS.2025.1412000028  
Received: 17 November 2025; Accepted: 20 December 2025; Published: 31 December 2025  
ABSTRACT:  
Mango is a vital tropical fruit crop, yet its productivity is often reduced by leaf diseases such as Powdery Mildew,  
Dieback, Anthracnose, Bacterial Canker, and Sooty Mold. These infections lower yield, degrade fruit quality,  
and cause major economic losses. Early detection is crucial but challenging for farmers with limited expert  
access.  
This study proposes an image-based classification system using Convolutional Neural Networks (CNN) for  
accurate disease recognition. A curated dataset of mango leaf images was pre-processed and augmented to  
address class imbalance. The CNN model outperformed traditional classifiers like Support Vector Machine  
(SVM) and Decision Tree in terms of accuracy, robustness, and efficiency.  
The system not only detects multiple diseases with high precision but also offers severity estimation, visual  
feedback, and farmer-friendly  
treatment recommendations. Designed for real-time use via smartphones or  
field cameras, it provides a scalable and accessible solution to support precision agriculture.  
Keywords: Mango leaf disease, CNN, Image classification, Deep learning, Precision agriculture, Real-time  
detection  
INTRODUCTION:  
Mango (Mangifera indica) is one of the most commercially valuable fruit crops in India and across the tropics.  
Despite its economic and nutritional importance, mango cultivation faces numerous challenges, with leaf  
diseases being one of the most severe threats [1,5]. Diseases such as Powdery Mildew, Dieback, Anthracnose,  
Bacterial Canker, and Sooty Mold adversely affect leaf health, reduce photosynthetic efficiency, and  
consequently lower fruit yield and quality, leading to substantial economic losses for farmers [1,4].  
Traditionally, disease detection in mango trees relies on manual inspection by farmers or agricultural experts.  
However, this approach is labour-intensive, time- consuming, and prone to errors, especially in large orchards  
or in areas where expert knowledge is scarce [5]. Moreover, delayed or inaccurate diagnosis can result in  
inappropriate treatment measures, which may worsen the infection and increase the cost of cultivation [4,5].  
Recent advances in computer vision and machine learning provide a promising alternative through automated  
disease detection systems [2,3,6]. These systems leverage image-based analysis to identify disease symptoms  
accurately and rapidly. Convolutional Neural Networks (CNN) [2,4,5], a type of deep learning model, have  
shown exceptional performance in image recognition tasks due to their ability to automatically learn hierarchical  
features from raw images. Classical machine learning algorithms, such as Support Vector Machine (SVM) and  
Decision Tree [5,6], also offer disease classification capabilities, particularly when combined with robust feature  
extraction techniques.  
This study aims to develop an automated, efficient, and user-friendly system for classifying five major mango  
leaf diseases using CNN [2,4]. Comparative analysis with SVM and Decision Tree classifiers is also performed  
www.ijltemas.in  
Page 317  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
to evaluate the relative performance and robustness of different approaches. The proposed system is designed  
for real-time application in the field, enabling farmers to capture leaf images via smartphones or cameras and  
receive instant, reliable diagnoses [2,3]. By providing timely detection and actionable insights, the system has  
the potential to support effective disease management, reduce crop losses, and enhance mango production  
efficiency [1,5].  
LITERATURE REVIEW:  
Mahmud et al. [1] developed a lightweight deep learning model capable of fast inference for mango leaf disease  
classification, emphasizing reduced computational requirements for mobile devices.  
Pathak and Kumar [2] proposed a robust CNN architecture specifically for mango leaf diseases, demonstrating  
high accuracy across multiple disease categories.  
Gautam and Kumar [3] introduced an ensembled stacked deep neural network that enhanced feature learning  
and improved classification performance for complex disease patterns.  
Kumar et al. [4] evaluated multiple CNN architectures and highlighted their strong capability for identifying  
diverse plant leaf diseases.  
Patil and Thorat [5] further supported CNN effectiveness by demonstrating early disease detection with high  
precision.  
Transfer learning has proven especially useful for limited datasets.  
Zhang et al. [6] showed that pre-trained CNN models significantly boost performance in plant disease  
identification.  
Rani et al. [7] integrated CNN models with IoT-based smart agriculture systems, enabling continuous monitoring  
and timely alerts.  
Majeed et al. [8] demonstrated the effectiveness of deep CNNs for mango leaf disease classification, setting the  
base for modern architectures.  
METHODOLOGIES:  
The proposed system for mango leaf disease classification is designed using a Convolutional Neural Network  
(CNN) as the primary deep learning model, with comparative evaluation against Support Vector Machine (SVM)  
and Decision Tree classifiers. The methodology consists of the following stages:  
Fig: Block Diagram  
www.ijltemas.in  
Page 318  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
Data Preprocessing:  
Images were resized into a fixed resolution (e.g., 224 × 224px) for uniformity.  
Techniques such as rotation, flipping, zooming, and contrast adjustment were applied for data augmentation,  
ensuring robustness to real-world variations.  
Noise reduction and background normalization were used to enhance disease-affected regions.  
Feature Extraction:  
CNN automatically extracts spatial features such as color patterns, texture, and leaf vein structures.  
For comparative study, handcrafted features like Histogram of Oriented Gradients (HOG) and color histograms  
were also tested with classical ML models (SVM, Decision Tree).  
Model Development:  
CNN Architecture: A custom CNN with multiple convolutional, pooling, and fully connected layers was  
designed for classification.  
Comparative Models: SVM and Decision Tree classifiers were trained using the same dataset to benchmark  
performance.  
Training and Validation:  
The dataset was split into training (80%), validation (20).  
Models were trained using cross-entropy loss and Adam optimizer with early stopping to prevent overfitting.  
Performance metrics included accuracy, precision, recall, F1-score, and confusion matrix.  
System Implementation:  
A real-time diagnostic application was developed, allowing farmers to capture leaf images via smartphone and  
receive instant classification results. The app provides:  
Visual feedback by highlighting infected areas, Interactive confidence charts to display prediction confidence,  
Farmer friendly interface supporting local languages.  
Evaluation:  
CNN’s performance was compared with traditional ML models to highlight the advantages of deep learning in  
image- based disease recognition.  
The proposed approach was tested under varying light and field conditions to assess real-world applicability.  
Furthermore, the model’s robustness was verified using data augmentation and cross- validation techniques to  
ensure consistent performance across diverse leaf types and disease stages. The evaluation also focused on key  
metrics such as accuracy, precision, recall, and F1-score, confirming the CNN’s superiority in handling complex  
visual patterns. This comprehensive assessment demonstrates the system’s capability to generalize well  
beyond the training dataset.  
www.ijltemas.in  
Page 319  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
Fig: Comparison Metrics  
Vector Machine (SVM) and Decision Tree models, CNNs achieve higher accuracy, precision, and recall, as they  
automatically extract hierarchical image features rather than relying on manually crafted ones.  
Confidence chart:  
Instead of just saying “infected,” the system provides Confidence chart. This helps farmers judge whether  
immediate action is required or if preventive care is enough.  
Fig: summery metrics  
RESULT AND DISCUSSION:  
Mango productivity in India is often affected by leaf diseases that are hard to identify without expert knowledge.  
Many farmers rely on guesswork, leading to ineffective treatments, higher costs, and crop damage.  
Deep learning provides a solution by teaching computers to recognize leaf diseases from images. Our system  
uses convolutional neural networks (CNNs) to classify major mango leaf diseases, including anthracnose,  
powdery mildew, die-back, bacterial canker, and sooty mould. Farmers can take or upload a leaf image using a  
smartphone, and the system provides an instant diagnosis. Compared to Support  
www.ijltemas.in  
Page 320  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
Fig: Confidence Chart  
Instant Diagnosis in the Field:  
The model is lightweight and works on mobile devices, giving results within seconds. Farmers can use it right  
in the field without waiting for experts or lab results.  
Fig: Recommended Actions  
Overall, the system reduces guesswork, saves time, lowers costs, and helps farmers manage mango leaf  
diseases  
more effectively, improving crop yield and efficiency.  
CONCLUSION:  
The mango leaf disease classification system effectively identifies multiple leaf diseases with high accuracy  
using deep learning models. By analysing leaf images, the system provides fast, reliable, and automated  
detection, reducing the need for manual inspection. Its confidence score and visual feedback help farmers  
understand the disease more clearly. The solution improves early diagnosis, supports timely treatment, and  
enhances overall crop health. With mobile and local-language support, the system becomes highly accessible  
for real-world agricultural use.  
REFERENCES:  
1. Mahmud,  
B. U., Al Mamun, A., Hossen, M. J., Hong, G. Y., & Jahan, B. (2024). Lightweight  
deep learning model for accelerating the classification of mango-leaf disease.  
2. Emerging Science Journal, 8(1), 28–42. https://doi.org/10.28991/ESJ-2024-08-01- 03  
3. Pathak, A. K., & Kumar, S. (2024). Development of a robust CNN model for mango leaf disease  
detection. ACS Agricultural Science & Technology, 4(1), 1–10. https://doi.org/10.1021/acsagscitech.4c0012  
2
4. Gautam, V., & Kumar, R. (2024). A novel ensembled stack deep neural network for mango leaf disease  
classification. Multimedia Tools and Applications, 83(4), 10989–11015. https://doi.org/10.1007/s11042-  
023-16012-6  
5. Kumar, A., Singh, M., & Jindal, N. (2023). Deep learning-based detection and classification of plant leaf  
diseases using CNN architectures. Computers and Electronics in Agriculture, 205, 107596.  
https://doi.org/10.1016/j.compag.2023.1075 96  
6. Patil, S. B., & Thorat, S. A. (2022). Early detection of plant leaf diseases using convolutional neural  
networks. Journal of Plant Pathology, 104(2), 567–576. https://doi.org/10.1007/s42161-022-01047- 8  
7. Zhang, Y., Chen, K., & Li, X. (2023). Transfer  
using pre-trained CNN models. IEEE  
https://doi.org/10.1109/ACCESS.2023.3255 124  
8. Rani, P., Sharma, A., & Singh, R. (2024). Smart agriculture: Real-time crop disease detection using CNN  
learning  
for  
plant  
disease identification  
Access,  
11,  
34521–34532.  
and IoT integration. Sustainable Computing: Informatics and Systems,  
https://doi.org/10.1016/j.suscom.2024.1008 82  
41,  
100882.  
9. Majeed, Y., Zhang, J., Li, J., & Karkee, M. (2020). Mango leaf disease classification using deep  
www.ijltemas.in  
Page 321  
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,  
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)  
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XII, December 2025  
convolutional neural networks. Computers and Electronics in Agriculture, 169, 105161.  
https://doi.org/10.1016/j.compag.2019.1051 61  
10. Rajput, V., & Yadav, V. (2021). Automated detection of mango leaf diseases using transfer learning  
models. International Journal of Advanced Computer Science and Applications (IJACSA),  
https://doi.org/10.14569/IJACSA.2021.012 0657  
12(6).  
11.Ramkumar, N., & Vijayalakshmi, P. (2023). Mango leaf disease detection using EfficientNet-B0 with data  
augmentation. Neural Processing Letters. https://doi.org/10.1007/s11063-022-10933- 1  
12. Arora, R., & Verma, R. (2021). Image-based mango leaf disease identification using deep learning and image  
segmentation. International Journal of Engineering Research & Technology (IJERT), 10(9).  
www.ijltemas.in  
Page 322