Enhanced Deep Learning Model for Automated Plant

DiseaseDr.Identification
C. Saravana Chandrasekaran
Asst. Professor, M.E., Ph.D, SRMIST
Agenda
• Abstract
• Motivation
• Problem Definition
• Existing work
• Proposed Methodology
• Result
• Conclusion
• Reference
Abstract
Plant diseases significantly reduce crop growth and productivity, leading to
economic losses. Early detection is essential to avoid potential damage. Automated
plant disease detection using deep learning, especially Residual Neural Networks
(ResNet), has made it possible to detect diseases efficiently and accurately. ResNet’s
residual connections enable deep architecture and prevent issues like vanishing
gradients, allowing the model to learn complex patterns such as color, texture, and
structure in leaf images.
Trained on a dataset of over 87,000 images across 38 plant categories, the model
achieves an impressive 99.3% accuracy in disease classification. Optimized for edge
computing using TPUs and VPUs, the system is designed for real-time field
applications, helping farmers manage diseases effectively and prevent crop losses.
Motivation

The rise of deep learning in plant disease detection presents a transformative

opportunity for modern agriculture. By harnessing advanced models like ResNet,
farmers can now detect plant diseases with unprecedented accuracy, allowing for
early intervention and minimizing crop loss. This shift empowers farmers, making
their work more efficient and helping to secure food supplies globally.
The system’s ability to perform in real-time through edge computing technologies,
like TPUs and VPUs, ensures its use even in remote or resource-limited settings. As
these innovations continue to evolve, they hold the potential to revolutionize crop
management and boost agricultural productivity.
Problem Definition

• Diverse Diseases and Symptoms: Different plant species present unique and
overlapping disease symptoms, requiring adaptable and sophisticated detection
models.
• Limited and Imbalanced Data: The collection of high-quality labeled images is
challenging, leading to imbalanced datasets that bias models toward common
diseases.
• Image Analysis Complexities: Variations in lighting, background distractions,
and image quality complicate accurate disease detection from images.
• Real-world Deployment Constraints: Deep learning models face challenges in
resource-constrained environments, requiring lightweight solutions for offline
inference and real-time application.
 Existing work
• The existing work focuses on the application of deep learning techniques,
particularly CNNs, MobileNet, EfficientNet, and transfer learning, for detecting
plant and crop diseases. Researchers have leveraged architectures like VGG16 and
ResNet, along with transfer learning, to enhance accuracy on small agriculture-
based datasets. Hybrid models have been developed by integrating different
architectures to handle agricultural image data more effectively.
• Some studies have applied CNNs and MobileNet for pest detection in crops like
citrus fruits, apples, and tomatoes, addressing limited dataset availability by fine-
tuning pre-trained models.
• Additionally, EfficientNet and deep learning approaches have been used for rice
plant disease detection, achieving high accuracy. Ensemble techniques and
transfer learning further improved model performance and adaptability across
various plant species and environments, making these techniques more applicable
in real-world agricultural settings.
 Proposed Methodology
Dataset:
•87,000+ RGB images of healthy and diseased crop leaves.
•Organized into 38 distinct categories of plant diseases.
•80/20 split for training and validation, with 33 images set aside for testing.
•Offline augmentation applied to increase data diversity (rotation, scaling, cropping).
Model Workflow:
•Systematic process: Data collection → preprocessing → model training → classification.
•Key models: Convolutional Neural Network (CNN) and ResNet.
•Workflow: Input data → Feature extraction → Model training → Disease classification.
•Hybrid model approach combines CNN and ResNet predictions for enhanced accuracy.
•Structured machine learning pipeline helps to streamline the process from raw data
to final output.
Figure 1 . Process Work Flow
ResNet Architecture:
•Deep learning model with residual blocks to avoid vanishing gradients.
•Identity skip connections allow smoother gradient flow across layers.
•Includes 3x3 and 1x1 kernels, ReLU activation, and batch normalization.
•Downsampling blocks handle larger images and complex features.
•ResNet excels at capturing intricate patterns in large datasets, improving classification
accuracy

Performance Evaluation:
•Metrics: Accuracy, Precision, Recall, F1-score.
•Cross-validation ensures robustness and reduces overfitting.
•Precision: Measures correctly predicted positive cases.
•Recall: Captures the true positive rate (sensitivity).
•F1-score: Balances precision and recall for a holistic model evaluation.
•Confusion matrix helps visualize classification results across different disease categories.
Figure 2. ResNet Architecture
Result

Figure 3. Accuracy vs. No. of epochs Figure 4. Loss vs. No. of epochs

Accuracy and loss were monitored during the training period, as shown in Figures 3 and
4 respectively, which illustrate the inverse connection between accuracy and epoch count
Figure 5: Accuracy of State-of-the-models Comparison with Pre-trained Models
with Accuracy

ResNet achieves the highest accuracy rates of 95-99%, followed closely by EfficientNet at 94-97%,
making both ideal for high-accuracy applications. In contrast, CNN and MobileNet score between
80-85% and 82-87%, respectively, serving as simpler alternatives for low-resource settings.
 Figure 6: Precision-Recall Curve

It appears that in the ranges where recall is near zero, the models are unstable in precision, and
they do not preserve the same precision level. Even if it becomes very close to zero, the models
become unstable in terms of precision. However, with an increase in recall , the precision levels
remain stable at 0.4-0.6 across the models
Conclusion

• This work presents a deep learning approach using a Residual Neural Network
(ResNet) for automatic plant disease detection. Achieving an impressive 99.3%
accuracy on a dataset of over 87,000 images in 38 categories, ResNet
outperforms other models like CNN, MobileNet, and EfficientNet. Residual
connections address the vanishing gradient issue, enabling efficient learning of
complex features.
• The model is optimized for edge computing on platforms like TPUs and VPUs,
making it suitable for real-time agricultural applications in environments with
limited computational resources. This highlights ResNet's potential in improving
crop disease detection through high precision and efficient scalability.
Reference
• Shamiul Islam, Md, Ummya Habiba, Md Abu Baten, Nazrul Amin, Imrus Salehin, and Tasmia Tahmida Jidney.
“Hybrid Convolution Neural Network with Transfer Learning Approach for Agro-Crop Leaf Disease Identification.”
In Lecture Notes on Data Engineering and Communications Technologies, 209–17. Cham: Springer Nature
Switzerland, 2023. http://dx.doi.org/10.1007/978-3-031-24475-9_18.

• Hadipour-Rokni, Ramazan, Ezzatollah Askari Asli-Ardeh, Ahmad Jahanbakhshi, Iman Esmaili paeen-Afrakoti, and
Sajad Sabzi. “Intelligent Detection of Citrus Fruit Pests Using Machine Vision System and Convolutional Neural
Network through Transfer Learning Technique.” Computers in Biology and Medicine 155 (March 2023): 106611.
https://doi.org/10.1016/j.compbiomed.2023.106611.

• Saleem, Muhammad Hammad, Johan Potgieter, and Khalid Mahmood Arif. “Plant Disease Classification: A
Comparative Evaluation of Convolutional Neural Networks and Deep Learning Optimizers.” Plants 9, no. 10
(October 6, 2020): 1319. https://doi.org/10.3390/plants9101319.
• Chandrasekaran, Saravanan, S. Athinarayanan, M. Masthan, Pallapothu Bhoga Venkata Ganesh, Sirigiri Sai Saran
Ganesh, and Mahesh T R. “Classification of Parkinson’s Disease Based on Acoustic Characteristics Using Feature
Selection.” In 2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems
(ADICS), 1–6. IEEE, 2024. http://dx.doi.org/10.1109/adics58448.2024.10533543.
• A, Hussain., and Balaji Srikaanth. P. “Disease Classification and Detection Techniques in Rice Plant Using Deep
Learning.” In 2022 8th International Conference on Smart Structures and Systems (ICSSS), 1–7. IEEE, 2022.
http://dx.doi.org/10.1109/icsss54381.2022.9782162.
• Yang, Qing, Shukai Duan, and Lidan Wang. “Efficient Identification of Apple Leaf Diseases in the Wild Using Convolutional Neural
Networks.” Agronomy 12, no. 11 (November 9, 2022): 2784. https://doi.org/10.3390/agronomy12112784.

• D, Femi, Murugasami R, Manikandaprabu N, Raja Paulsingh J, and Vanaja P. “Identification and Classification of Tomato Leaf Diseases
Using Machine Learning Techniques.” Webology 18, no. 05 (October 29, 2021): 1168–75. https://doi.org/10.14704/web/v18si05/web18297.
• Mahesh, T. R., Saravanan Chandrasekaran, V. Ashwin Ram, V. Vinoth Kumar, V. Vivek, and Suresh Guluwadi. “Data-Driven Intelligent
Condition Adaptation of Feature Extraction for Bearing Fault Detection Using Deep Responsible Active Learning.” IEEE Access 12 (2024):
45381–97. https://doi.org/10.1109/access.2024.3380438.
• Kibriya, Hareem, Rimsha Rafique, Wakeel Ahmad, and S.M Adnan. “Tomato Leaf Disease Detection Using Convolution Neural Network.”
In 2021 International Bhurban Conference on Applied Sciences and Technologies (IBCAST). IEEE, 2021.
http://dx.doi.org/10.1109/ibcast51254.2021.9393311.
• Sharada, K. A., T. R. Mahesh, Saravanan chandrasekaran, R. Shashikumar, V. Vinoth Kumar, and Jonnakuti Rajkumar Annand. “Improved
Energy Efficiency Using Adaptive Ant Colony Distributed Intelligent Based Clustering in Wireless Sensor Networks.” Scientific Reports 14,
no. 1 (February 22, 2024). https://doi.org/10.1038/s41598-024-55099-1.
• Li, Xin, and Laxmisha Rai. “Apple Leaf Disease Identification and Classification Using ResNet Models.” In 2020 IEEE 3rd International
Conference on Electronic Information and Communication Technology (ICEICT). IEEE, 2020.
http://dx.doi.org/10.1109/iceict51264.2020.9334214.
• Shamiul Islam, Md, Ummya Habiba, Md Abu Baten, Nazrul Amin, Imrus Salehin, and Tasmia Tahmida Jidney. “Hybrid Convolution Neural
Network with Transfer Learning Approach for Agro-Crop Leaf Disease Identification.” In Lecture Notes on Data Engineering and
Communications Technologies, 209–17. Cham: Springer Nature Switzerland, 2023. http://dx.doi.org/10.1007/978-3-031-24475-9_18.
• Chandrasekaran, Saravanan, S. Athinarayanan, M. Masthan, Anmol Kakkar, Pranav Bhatnagar, and Abdul Samad. “Edge Intelligence
Paradigm Shift on Optimizing the Edge Intelligence Using Artificial Intelligence State-of-the-Art Models.” In Advances in Computer and
Electrical Engineering, 1–18. IGI Global, 2024. http://dx.doi.org/10.4018/979-8-3693-3739-4.ch001.
• Singh, Nidhi. “Advancing Plant Disease Detection: Harnessing Deep Learning and Machine Vision.” International Journal of Science and
Research (IJSR) 13, no. 3 (March 5, 2024): 1272–77. https://doi.org/10.21275/sr24319153135.
• Picon, Artzai, Aitor Alvarez-Gila, Maximiliam Seitz, Amaia Ortiz-Barredo, Jone Echazarra, and Alexander Johannes. “Deep Convolutional Neural
Networks for Mobile Capture Device-Based Crop Disease Classification in the Wild.” Computers and Electronics in Agriculture 161 (June 2019): 280–
90. https://doi.org/10.1016/j.compag.2018.04.002.

• Chandrasekaran, Saravanan, M. Masthan, Mahesh T R, S. Athinarayanan, Ashwin Ram, and Vivek V. “Uncertainty-Aware Functional Analysis for
Electricity Consumption Prediction Using Multi-Task Optimization Learning Model.” In 2023 International Conference on Computer Science and
Emerging Technologies (CSET), 1–7. IEEE, 2023. http://dx.doi.org/10.1109/cset58993.2023.10346821.

• Srinivas, Kalyanapu, Laxman Singh, Subba Reddy Chavva, Bhasker Dappuri, Saravanan Chandrasekaran, and Shamimul Qamar. “Multi-Modal Cyber
Security Based Object Detection by Classification Using Deep Learning and Background Suppression Techniques.” Computers and Electrical
Engineering 103 (October 2022): 108333. https://doi.org/10.1016/j.compeleceng.2022.108333.

• Abade, Andre, Ana S. de Almeida, and Flavio Vidal. “Plant Diseases Recognition from Digital Images Using Multichannel Convolutional Neural
Networks.” In Proceedings of the 14th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications.
SCITEPRESS - Science and Technology Publications, 2019. http://dx.doi.org/10.5220/0007383900002108.

• Vishnoi, Vibhor Kumar, Krishan Kumar, and Brajesh Kumar. “Plant Disease Detection Using Computational Intelligence and Image Processing.”
Journal of Plant Diseases and Protection 128, no. 1 (August 30, 2020): 19–53. https://doi.org/10.1007/s41348-020-00368-0.

• Marzougui, Fatma, Mohamed Elleuch, and Monji Kherallah. “A Deep CNN Approach for Plant Disease Detection.” In 2020 21st International Arab
Conference on Information Technology (ACIT). IEEE, 2020. http://dx.doi.org/10.1109/acit50332.2020.9300072.