Leaf Disease Detection Using Grad-Cam

Sandeep.kr.yadav Shubham kumar Nitin kumar Priyanshu Saini

[B.Tech IT branch] [B.Tech IT branch] [B.Tech IT branch] [B.Tech IT branch]
[IMS Engineering [IMS Engineering [IMS Engineering [IMS Engineering
college, Ghaziabad] College Ghaziabad] Colloge, Ghaziabad] College, Ghaziabad]

Traditional Disease Detection Methods

Abstract- Leaf diseases significantly impact global Traditional methods for leaf disease detection primarily
agricultural productivity, leading to economic losses
rely on visual inspection by farmers or agricultural
and food insecurity. Early and accurate detection of
experts. However, these approaches have several
plant diseases is crucial for effective crop
management. Traditional disease identification limitations:
methods rely on human expertise and visual Time-Consuming: Manual inspection requires
inspection, which are timeconsuming, subjective, and significant time and effort, especially in large farms.
inefficient on a large scale. In this study, we propose Subjective and Error-Prone: Diagnosis varies based on
an optimized solution for automatic leaf disease the experience and expertise of the inspector. Limited
detection. The model is trained on a large dataset Scalability: Monitoring thousands of plants in real-time
comprising images of healthy and diseased leaves. We is impractical using manual techniques.
compare the proposed approach with traditional
machine learning methods such as Support Vector
Machine Learning in Agriculture
Machines (SVM) and Random Forest (RF), as well as
basic Convolutional Neural Networks (CNNs). The Machine Learning (ML) and Deep Learning (DL) have
experimental results demonstrate that our proposed revolutionized various domains, including agriculture.
model outperforms previous solutions, achieving These technologies can automatically detect, classify,
higher accuracy, robustness, and scalability. Our and predict diseases based on image analysis.
approach leverages deep learning techniques,
Convolutional Neural Networks (CNNs) have proven
particularly the ResNet-50 architecture, to extract
particularly effective in feature extraction and disease
intricate leaf patterns and classify diseases with
minimal human intervention.. This research classification, surpassing traditional machine learning
contributes to the advancement of precision models.
agriculture, offering an automated, cost-effective, and
scalable solution to improve crop health monitoring Objectives of This Research
and yield optimization. Develop a deep learning-based model for automated
Keywords- Leaf Disease, Deep Learning, leaf disease detection. Compare the proposed
Convolutional Neural Networks (CNNs), ResNet- approach with existing image processing and ML-based
50, Image Processing, Precision Agriculture, techniques. Improve the accuracy, efficiency, and
Machine Learning. scalability of disease classification.
I. INTRODUCTION Enable real-time disease detection for farmers
through a deployable solution.
Agriculture is the backbone of many economies
worldwide, providing food, employment, and raw
materials. However, plant diseases pose a significant
threat to agricultural productivity, often resulting in
reduced yields, economic losses, and food
shortages. Early detection and treatment of plant
diseases are essential to mitigating these adverse
effects.
Fig: Normal Leaf v/s Infected Leaf
 (C) Deep Learning-Based Approaches (Proposed
Overview of the disease detection pipeline. Solution)
Deep Learning, particularly Convolutional Neural
Networks (CNNs), has demonstrated state-of-the-art
performance in plant disease detection. CNNs
automatically learn hierarchical features from images,
eliminating the need for manual feature extraction.
Study Method
Accuracy
(%)

[1] 75.4
Image Processing +
SVM
II. LITERATURE REVIEW
This literature review contains the tabular analysis [2] Random Forest 81.2
of previous solutions and proposed solution Existing
Approaches for Leaf Disease Detection
Several studies have explored various techniques for [3] CNN (Basic) 87.5
plant disease identification:

(A) Image Processing-Based Approaches

Proposed Deep Learning 94.3
These methods rely on color, texture, and shape
Approach (CNN)
features for disease detection. Popular techniques
(ResNet-50)
include:
K-means clustering for segmenting infected
areas.
Histogram-based thresholding for feature
extraction. Advantage: CNN-based models generalize better and
Gray-Level Co-Occurrence Matrix (GLCM) for perform robust classification across multiple disease
texture analysis. types.

Limitations: Image processing techniques are highly

sensitive to background variations, lighting
conditions, and leaf orientation.

(B) Machine Learning-Based Approaches

Support Vector Machines (SVM): Uses

handcrafted features like color and texture for
classification.
Fig: Comparative bar chart of leaf disease detection
Random Forest (RF): Employs multiple decision
accuracy
trees for robust disease detection.
III. PROBLEMS WITH PREVIOUS
K-Nearest Neighbors (KNN): Classifies leaf diseases
SOLUTIONS
based on feature similarity.
Manual Inspection is Inefficient → Prone
Limitations: These methods require extensive feature to human error and time-consuming. Image
engineering and struggle with large datasets. Processing Techniques are Rigid → Sensitive to
lighting, background noise, and shape
variations.
 Traditional ML Models Lack Feature 3. Feature Extraction: ResNet-50 learns deep spatial
Extraction Power → Cannot automatically learn features.
deep spatial patterns. 4. Classification: Softmax activation for multi-class
Shallow CNN Models Overfit Small disease prediction.
Datasets → Lack generalizability for real-world
applications. Why ResNet-50?
Deeper network with skip connections → Prevents
vanishing gradient.
Manual
Inspection Higher accuracy compared to basic CNN models.
- Time-
Start Algorithm for Leaf Disease Detection Using
consuming
- Prone to human ResNet-50
error 1. Load and Preprocess Dataset Acquire a dataset
of leaf images with labeled disease categories.
Resize all images to a uniform dimension
(e.g., 224x224 pixels) for compatibility
with ResNet-50. Normalize pixel values
(e.g., scale between 0 and 1) for consistent
Traditional ML model training.
Problems Image Processing Augment dataset with transformations
- Requires manual Issues
features (e.g., rotation, flipping, brightness
- Sensitive to
adjustment) to enhance generalization.
- Cannot learn lighting
Split the dataset into training, validation,
deep patterns - Background
noise and test sets (e.g., 70% train, 20%
- Shape variations validation, 10% test).
2. Implement ResNet-50 Architecture Load a pre-
trained ResNet-50 model (with or without
ImageNet weights). Modify the fully connected
(FC) layer to match the number of disease
classes. Apply dropout and batch normalization
Shallow CNN Models layers to reduce overfitting.
- Overfit small datasets Choose an appropriate activation function
End
- Lack real -world (e.g., softmax for multi-class
generalization classification).
3. Train the Model
Compile the model with a suitable
Fig: Flowchart illustrating the problems in existing optimizer (e.g., Adam, SGD) and loss
approaches function (e.g., categorical crossentropy).
Train the model using the training
IV. PROPOSED SOLUTION dataset with batch processing (e.g.,
We propose a deep learning-based approach using batch size of 32).
ResNet-50, a state-of-the-art CNN model. Monitor validation accuracy and loss
Architecture Overview to assess performance. Implement early
stopping and learning rate scheduling to
1. Image Acquisition: Dataset of healthy and diseased
optimize training. 4. Validate Model
leaves.
Performance Evaluate the trained model on the
2. Preprocessing: Noise reduction, validation dataset.
image augmentation for better generalization.
Compute accuracy, precision, recall,
and F1-score for classification
 performance. Generate a confusion F1 Score 93.7%
matrix to analyze misclassifications.
Perform test dataset evaluation to
ensure model generalization.
5. Deploy for Real-Time Prediction Convert
the trained model into a deployable
format (e.g., TensorFlow Lite,
ONNX).
Develop a web or mobile application
to accept user-uploaded images.
Preprocess incoming images similar to
the training phase.
Perform real-time inference and
display disease classification results.
Provide recommendations for disease
treatment based on prediction results.

V. EXPERIMENTAL RESULTS Fig: Confusion Matrix

We evaluated our approach on the Plant Village
dataset, containing 38 different plant disease classes.

Accuracy Comparison

Model Accuracy (%)

SVM 75.4

Random Forest 81.2

CNN (Basic) 87.5

Fig: Comparison Graph

VI. CONCLUSION
Proposed Model 94.3
(ResNet50) This research introduces a deep learning-based leaf
disease detection system utilizing the ResNet-50 CNN
architecture. The model achieves an impressive 94.3%
Performance Metrics accuracy, significantly outperforming traditional image
processing and machine learning-based approaches.
Metric Value
Unlike conventional methods that require extensive
manual feature engineering, ResNet-50 automatically
Precision 92.5% learns hierarchical features, enhancing its robustness
across various plant species and environmental
conditions. Additionally, the model's ability to generalize
Recall 95.1% well across multiple disease types enables real-time and
scalable disease classification, making it a practical
solution for precision agriculture. The findings of this
study highlight the potential of deep learning in biochemical properties of infected leaves, further
agricultural diagnostics, paving the way for smarter and improving disease detection accuracy. Cross-Dataset
more efficient disease detection systems. Generalization: Enhancing the dataset by including
Despite its effectiveness, the system has certain diverse plant species, leaf textures, and environmental
limitations that need to be addressed. The model's conditions will improve the model’s adaptability to real-
performance could be affected by poor-quality world agricultural scenarios.
images, variations in lighting, and complex Explainable AI in Plant Disease Detection:
backgrounds. Additionally, the computational Integrating explainable AI techniques will help farmers
requirements of deep learning models, especially understand why a leaf is classified as diseased, making
during training and inference, pose challenges for the model more transparent and user-friendly.
deployment on resource-constrained devices.
Addressing these limitations by optimizing model REFERENCES
efficiency and integrating edge computing solutions
[1] Plant Leaf Disease Detection, Classification, and
could further improve real-time applications. Diagnosis Using Machine Learning and Deep Learning
Moving forward, the proposed approach can be Techniques.
extended to include multispectral and hyperspectral
imaging techniques, enabling early-stage disease [2] Machine Learning for Leaf Disease Classification: Data,
Techniques, and Challenges.
detection before visible symptoms appear.
Furthermore, integrating IoT-based smart farming [3] Image-Based Crop Disease Detection Using Machine
solutions with automated drones and sensor Learning.
networks can enhance real-time crop monitoring and
[4] Identification of Plant Leaf Diseases by Deep Learning
disease prevention strategies. Future research can Based on Improved Convolutional Neural Network.
also explore the use of explainable AI (XAI)[20] to
provide interpretable disease diagnoses, making the [5] Detection and Identification of Plant Leaf Diseases Using
system more accessible to farmers and agricultural YOLOv4.
experts. In conclusion, this study highlights the
[6] Detection of Plant Leaf-Based Diseases Using Machine
transformative potential of deep learning in Learning Approach.
agricultural diagnostics.
[7] Plant Leaf Disease Detection and Classification Using
VII. FUTURE SCOPE Deep Learning Models.

Although the proposed model demonstrates high [8] ViTaL: An Advanced Framework for Automated Plant
accuracy and robustness, there is still room for Disease Identification in Leaf Images Using Vision
Transformers and Linear Projection for Feature
improvement and expansion: Reduction.

Mobile Application Deployment: [9] Real-Time Plant Health Assessment via Implementing
Implementing the trained model in mobile CloudBased Scalable Transfer Learning on AWS
DeepLens.
applications will allow farmers and agricultural
experts to detect plant diseases in the field using just [10] Leaf Tar Spot Detection Using RGB Images.
a smartphone camera, providing instant
feedback and treatment [11] Leaf-Based Plant Disease Detection and Explainable AI.
recommendations.
IoT-Based Smart Farming Integration: The [12] A Comprehensive Survey on Plant Leaf Disease Detection
Using Deep Learning, Computers and Electronics in
system can be integrated into IoT-enabled smart farming Agriculture , 2021.
solutions, where automated drones and sensor-based
cameras continuously monitor crop health and detect [13] Deep Learning-Based Plant Disease Detection: A Review,
Journal of Plant Pathology , 2020.
diseases in real time, reducing human effort.
Multispectral and Hyperspectral Imaging: Future [14] Transfer Learning with Convolutional Neural Networks
research can extend the model to analyze multispectral for Plant Disease Diagnosis, Computers and Electronics
and hyperspectral images, capturing hidden patterns and in Agriculture , 2020.
 [32] A Mobile-Based System for Maize Plant Leaf Disease
[15] Automated Leaf Disease Detection with Deep Detection and Classification Using Deep Learning, Frontiers in Plant
Convolutional Neural Networks, Remote Sensing , 2021. Science , 2023.

[16] Plant Disease Detection Using Deep Learning and [33] Plant Leaf Disease Detection and Classification Using Deep
Convolutional Neural Networks, International Journal of Learning, arXiv preprint , arXiv:2501.03305, 2025.
Advanced Science and Technology , 2020.
[34] Deep Learning-Based Leaf Disease Detection in Crops
[17] Deep Learning Approaches for Plant Disease Detection Using Convolutional Neural Networks, Agronomy , 2022.
and Diagnosis, Plant Pathology Journal , 2021.
[35] A Hybrid Framework for Plant Leaf Disease Detection and
[18] A Review on Image Processing Techniques for Plant Classification, Complex & Intelligent Systems , 2024.
Disease Detection, International Journal of Computer
Applications , 2020. [36] A Novel Plant Type, Leaf Disease, and Severity
Identification Framework Using Deep Learning, Scientific Reports ,
[19] Convolutional Neural Networks for Image-Based Plant 2024.
Disease Detection, Biosystems Engineering , 2021.
[37] Identification of Plant Leaf Diseases by Deep Learning
[20] Plant Disease Detection Using Image Processing and Based on Improved Convolutional Neural Network, Frontiers in Plant
Machine Learning, International Journal of Computer Science , 2022.
Science and Information Security , 2020.
[38] Plant Disease Detection and Classification by Deep
[21] Advancements in Plant Disease Detection Using Deep Learning—A Review, IEEE Access , 2021.
Learning, Artificial Intelligence in Agriculture , 2021.
[39] ViTaL: An Advanced Framework for Automated Plant
[22] A Survey on Deep Learning in Agriculture: Techniques Disease Identification in Leaf Images Using Vision Transformers and
and Applications, Computers and Electronics in Linear Projection for Feature Reduction, arXiv preprint ,
Agriculture , 2020. arXiv:2402.17424, 2024.

[23] Deep Learning for Image-Based Plant Disease Detection, [40] Deep Learning for Plant Identification and Disease
Frontiers in Plant Science , 2021. Classification from Leaf Images: Multi-Prediction Approaches, arXiv
preprint , arXiv:2310.16273, 2023.
[24] Machine Learning Techniques for Plant Disease
Detection: A Review, International Journal of [41] S. P. Mohanty, D. P. Hughes, and M. Salathé, "Using Deep
Engineering Research and Technology , 2020. Learning for Image-Based Plant Disease Detection," *Frontiers in
Plant Science*, vol. 7, p. 1419, 2016.
[25] Deep Learning Models for Plant Disease Detection and
Diagnosis, Computers and Electronics in Agriculture , [42] L. Liu, C. Zhang, J. Wang, and X. He, "Plant Disease
2021. Detection Using Deep Learning: A Systematic Review," *Computers
and Electronics in Agriculture*, vol. 185, p. 106125, 2021.
[26] Plant Disease Detection Using Deep Learning: A Review,
International Journal of Computer Applications , 2020. [43] P. Oppenheim, R. Shavit, and I. Klein, "Automated Plant
Disease Detection Using Transfer Learning," *IEEE Transactions on
Computational Imaging*, vol. 6, pp. 328–337, 2020.
[27] Applications of Deep Learning in Plant Disease
Detection, Computers and Electronics in Agriculture ,
2021. [44] A. Sharma and M. K. Verma, "Hybrid Deep Learning Model
for Leaf Disease Detection in Agricultural Crops," *Artificial
Intelligence in Agriculture*, vol. 7, pp. 55–65, 2023.
[28] A Comprehensive Review on Plant Disease Detection
Using Image Processing and Machine Learning,
International Journal of Computer Science and [45] K. A. Khan, S. Khan, and M. S. Rahman, "An Ensemble
Information Security , 2020. Approach for Crop Disease Identification Using Convolutional Neural
Networks," *Biosystems Engineering*, vol. 204, pp. 198– 209, 2022.
[29] Deep Learning Techniques for Plant Disease Detection: A
Review, Biosystems Engineering , 2021.

[30] Recent Advances in Plant Disease Detection Using Deep

Learning Approaches, Artificial Intelligence in
Agriculture , 2021. [31] Plant Leaf Detection and Disease
Recognition Using Deep Learning, IEEE Conference
Publication , 2020.