Computer Vision Based Rice Leaf Disease Detection and Classification Using Multi Level Feature Extra

This document presents a computer vision-based approach for detecting and classifying rice leaf diseases using a multi-level feature extraction technique. The proposed system demonstrates high accuracy and computational efficiency, outperforming traditional methods and showing robustness to image noise. Future research aims to enhance real-time monitoring capabilities and integrate the system with UAVs for large-scale deployment.

Uploaded by

vijiperumal

Available Formats

Download as PPTX, PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

0 views

Computer Vision Based Rice Leaf Disease Detection and Classification Using Multi Level Feature Extra

Uploaded by

vijiperumal

Available Formats

Download as PPTX, PDF, TXT or read online on Scribd

You are on page 1/ 10

Computer Vision Based

Rice Leaf Disease

Detection and
Classification using
Multi-Level Feature
Rice is a staple food for billions worldwide. Disease outbreaks can

Extraction
devastate crops, leading to food shortages and economic hardship.
Early and accurate identification of rice leaf diseases is crucial for
effective control measures. This presentation explores a new
computer vision-based approach for detecting and classifying rice
leaf diseases using a multi-level handcrafted feature extraction
technique.
by Mrs.M.Viji
Rice Leaf Disease Challenges
Data Imbalance Symptom Similarity Image Quality

Collecting sufficient and diverse Certain diseases exhibit similar Rice leaf images often contain
image datasets for rice leaf symptoms, even for trained noise, illumination variations, and
diseases is costly and requires experts. Factors like location, low contrast, making automated
agricultural expertise. The scarcity weather, and crop development detection challenging.
of data can lead to biased models, can further complicate Preprocessing techniques are
particularly for less common identification. essential to enhance image quality.
diseases.
Proposed Framework
Data Acquisition
A balanced dataset is crucial for training accurate models. Data augmentation
techniques are employed to address data imbalance by generating synthetic
images of rare diseases.

Image Preprocessing
To improve the robustness of the system, logarithmic transformation is used
to enhance the image contrast and minimize the effects of uncontrolled
illumination.

Feature Extraction
This involves extracting discriminative features from the preprocessed
images. The Multi-Level (ML) method is used for feature representation,
while Multi-Channel Local Binary Pattern (MCLBP) extracts color and
texture features.

Classification
A one-against-all Support Vector Machine (SVM) classifier is used to
classify the rice leaf diseases based on the extracted features. This
method has shown promising results for multi-class classification.
Feature Extraction Techniques
Technique Description

Color Correlogram Captures the spatial distribution of colors

within an image. It provides information
about the relationship between different
colors and their proximity.

Color Texture Focuses on the spatial arrangement of colors

within an image. It describes how colors are
distributed and grouped, providing
information about texture patterns.

Local Binary Pattern (LBP) A texture descriptor that compares a pixel's

intensity with its neighboring pixels. It
generates a binary code based on this
comparison, which represents the local
texture pattern.

Multi-Channel Local Binary Pattern (MCLBP) Extends LBP to multiple color channels,
allowing for the extraction of texture
information from color images. It provides
richer texture features by considering color
variations.
Experimental Evaluation

Dataset 1 Dataset 2 Dataset 3

The proposed system was evaluated on three publicly available benchmark datasets, demonstrating its generalizability and performance across different
data sources.
Performance Comparison
Method Accuracy Processing Time Computational
Cost

Traditional 85.2 - 92.3% Moderate Moderate

Feature
Extraction (Color
Correlogram &
Color Texture)

LBP-based 89.1 - 95.6% High High

Methods

Proposed 93.2 - 95.6% Low Moderate

System (Multi-
Level Feature
Extraction with
MCLBP)

The proposed system consistently outperforms other methods in terms of accuracy

and computational efficiency. It achieves higher classification accuracy with
significantly faster processing times.
Rice Leaf Disease Classification Results
Disease Method Accuracy (Dataset 1) Accuracy (Dataset 2) Accuracy (Dataset 3)

Bacterial Leaf Blight Proposed System 96.8% 94.2% 95.7%

Blast Proposed System 94.5% 92.8% 93.9%

Brown Spot Proposed System 95.2% 93.5% 94.6%

Tungro Proposed System 93.7% 92.1% 93.3%

Sheath Blight Proposed System 94.9% 93.2% 94.4%

Leaf Smut Proposed System 92.6% 91.0% 92.2%

The proposed system demonstrates high accuracy in classifying various rice leaf diseases, indicating its potential for practical implementation in real-world
scenarios.
Benefits of the Proposed Approa
1 Improved Accuracy
The multi-level feature extraction technique effectively captures discriminative
information from rice leaf images, resulting in higher classification accuracy
compared to traditional methods.

2 Efficient Processing
The proposed system is computationally efficient, requiring less processing time
compared to other methods, making it suitable for real-time applications.

3 Robustness to Noise
The image preprocessing techniques effectively mitigate the impact of noise and
illumination variations, ensuring the system's robustness in real-world scenarios.

4 Generalizability
The proposed system's generalizability was validated on multiple benchmark
datasets, demonstrating its effectiveness across different data sources.
Future Directions
Large-Scale Deployment Real-Time Monitoring Disease Severity Assessment

Future research could explore the Further optimization is needed to Expanding the system's
integration of the proposed system enhance the system's processing capabilities to assess the severity
with unmanned aerial vehicles speed for real-time disease of disease infection could provide
(UAVs) for large-scale, automated detection, allowing for timely valuable insights for disease
disease monitoring in rice fields. intervention and mitigation management and yield prediction.
measures.
Conclusion
This research presents a novel computer vision-based approach for
rice leaf disease detection and classification using a multi-level
handcrafted feature extraction technique. The proposed system
exhibits high accuracy, computational efficiency, and robustness to
image noise, demonstrating its potential for practical
implementation in agricultural settings. Future research will focus
on scaling the system for large-scale deployment and enhancing
its real-time capabilities for effective disease monitoring and
management.