Leaf disease detection and classification in food crops with efficient feature dimensionality reduction.

Area of Science:

• Computer Vision
• Machine Learning
• Agricultural Science

Background:

• Computer vision relies on feature extraction for image classification.
• Dimensionality reduction is crucial for computational efficiency in deep learning models.
• Leaf image analysis faces challenges with high dimensionality, impacting disease identification accuracy.

Purpose of the Study:

• To propose a novel computer vision system integrating Convolutional Neural Networks (CNNs) and Bidirectional Long Short-Term Memory (BiLSTM) networks.
• To address the high dimensionality of leaf image data through effective feature reduction.
• To enable early and accurate identification of plant diseases for improved crop management.

Main Methods:

• Feature extraction using CNNs.
• Temporal dependency modeling using BiLSTM networks.
• Integration of label information as constraints for discriminative feature learning.
• Dimensionality reduction of extracted features using the proposed ELFDR-LDC-CNN-BiLSTM model.

Main Results:

• Achieved 99.37% classification accuracy on pepper and maize leaf image datasets.
• Demonstrated superior performance compared to existing dimensionality reduction techniques.
• Successfully reduced feature dimensionality while enhancing classification effectiveness.

Conclusions:

• The proposed ELFDR-LDC-CNN-BiLSTM model offers a cost-effective solution for automated leaf disease detection.
• The system can be integrated into precision agriculture for enhanced crop monitoring and yield improvement.
• This approach facilitates sustainable farming practices through early and accurate disease identification.