Leaf Classification for Crop Pests and Diseases in the Compressed Domain

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Crop pests and diseases significantly reduce food production and threaten food security.
• Traditional neural networks require extensive data processing, limiting efficiency.
• Compressed sensing (CS) offers a method to reduce data processing by analyzing only a fraction of the data.

Purpose of the Study:

• To develop an efficient and accurate method for classifying and identifying pest images.
• To reduce the computational burden of pest identification using neural networks.
• To explore the application of compressed sensing within neural network architectures for image analysis.

Main Methods:

• A novel network model, CSBNet, was proposed, integrating compressed sampling and classification.
• Compression was embedded within the neural network, unlike conventional CS methods.
• An attention mechanism was incorporated to enhance feature representation in the compressed domain.
• Recognition was performed directly on compressed data without image reconstruction.

Main Results:

• CSBNet achieved a maximum accuracy of 96.32% at a sampling rate of 0.7.
• The model demonstrated superior accuracy compared to other methods (93.01%, 83.58%, 87.75%) under the same conditions.
• CSBNet required substantially fewer trainable parameters than competing models.

Conclusions:

• The proposed CSBNet effectively classifies pest images in the compressed domain, offering a computationally efficient solution.
• Integrating compressed sensing principles directly into neural networks enhances pest identification accuracy and reduces resource requirements.
• This approach holds significant potential for improving agricultural monitoring and ensuring food security.