A lightweight dual-attention network for tomato leaf disease identification.

Area of Science:

• Agricultural science
• Computer vision
• Machine learning

Background:

• Tomato disease recognition is vital for agriculture, but current deep learning methods struggle with imbalanced data, unclear features, and variations within and between classes.
• Existing machine vision techniques for plant disease identification face limitations in accurately capturing subtle disease indicators and handling data inconsistencies.

Purpose of the Study:

• To develop an advanced machine vision method for accurate tomato leaf disease classification and recognition.
• To overcome challenges in current deep learning models, including imbalanced datasets, feature ambiguity, and label noise.

Main Methods:

• Image enhancement using piecewise linear transformation and oversampling to address dataset imbalance.
• Introduction of a lightweight model, LDAMNet, incorporating a Dual Attention Convolutional block (DAC Block) with Hybrid Channel Attention (HCA) and Coordinate Attention (CSA).
• Implementation of a Robust Cross-Entropy (RCE) loss function to mitigate the impact of noisy labels during training.

Main Results:

• Achieved an average recognition accuracy of 98.71% on a tomato disease dataset, demonstrating effective disease information retention and area capture.
• The proposed method showed strong generalization performance on rice crop disease datasets, indicating broad applicability across different crops.
• The LDAMNet model with the DAC Block and RCE loss function significantly improved feature extraction and classification accuracy.

Conclusions:

• The developed machine vision method offers a robust and accurate solution for tomato leaf disease recognition, outperforming existing approaches.
• The study provides novel insights and techniques applicable to crop disease recognition across various agricultural applications.
• Future work should focus on optimizing model efficiency and validating performance in real-world agricultural settings.