LightMixer: A novel lightweight convolutional neural network for tomato disease detection

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Tomato cultivation is globally significant, but diseases threaten crop health and yield.
• Traditional deep learning models for disease identification are computationally intensive and parameter-heavy.
• Computer vision offers a promising solution for automated disease detection in crops.

Purpose of the Study:

• To design a lightweight and efficient model for tomato leaf disease identification.
• To address the high computational costs associated with traditional deep learning methods.
• To enable on-device automatic disease diagnosis for improved crop management.

Main Methods:

• Development of the LightMixer model, incorporating depthwise convolution with a Phish module and a light residual module.
• The Phish module utilizes spliced nonlinear activation functions with depthwise convolution for feature extraction.
• Light residual blocks were employed to enhance computational efficiency and minimize feature information loss.

Main Results:

• The LightMixer model achieved an accuracy of 99.3% on public datasets.
• The model requires only 1.5 million parameters, significantly fewer than classical and other lightweight models.
• Demonstrated superior performance compared to existing convolutional neural network and lightweight models.

Conclusions:

• LightMixer is a highly accurate and computationally efficient model for tomato leaf disease identification.
• Its lightweight nature makes it suitable for deployment on mobile devices for real-time crop monitoring.
• The model represents a significant advancement in applying computer vision for sustainable agriculture and disease management.