LCGSC-YOLO: a lightweight apple leaf diseases detection method based on LCNet and GSConv module under YOLO framework

Area of Science:

• Computer Vision
• Agricultural Technology
• Deep Learning

Background:

• Current deep learning models for plant disease detection are computationally intensive.
• Apple leaf disease identification presents complex scenarios for automated systems.

Purpose of the Study:

• To develop a lightweight and efficient deep learning model for apple leaf disease detection.
• To reduce model parameters and computational load while maintaining high accuracy.

Main Methods:

• The study proposes LCGSC-YOLO, a modified YOLO framework incorporating LCNet (Lightweight CPU Convolutional Neural Network) as the backbone.
• Integration of GSConv (Group Shuffle Convolution) and VOVGSCSP modules in the neck network for efficient feature fusion.
• Inclusion of coordinate attention mechanisms to mitigate accuracy loss from model lightweighting.

Main Results:

• LCGSC-YOLO achieved a mean average precision (mAP) of 95.5% on combined datasets (Plant Pathology 2021 and AppleLeaf9).
• The model demonstrated a detection speed of 53 frames per second (FPS).
• LCGSC-YOLO significantly reduced the number of parameters and Floating Point Operations (FLOPs) compared to other algorithms.

Conclusions:

• LCGSC-YOLO offers a highly efficient and accurate solution for apple leaf disease detection.
• The lightweight design makes it suitable for resource-constrained environments.
• This method addresses the trade-off between model complexity and detection performance in agricultural applications.