LSD-YOLO: Enhanced YOLOv8n Algorithm for Efficient Detection of Lemon Surface Diseases

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Lemon cultivation is vital globally, but diseases significantly reduce yield and quality.
• Early detection of lemon diseases is crucial for effective management and crop preservation.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for accurate and early detection of lemon diseases.
• To enhance feature extraction and multi-scale feature fusion for improved disease recognition.

Main Methods:

• A comprehensive dataset of 2022 lemon images (healthy and diseased) was curated.
• A novel model, lemon surface disease YOLO (LSD-YOLO), was proposed, integrating Switchable Atrous Convolution (SAConv) and Convolutional Block Attention Module (CBAM).
• The LSD-YOLO model incorporates C2f-SAC and a small-target detection layer for enhanced feature processing.

Main Results:

• The LSD-YOLO model achieved an accuracy of 90.62% and mAP@50-95 of 80.84% on the lemon disease dataset.
• The model demonstrated improved performance compared to the original YOLOv8n, particularly in mAP@50 and mAP@50-95 metrics.
• Enhanced feature extraction and multi-scale fusion contributed to superior detection capabilities.

Conclusions:

• The proposed LSD-YOLO model offers a highly accurate solution for identifying healthy and diseased lemons.
• This advancement effectively addresses the challenge of lemon disease detection, supporting agricultural productivity.
• The study highlights the potential of integrating advanced convolutional modules for agricultural disease diagnostics.