Improved YOLO v5s-based detection method for external defects in potato.

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Manual potato defect sorting is inefficient and biased.
• Automated systems require high accuracy and speed, posing resource challenges.
• Real-time potato defect detection is crucial for agricultural efficiency.

Purpose of the Study:

• To develop an enhanced YOLO v5s model (YOLO v5s-ours) for real-time potato defect detection.
• To improve detection accuracy and computational efficiency for automated sorting.
• To address limitations of manual sorting and existing automated systems.

Main Methods:

• Integration of Coordinate Attention (CA), Adaptive Spatial Feature Fusion (ASFF), and Atrous Spatial Pyramid Pooling (ASPP) modules into YOLO v5s.
• Development of a specialized model (YOLO v5s-ours) for real-time defect identification.
• Evaluation of model performance across six defect categories: healthy, greening, sprouting, scab, mechanical damage, and rot.

Main Results:

• The YOLO v5s-ours model achieved 82.0% precision, 86.6% recall, 84.3% F1-Score, and 85.1% mean average precision.
• Significant accuracy improvements of 24.6% (precision), 10.5% (recall), 19.4% (F1-Score), and 13.7% (mAP) over the baseline model.
• Maintained computational efficiency with a frame rate of 30.7 fps, despite a moderate increase in memory usage.

Conclusions:

• The enhanced YOLO v5s-ours model significantly improves real-time potato defect detection accuracy.
• The model provides a viable solution for developing efficient automated potato sorting systems.
• This research advances agricultural technology by overcoming limitations of traditional sorting methods.