The research on the recognition method of foreign fibers in cotton based on improved lightweight YOLOv8

To address the challenges of high detection difficulty and labor costs in foreign fiber detection in seed cotton, this study proposes an improved lightweight YOLOv8-based recognition method. First, we introduce a lightweight feature extraction network module, GRGCE, to reduce model complexity. To enhance model accuracy while maintaining lightweight performance, the FasterBlock and CGLU were integrated into the neck network, forming a C2f-Faster-CGLU module. This effectively mitigated the accuracy loss caused by parameter reduction. Experimental results demonstrate that the improved model achieves an [email protected] of 0.682 for fiber detection (a 6.4% improvement over the original YOLOv8n) with only 1.37M parameters, while maintaining an inference speed of 65 FPS. The proposed method significantly outperforms comparison models such as PP-YOLOE-S. This approach provides the first automated detection solution for ginning production lines that combines high accuracy, real-time performance, and low deployment costs.