Research on glass surface defect detection method based on shadowgraphy imaging and improved YOLOv11n

Addressing the challenges of observing and detecting defects on ordinary glass surfaces, this paper proposes a defect detection method combining shadowgraphy technology with the YOLO_FSA algorithm. By constructing a glass surface defect image dataset using a shadowgraphy system and adopting YOLOv11n as the base model, the C3k2_FWD module is introduced to achieve efficient lightweight spatial-channel feature extraction and detection of elongated defects.A GSCSA attention mechanism is constructed to enhance perception of faint defect features. ADown convolution replaces traditional downsampling, achieving model lightweighting while maintaining detection accuracy. Experimental results show that the YOLO_FSA algorithm achieves precision, recall, and mAP50 of 74.8%, 80.7%, and 85.5%, respectively. Compared to YOLOv11n, these metrics improved by 8.5%, 6.4%, and 12.6%, respectively. Concurrently, the model parameters were reduced by 31.4% to 1.77 million, and computational complexity decreased by 25.4% to 4.7 GFLOPs, achieving an effective balance between detection performance and computational efficiency.