YOLOv8s-D2: A Novel Object Defect Detection Algorithm Based on Improved YOLOv8s

Addressing the issues of low efficiency and poor accuracy in current surface defect detection systems, as well as the challenges posed by existing defect detection models, such as large parameter counts, low generalization capabilities, and poor real-time performance, this paper proposes a novel real-time target defect detection method named YOLOv8s-D2, based on the YOLOv8s deep learning object detection algorithm. This approach optimizes and enhances the YOLOv8s object detection neural network in several key ways. Firstly, the YOLOv8s-D2 model incorporates a lightweight Convolutional Block Attention Module (CBAM) into the output feature maps of the backbone network, enhancing the network's focus on defect-related features. Secondly, it employs depthwise separable convolutions and a Squeeze-and-Excitation (SE) network to reconstruct the feature interaction module. This effectively preserves spatial and channel features while suppressing redundant information. Lastly, the three decoupled heads of YOLOv8s are optimized into four lightweight object detection heads, thereby increasing sensitivity to defective targets. The experimental results show that compared with the YOLOv8s algorithm model, the improved model has a corresponding reduction of 3.3M in parameters and 7.2G in computational complexity, while the mean average precision (mAP) value increases by 7.4%. This verifies the superiority of the improved algorithm presented in this paper.