Improved Detection Algorithm of Floating Foreign Bodies in Xilin Bottles Based on YOLOv8

To address the challenges of minimal class differences, confusion between targets and background, and difficulty in identifying small targets during the detection of floating foreign bodies in Xilin bottles, this paper proposes an enhanced target detection model, YOLOv8-PSP, based on the YOLOv8 algorithm. To tackle the issue of small differences between floating foreign bodies, the PPHGNet network module is incorporated into the backbone network, which enhances the semantic information transmission across scales, thereby improving feature fusion and the recognition of inter-class differences. To address the weak characteristics of floating foreign bodies and their susceptibility to background interference, the SCAM attention mechanism is introduced into the neck network, enhancing the feature representation of small targets and effectively distinguishing between foreground and background. For the challenge of small and hard-to-identify floating foreign bodies, the PIoU loss function is employed to accommodate the aspect ratio variations of small targets, improving regression accuracy and promoting faster model convergence. Experimental results demonstrate that the improved model achieves 96.0% accuracy, 90.0% recall, and 94.4% [email protected] on the Xilin bottles floating foreign body defect dataset, representing increases of 6.4%, 6.1%, and 3.8% respectively compared to previous methods. The model is also validated on the NEU-DET dataset.