Dangerous Goods Detection in X-Ray Security Inspection Images Based on Improved YOLOv8

The article presents an enhanced YOLOv8 model, YOLOv8-APSE, for detecting dangerous goods in X-ray security images. YOLOv8-APSE integrates the Adaptive Feature Pyramid Network (AFPN) and Squeeze-and-Excitation (SE) detection heads, focusing on efficient multi-scale feature fusion and channel-wise feature recalibration, respectively. This approach addresses the challenges of traditional feature fusion complexities and improves detection accuracy, especially in cluttered scenes. Tested on the PIDray dataset, YOLOv8-APSE demonstrates superior performance in detecting small and overlapping objects against complex backgrounds, contributing valuable insights and methodologies for future developments in target detection models.