Optimized YOLOV10 for X-Ray Image Dangerous Object Detection
Yulong Zhang, Jiaqi Niu, Yuepeng Chen, Jin Chen · 2024
Accurate analysis of X-ray images is crucial for detecting potential hazardous items in luggage and packages in security screening. With the advancement of deep learning, automated object detection algorithms like YOLOv10 have demonstrated remarkable performance. This study aims to further enhance YOLOv10's detection capabilities, particularly in recognizing objects with complex backgrounds and small targets in X-ray images. By incorporating SPDConv and Bidirectional Feature Pyramid Network (BiFPN), we significantly improved the model's performance in detecting prohibited items in X-ray images. Additionally, we compared the optimized YOLOv10 model with other mainstream object detection methods, including SSD and YOLOv5, and conducted ablation experiments to evaluate the impact of different components on model performance. Experimental results show that the optimized YOLOv10 achieved significant improvements across multiple evaluation metrics, validating the effectiveness of the adopted techniques.