Research on Prohibited Item Detection Algorithms in X-ray Images

To address challenges such as target overlap, complex backgrounds, and missed detections of small objects in X-ray image contraband detection, we propose an improved YOLOv7-based contraband detection model. Firstly, in the feature fusion stage, we introduce a Coordinate Attention (CA) module to enhance the network's ability to extract essential features, mitigating background interference caused by overlaps. Secondly, we design an MPC2F module to reduce feature loss and enhance the network's capability to detect small objects. Lastly, to expedite model convergence and enhance its generalization, we employ the AdaN adaptive momentum optimizer during training.The improved model achieves an average detection accuracy of 93.4% and 84.2% on the SiXary and HiXary contraband datasets, respectively, representing a 1.3% and 1.4% improvement over the baseline YOLOv7 model. Experimental results demonstrate that our proposed model effectively enhances contraband detection performance in X-ray images compared to existing mainstream networks, significantly reducing missed detection rates caused by occlusion and small target issues.