X-Ray Image Detection Algorithm for Prohibited Items Based on Feature Enhancement and Loss Optimization

To address the issue of low detection accuracy of prohibited items in X-ray security images caused by varying orientations, different scales, and the intertwining of targets with backgrounds, we propose a novel X-ray image detection algorithm based on feature enhancement and loss optimization. The model is built upon the ConvNext network and incorporates a Directional Channel Attention (DCA) mechanism, which efficiently captures the interaction information of local channels in different directions, thereby enhancing the accuracy of the detection model for prohibited items. Additionally, a multi-scale fusion bypass (MFB) branch is designed after the backbone network to integrate information from feature maps at different layers, thereby mitigating the interference of scale variations on the model. Furthermore, the loss function is redesigned to enable the model to automatically adjust its focus on hard samples, improving the overall detection performance. Experimental results on the SIXray dataset demonstrate that the proposed model achieves a mean Average Precision (mAP) of 91.43%, representing a 9.17% improvement over the original algorithm, thereby validating the effectiveness of the proposed method.