YOLO-SRW: An Enhanced YOLO Algorithm for Detecting Prohibited Items in X-Ray Security Images

To address the challenges of false positives and false negatives in prohibited item detection within X-ray security images, caused by complex backgrounds, poor image quality, and varying scales, this paper proposes an improved algorithm based on YOLOv8, named YOLO-SRW, to improve the accuracy of detecting prohibited items. First, inspired by the receptive field attention mechanism and integrating the Large Separable Kernel (LSK) attention mechanism, we designed the Receptive Field Large Separable Kernel Attention (RFLSKA) module. This module dynamically adjusts the spatial receptive field to capture contextual information at different scales, thereby enhancing the understanding and the response to small objects and providing a more comprehensive scene interpretation. Then, we integrate the Shallow Robust Feature Downsampling (SRFD) module to enhance the shallow feature extraction in YOLOv8, enhancing the model’s ability to extract features from low-resolution and feature-sparse targets, thus reducing object information loss. Finally, by combining SCYLLA-IoU (SIoU) and Wise-IoUv3 losses, we design the Wise-SIoU loss function to reduce false negatives and false positives in Prohibited item detection, enhancing the model’s generalization ability. Experimental results on the SIXray dataset demonstrate that YOLO-SRW achieves an mAP of 92.8%, surpassing the original YOLOv8 by 2.4%. Compared to existing mainstream algorithms, YOLO-SRW offers higher detection accuracy, effectively reducing false negatives and false positives for small and multi-scale objects in X-ray security images.