Detection of X-ray prohibited items based on Improved YOLOv7
Li Song, Yasenjiang Musha · 2023
In order to improve the detection ability of the algorithm for overlapping objects and small targets in the X-ray prohibited items image, we propose an X-ray prohibited items detection method based on improved YOLOV7. Firstly, we introduce Bottleneck Transformers to enable networks to locate accurately in high-density scenarios. In order to further improve the effects of self -attention in Bottleneck Transformers, while reducing the calculation amount, using Hydra Attention to replace the multi-head attention of the original Bottleneck Transformers. Secondly, the convolution module is integrated in Hydra Attention to enhance the local details feature. The experiment shows that the detection accuracy of the improved YOLOv7 is increased from 90.6% to 95.2%. Compared with some of the mainstream algorithm, our algorithm has certain advantages.