X-ray Security Inspection Image Detection Algorithm Based on Improved YOLOv4

Abstract

The existing object detection algorithms have low recognition accuracy for prohibited items due to the complex background, large variation of target scale, and mutual occlusion of objects in X-ray security inspection images. In order to accurately identify prohibited items in real-time, an X-ray security inspection image detection algorithm based on improved YOLOv4 is proposed. Firstly, deformable convolution is introduced into the network to improve the feature extraction ability of prohibited items. Then, GHM loss is used to optimize the loss function, so that the model can focus on the difficult classification samples that are more effective for training improvement. Finally, the non-maximum suppression method combining soft NMS and DIoU NMS is used to improve the detection ability of the algorithm for occluded targets. Experiments on the X-ray security inspection image dataset show that the mAP of the improved algorithm reaches 91.4%, which is 3.3% higher than the YOLOv4, and the detection speed meets the real-time requirements.