Improved YOLOX detection algorithm for contraband in X-ray images

• DOI: 10.1364/ao.461627
• Publication Date: 2022-06-30T15:01:13Z
• Authors: Yinsheng Zhang; Wenxiao Xu; Shanshan Yang; Yongjie Xu; Xinyuan Yu
• Keywords: Benchmark (surveying); Feature (linguistics); Public security

It is important to perform contraband inspections on items before they are taken into public places in order ensure the safety of people and property. At present, mainstream method judging that security inspectors observe X-ray image objects judge whether belong contraband. Unfortunately, often hidden under other normal objects. In a high-intensity working environment, very prone missed detection wrong detection. To this end, framework based computer vision technology proposed, which trained improved basis current state-of-the-art YOLOX object network, adopts strategies such as feature fusion, adding double attention mechanism classifying regression loss. Compared with benchmark YOLOX-S model, proposed achieves higher average accuracy, an improvement 5.0% SIXray dataset, opening way large-scale automatic places.