WCSE

Research on Small Target Detection Algorithm Based on Improved YOLOX

Abstract— Aiming at the problem of insufficient feature extraction and insufficient accuracy in different stages of feature fusion in the newly launched YOLOX algorithm, an improved YOLOX target detection algorithm is proposed. This method first incorporates the convolutional attention module in the feature extraction stage, which can better capture the original rich information of the feature. Then integrate the attention mechanism in the path fusion module to further improve the feature fusion effect. Finally, the complete intersection ratio loss function is introduced in the bounding box regression process to improve the convergence speed and accuracy of the regression process. In order to test the detection effect of the algorithm, experiments were performed on the MS COCO data set and the PASCAL VOC data set. Compared with the improved YOLOX algorithm, the average accuracy of the proposed algorithm on the two data sets is increased by 1.9% and 4%, respectively, and the effect is improved significantly.

Index Terms— deep learning, object detection, ciou, yolox, mff

Caixia Meng
Xi'an University of Posts & Telecommunication
Hongpeng Chu
Xi'an University of Posts & Telecommunication
Jiabao Zhang
Xi'an University of Posts & Telecommunication
Kaijie Xi
Xi'an University of Posts & Telecommunication

[Download]

Cite: Caixia Meng, Hongpeng Chu, Jiabao Zhang, Kaijie Xi, " Research on Small Target Detection Algorithm Based on Improved YOLOX, " WCSE 2022 Spring Event: 2022 9th International Conference on Industrial Engineering and Applications, pp. 507-516, Sanya, China, April 15-18, 2022.