TRAFFIC SIGN DETECTION BASED ON IMAGE MASK AND FEATURE FUSION

In order to solve the problem of small object detection in traffic sign detection, a feature recurrent fusion object detection model based on YOLOX is proposed. The feature recurrent fusion method was used to deepen the mutual complementation of position information and semantic information between different feature maps, and the channel attention mechanism was used to deepen the model’sattention to small targets in the fusion process. The original image was used to generate image mask, and the mask was used to optimize loss function, thereby reducing the false detection of non-traffic signs by the model, and using the data-balanced data augmentation method to improve the overall detection performance of the model. Experimental results on public datasets show that the proposed method can effectively improve the detection accuracy of small targets in traffic signs.