Traffic signs recognition model over data augmentation based on Yolov5

Abstract

Due to the long time consuming of data collection, it is difficult to label small targets of data samples, the amount of data is small, and the sample distribution is uneven. At the same time, the proportion of small targets is small, the missed detection rate is high, and the model feature fusion is insufficient. In order to pay more attention to the detection target and improve the feature extraction ability of the algorithm. In this regard, this paper proposes a method to generate a new sample dataset by expanding the existing dataset samples, and integrates the Convolutional Block Attention Module (CBAM) into the backbone feature extraction network. The scale feature fusion module, combined with the yolov5 target detection model, achieves the purpose of improving the detection rate of individual identification and enhancing the generalization ability of the target detection model. This data augmentation method enriches the traffic sign dataset and improves the robustness of the model, making it more suitable for practical scenarios. Taking the TTIOOK dataset as an example, its experimental results demonstrate the effectiveness and superiority of the proposed method compared with the unimproved method.