关键词: 关键词: 卷积神经网络, 交通标志识别, 尺度依赖下采样, 特征融合, 空间金子塔池化

Abstract: Abstract: The traffic sign has a small proportion in the natural scene, the extracted features are inadequate and the recognition accuracy is low,In this paper, an improved scale-dependent pooling model is proposed for the recognition of small-scale traffic images. Firstly, based on the contour information and class feature of the deep convolution layer of the neural network, we use supplemented deep convolution layer characteristic scale dependent pooling (SD-SDP) model extract feature based on the extraction of shallow convolution feature information of scale dependent pooling (SDP) model,enriching feature information;Secondly, when The characteristics of fusion are pooled fixed dimension, the multi-scale sliding window pooling (MSP) method is used instead of the single-layer space pyramid method in the original SDP algorithm, making up the edge information of the target object.Finally, the improved SDP model is applied to the recognition of traffic signs. The experimental results show that compared with other algorithms, the accuracy of recognition is improved when the neural network classification time is reduced as little as possible.

Key words: Keywords: convolution neural network, traffic sign recognition, scale dependent pooling, feature fusion, spatial pyramid pooling