As shown in Figure 8, there is a big difference between the actual Expressway scene data set and the sample data set used for training. Although the algorithm in this paper adds the domain adaptive component, it is still not ideal for small-scale target detection, because fast In the r-cnn model, although the RPN structure is introduced to extract multi-scale vehicle feature map, the RPN can only receive the features extracted from the last layer of the convolutional layer network, which seriously reduces the detection accuracy of the model for small-scale targets. For this problem, the fifth chapter will propose solutions based on the actual Expressway case analysis in Jiangxi Province. The algorithm proposed in this paper can basically detect large-scale vehicle targets in a relatively small number of scenes and a good weather environment. Generally speaking, the algorithm proposed in this paper basically solves the problem of domain difference due to the lack of data set samples and replaces the manual feature extraction in the traditional vehicle detection algorithm, which improves the accuracy of model detection to a certain extent. Therefore, the detection algorithm proposed in this paper performs well in the whole body.<br>
正在翻译中..