基于卷积神经网络的车辆型号识别研究

Research on vehicle model identification based on convolutional neural network

为了解决交通系统中车辆型号识别率还不够高的情况,通过可视化手段优化了特征提取的步骤,同时设计了车辆识别的分类器模型和一系列训练策略。运用选择性搜索方法对样本进行分析,由此得出候选区域,之后利用融合算法和边框回归算法得出真实车辆所在区域的候选窗口。在车辆候选窗口被标出后,利用卷积神经网络对候选窗口的特征进行提取,送入到神经网络中进行分类,最终得出车辆的具体型号。通过实验表明,提出的基于卷积神经网络的图像识别算法与传统的卷积神经网络以及SVM比较,在车辆识别上都有更好的识别率。

In order to solve the problem that the vehicle type recognition rate is not high enough in the traffic system,in this paper,steps of feature extraction are optimized by visualization means,and the classification model for vehicle recognition and a series of training strategies are designed at the same time.Apply the selective search algorithm to analysis of samples,then get the candidate region,and further,derive the candidate window of the actual area of vehicles by a fusion algorithm and a bounding box regression algorithm.After a vehicle candidate window is marked out,extract the feature of the candidate window by a convolutional neural network,and send the feature to the neural network for classification,and finally obtain the specific model of vehicle.Experiments show that the proposed image recognition algorithm based on convolutional neural network has better recognition rate than the vehicle based on traditional convolution neural network and SVM.