Object Detection Research Based On Convolutional Neural Network

A general object detection method based on convolution neural network is proposed.The improvement mainly involves two aspects,including how to select the candidate area extraction method and the optimization of reclosing window merging mechanism.First,on the problem of candidate region extraction,based on the anchors mechanism of RPN,a clustering anchors mechanism based on ground truth boxes is proposed,which makes the selection of candidate regions more purposeful,so that the classifier can better learn the target features.Second,a detection frame merging mechanism,which is different from non maximum value suppression,is proposed.By reducing the fraction of the non maximum score detection frame instead of the direct suppression,the mechanism of recalculating the score is to calculate the coincidence degree between the detection frame and the frame with the fractional maximum value by using the Gauss function..Aiming at two improvements,a series of experiments have been carried out in the dataset of the universal object detection database VOC2007?which is widely used internationally.The experimental results show that the proposed two improved methods can effectively improve the detection performance.Under the same test conditions,two improved methods have improved the detection performance by 1.7%compared with the R-FCN algorithm,reaching a 81.2%detection result.A classification method of traffic speed limit signs based on convolution neural network is proposed.First,the speed limit label database with different speed limits is set up.The network is designed by network,and the parameters of the neural network are initialized by the public library GTSRB,then the parameters are adjusted in the custom training set and the test eoncentration is classified.Device performance verification.In the experiment,the sample was expanded and the number of negative samples was changed.The average accuracy of the final CNN classifier on the test set of the traffic speed limit sign was 99.7%.Experimental results show that the convolution neural network model proposed in this paper achieves better classification accuracy in speed limit sign classification tasks.