Battlefield Reconnaissance Radar Target Classification Method Based On CN

Battlefield reconnaissance radar radar is one of the important means of battlefield intelligence acquisition,but its low resolution characteristics improve the difficulty of target recognition.Also,the battlefield environment limits the resources and power consumption of the classifier.In this paper,aiming at the application requirements of battlefield reconnaissance radar target recognition,based on the time domain and frequency domain signals of the one dimensional radar target echo,one dimensional radar target signal is constructed into two dimensional data plane,the construction is based on 2D convolutional neural network,then the network is lightweight optimized.The main contents are as follows:(1)According to the characteristics of radar target echo signal,one dimensional radar target signal is constructed into two dimensional data plane based on Toeplitz matrix feature.Then,by combining multi receptive field feature fusion with residual network structure,a 2D weighted residual CNN is proposed.The simulation results show that the recognition accuracy of this network reaches 96%.(2)The network structure is optimized,according to the requirements of storage space,computing power and classification speed in actual usage scenarios.A new CNN is obtained by using convolution kernel series substitution method,convolution kernel multiplexing and a new weight calculation method.The simulation results show that new network reduces the storage space occupation by 69.5% and the running time by 28.6% without reducing the classification performance.(3)The causes of over fitting caused by the data set constructed by Toeplitz matrix method are analyzed.On this basis,a two dimensional data construction method based on parity alternating cyclic right shift method is proposed.The simulation results show that the data constructed by this method can improve the classification performance of CNN and effectively suppress the over-fitting problem,the classification accuracy achieves 98.4% in the test set.