One-dimensional convolutional neural network ground radar target classification method based on fusion features

Abstract

The invention discloses a one-dimensional convolutional neural network ground radar target classification method based on fusion features. The method comprises the following steps: preprocessing radarecho data samples in a training sample set and a test sample set, obtaining an amplitude spectrum and a power spectrum of echo data through FFT, performing power transformation on the amplitude spectrum, and taking three groups of feature vectors as three channels of feature input; determining a one-dimensional convolutional neural network architecture, and sending the extracted three-channel features to a full connection layer; performing classifying by utilizing a softmax classifier, and performing calculating to obtain an error between an output prediction label and a real label; correcting the network weight coefficient by using a gradient descent method until the maximum number of iterations is reached, extracting the corresponding network weight coefficient, and determining a finalclassifier model; and sending a to-be-identified test set radar echo sample into the classifier obtained by training to finish target classification. The method has fewer parameters, and can efficiently and accurately finish target classification.

Description

technical field [0001] The invention belongs to the technical field of radar digital signal processing, in particular to a one-dimensional convolutional neural network ground radar target classification method based on fusion features. Background technique [0002] At present, low-resolution radar target recognition is mostly based on artificial feature extraction of radar target data, and the recognition performance depends on the quality of features. For pulse Doppler radar, the widely used target identification method is the radar audio signal generated by the radar operator to identify different ground moving targets. However, this method puts forward high training requirements for radar operators, and the sensory differences and subjectivity of operators will lead to non-real-time classification, which is not suitable for real-time operation. In addition, low-resolution radar can also obtain the power spectrum through Fourier transform, and then artificially extract fe...

AI Technical Summary

Problems solved by technology

However, SVM only gets better results on small sample training sets. When the training sample size is large, the storage and calculation of SVM will consume a lot of machine memory and computing time, and the selection of its kernel function is still an unsolved problem.

③Artificial neural network algorithm: The neural network has the ability of self-learning, without manual feature extraction, but the amount of calculation is large, and there is a risk of the training result falling into a local optimum, and the training result is unstable

Although this method has achieved a good classification effect, due to the resource requirements for image processing and the cost and volume limitations of low-resolution radar, it will not be able to guarantee real-time processing in actual engineering implementation

Images