Radar target recognition method based on deep residual multi-scale one-dimensional convolutional neural network

Abstract

The invention proposes a radar target recognition method based on deep residual multi-scale one-dimensional convolutional neural network. Radar targets are difficult to recognize due to their translation sensitivity and attitude angle sensitivity. This method can automatically extract the HRRP invariant features of radar targets. In this method, two kinds of one-dimensional residual multi-scale blocks and two kinds of one-dimensional multi-scale down-sampling layers are designed by using the characteristics of convolution kernel and down-sampling operation, and the neural network model is built with the two as the core. On this basis, A new loss function is proposed to improve feature separability. This method can effectively extract the depth-invariant features of the target, has a high recognition accuracy rate, and has good robustness and generalization performance.

Description

A radar target recognition method based on deep residual multi-scale one-dimensional convolutional neural network technical field The invention belongs to radar target automatic identification technology, and the target HRRP feature extraction and analysis under the condition of design full-angle domain. Class recognition problem, provides a radar target recognition method based on a deep learning framework. Background technique The existing HRRP feature extraction methods for radar targets based on deep learning are basically based on self-encoding models. type and its variants, the autoencoder consists of two parts, an encoder and a decoder. The encoding process is the mapping of the input to the feature layer, The output of the encoder is the feature extracted by the self-encoding. When the number of neurons in the feature layer (hidden layer) is smaller than the input data dimension , encoding can be viewed as a dimensionality reduction operation, similar to Pri...

AI Technical Summary

Problems solved by technology

[0003] At present, the problems existing in radar target HRRP recognition are mainly divided into two aspects. First, HRRP has attitude angle sensitivity, and the recognition process needs to be divided into angle domains. Second, the generalization performance of the recognition model is not good.

Therefore, using self-encoding for recognition requires increasing the number of hidden layers and their neurons, but due to the limited number of training data, the model is prone to overfitting

Images