A SAR Automatic Target Recognition Method Based on Capsule Network

Abstract

The invention discloses a capsule network-based SAR automatic target recognition method and belongs to the field of radar target recognition. The main process of this method: firstly crop the original SAR image, then perform simple convolution processing on the cropped image, then use convolution kernels with different expansion rates to extract multi-scale features, and then use adaptive features The refinement module enhances important features, and then fuses the enhanced multi-scale features through a pixel-by-pixel fusion strategy, and then inputs them to the network layer based on capsule units for more abstract feature learning and retains the spatial relationship between features, and finally the encoder network The output features are input to a decoder network consisting of four transposed convolutional layers for SAR target reconstruction to improve the learning ability of the encoder; the SAR target discrimination result is output in the last layer of the encoder network. Compared with the existing deep convolutional neural network algorithm, the present invention has higher precision.

Description

technical field [0001] The invention is applied to the field of automatic target recognition of synthetic aperture radar (SAR), and specifically relates to a method for automatic target recognition of SAR based on deep learning. Background technique [0002] Due to its all-weather, all-time and high-resolution characteristics, synthetic aperture radar has been widely used in geological exploration, environmental monitoring, military target detection and identification and other fields. Automatic Target Recognition (ATR) is one of the important applications of SAR image interpretation. [0003] In recent years, with the continuous development of deep learning technology, in the field of SAR automatic target recognition, various ATR algorithms based on deep learning models have been proposed and obtained better performance than traditional methods under some standard conditions (Standard operation condition, SOC). recognition performance. However, under extended operation co...

AI Technical Summary

Problems solved by technology

However, under extended operation conditions (Extended operation conditions, EOCs), such as in noisy environments, various complex noises will seriously affect image feature extraction; in most real SAR scenes, it is difficult to collect a large number of training samples, and In the case of insufficient training samples, the data-driven deep model is easy to overfit the classification model; in combat scenes, partial target occlusion and camouflage are very common, and extracting robust discriminant features from partial target occlusion is a valuable tool. Challenging

Images