SAR target recognition method based on sparse complex image

Abstract

The invention discloses an SAR target recognition method based on a sparse complex image. The SAR target recognition method comprises the following steps: constructing a brand new sparse image data set; adopting a deep learning algorithm of a YOLO series, and training a neural network based on the sparse image data set constructed in the step S1; and S2, inputting a to-be-identified SAR image into the neural network trained in the step S2 to obtain the category probabilities of all targets in the SAR image and the corresponding positions thereof. According to the method, the sparse reconstruction algorithm and deep learning are combined, the system calculation complexity can be reduced, the image quality can be improved, more effective information is provided for a target recognition task, and rapid and accurate recognition of the target is achieved.

Description

technical field [0001] The invention belongs to the field of radar image processing and target recognition, in particular to a SAR target recognition method based on sparse complex images. Background technique [0002] Synthetic Aperture Radar (SAR) is a high-resolution microwave imaging system that realizes earth observation by actively transmitting and receiving electromagnetic waves. It is not affected by time and weather, and can obtain observation scenes all day and all day Therefore, SAR is widely used in reconnaissance, resource exploration, terrain mapping and disaster monitoring. [0003] With the development of science and technology, the amount of data has increased rapidly, and the method of manually interpreting SAR images is obviously no longer applicable. Since the first deep convolutional neural network (Convolution Neural Network, CNN for short) proposed by Krizhevesky et al. in 2012 achieved a top-5 error rate of 17.0%, CNN has been widely used for target ...

AI Technical Summary

Problems solved by technology

[0004] However, the quality of the SAR image reconstructed based on the matched filter algorithm is usually not high, and it contains serious clutter and side lobes, which is not conducive to the detection and recognition of targets.

Images