SAR image target recognition method based on wavelet threshold denoising combined with convolutional neural network

Abstract

A SAR image target recognition method based on wavelet threshold denoising combined with a convolutional neural network includes the following steps: a step 1, performing 2-layer 2D wavelet decomposition on a SAR target image; a step 2, using a Bayesian estimated threshold to quantize a high frequency coefficient after wavelet decomposition; a step 3, performing wavelet image reconstruction on a low-frequency coefficients after wavelet decomposition and the high-frequency coefficient after threshold quantization; and a step 4, using the convolutional neural network to automatically learn multi-layer features from the reconstructed SAR target image to characterize the image, and using a Softmax classifier to recognize a target type. The method can overcome the influence of speckle noise oncontour details of the image and reduce the dependence on artificial design, selection features and classifiers, and improve the performance of SAR image target recognition.

Description

technical field [0001] The invention relates to the fields of image processing, feature extraction, target recognition and the like, and in particular to the field of synthetic aperture radar image target recognition. Background technique [0002] Synthetic Aperture Radar (SAR) has the advantages of all-weather, all-day and long-distance detection, and can detect long-distance targets under any weather conditions and at night. At the same time, SAR has high resolution in range and azimuth, can obtain two-dimensional high-resolution radar images in the irradiation area, provide detailed ground detection data, and is widely used in military and civilian fields. Among them, one of the main applications of SAR in civilian and military applications is to discover and identify military targets, such as aircraft, tanks, airports and aprons, armored vehicles, missile launchers, various vehicles, ships, etc. [0003] SAR image target recognition refers to the technology of using SAR...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the problem that the existing SAR image target recognition method does not effectively remove image speckle noise and retain target contour information, and requires manual design, feature selection and classifier for target recognition, the present invention proposes a wavelet threshold denoising combined with convolutional neural network The SAR image target recognition method, which can overcome the influence of speckle noise on image contour details and reduce the dependence on manual design, feature selection and classifier, and improve the performance of SAR image target recognition

Images