Research Of Target Recognition Of SAR Images Based On Update Learning Mechanism

Synthetic Aperture Radar(SAR)has a broad research and application prospects.It has been widely used in geological prospecting,topographic surveying and mapping,marine application,etc.In the military field,the SAR can be used in all-day and allweather global strategic reconnaissance and military monitoring.Researches about target recognition mainly rely on large amounts of labeled data.However,it costs a lot of resources to obtain these data with correct labels.What's more,training classifiers based on vast quantities of data costs a lot and requires challenge.Therefor,under the premise of guaranting performance of recognition,cost reduction becomes new research direction.Classifiers trained based on a small amout of training data cannot perform well.Thus,the lack of SAR images restricts the development of target recognition on SAR images.However,the number of SAR images will increase over time and these new images can be used directly to improve the performance of original classifiers instead of training new classifiers with original images.In shis way,training cost can reduce substantially.Define the method that uses new images in iterative training to improve original classifiers as update learning.This thesis aims at target recognition problem with update learning and researches about features extraction of SAR images based on convolutional neural networks.Main contents of this thesis have been shown as follow:1.Introduce the background of target recognition on SAR images,analyse the significance of this topic research and primary researches of target recognition on SAR images of scholars at home and abroad.2.Analyse of SAR image features and feature extraction methods,introduce maching learning theory and MSTAR database;3.Implement SAR image target recognition method based on support vector machine and introduce the applications of deep learning on target recognition,use convolutional neural network on SAR image target recognition and compare its result with other widely cited methods in literature;4.Implement update leaning mechanism with convolutional neural network,and use support vector machine as the auxiliary classifier.