Study On SAR Images Target Recognition Algorithm Based On Deep Learning

SAR has the characteristics of high resolution,so can be widely used in various fields of military and civil,especially in the field of military reconnaissance,SAR image recognition system has an irreplaceable role.In the process of SAR image target recognition, SAR image of speckle denoising,image compression and image target segmentation are important steps.and it is based on SAR target recognition as the background,this paper mainly studied the SAR image denoising,compression and target segmentation,etc.First,this paper describes the unique speckle multiplicative noise model of SAR image,because it is the basis of SAR image analysis.Lee filtering,Kuan filtering,wavelet transform filtering and Contourlet filtering are introduced,and analyzes their respective advantages and disadvantages.and then puts forward a brainstorming threshold optimization of NSCT adaptive image denoising method,compared with the traditional methods,this method can obtain very good filtering denoising effect.In the aspect of SAR image compression,a new SAR image compressed sensing algorithm based on self-adaptive of two-dimensional implicit sparse sampling is proposed.and compared the performance index “ the magnitude of consistency and the sampling data”with the traditional compressed sensing algorithm,the compressed sensing algorithm based on wavelet and the compressed sensing algorithm based on DCT transform.The proposed algorithm not only compression effect is obvious,but also to ensure the image quality of imaging.Finally in the segmentation of SAR images,first gives the definition of image segmentation,simply introduces the basic ideas of deep learning,deep belief network and network learning and other related knowledge of deep learning.On this basis,a new method of SAR image segmentation is proposed.Compared with the traditional deep learning neural network recognition algorithm ? SVM support vector machine recognition algorithm and multi-scale SOM neural network recognition algorithm.the recognition rate of the new algorithm is 97.3%,the performance is significantly higher than the other three methods,it is an effective method for SAR image target recognition.