SAR Image Classification Based On Convolutional Neural Network

As the key of Synthetic Aperture Radar(SAR)image interpretation,SAR image classification is to classify the abundant objects in the image through relevant classification methods to obtain valuable information.With the development of artificial intelligence,deep learning increasingly shows the advantages of processing information,and has been widely used in the field of image classification.Compared with the traditional classification methods,deep learning can more fully exploit the potential information of the massive data to improve the accuracy of image classification.Convolutional neural network(CNN)can retain the local spatial correlation of the input image,so this paper mainly studies SAR image classification based on CNN.The research contents are as follows:(1)Pixel-level SAR image classification based on CNN utilizes the neighborhood information of the input image patches,but it does not highlight the influence of neighborhood pixels on the classification result of central pixels,which leads to the misclassification of central pixels under high noise conditions.This paper propose a novel SAR image classification method based on intensity similarity and convolution neural network.This method compares the intensity differences between neighborhood pixels and center pixels.The pixels with smaller intensity difference are given higher weighting factors,and pixels with larger intensity difference are given lower weighting factors.Experimental results show the proposed method can suppress the influence of speckle noise and improve the classification accuracy compared with traditional CNN,DBN and SVM.(2)The scene of real SAR image is complex,and the edge region has obvious structural characteristics-Considering only the influence of intensity differences between individual pixels on classification lacks the use of image structure information,which is disadvantageous for edge region classifications.This paper propose a novel SAR image classification method based on point feature similarity and convolution neural network.This method compares the image window distribution between neighborhood pixels and central pixels.The pixels with similar intensity and structure are given higher weighting factors,which highlights the influence of neighborhood pixels on the classification results of central pixels.Experimental results show that compared with the traditional CNN-based SAR image classification,the proposed method can make full use of neighborhood information,so as to suppress the influence of speckles more effectively and improve the classification accuracy of homogeneous regions.On the other hand,it can fully retain the structural information of the image patches by means of block matching to improve the positioning accuracy of the boundary.