Study On Despeckling And Segmentation Algorithms Of SAR Image

Synthetic Aperture Radar(SAR)has the ability to image all-weather,all-weather and long-distance.Because of its good performance,SAR has been widely used in both military and civilian fields.The SAR original image needs to be interpreted to obtain the target of interest,and the SAR image despeckle and image segmentation are the key steps in the SAR image interpretation preprocessing stage.In the SAR system,speckle will obscure the details of SAR images and limit the interpretation of SAR images.In severe cases,the interpretation of SAR images will fail.SAR image segmentation is the basis of SAR image target recognition and feature extraction.Accurate segmentation of SAR images can improve the readability of SAR images,so SAR image segmentation can promote the application of SAR images in various fields.The main work of this article is as follows: In order to improve the performance of SAR image despeckle,a non-local mean(NLM)algorithm for image denoising was developed.Aiming at the problem of poor robustness and performance degradation caused by non-uniform samples when performing NLM algorithm coherent speckle suppression,a weighted non-local mean algorithm based on maximum ratio distance is proposed.This method measures the similarity between image sub-blocks by designing the maximum ratio distance based on the NLM algorithm,derivation proves that the similarity measurement method is more suitable for multiplicative noise environment in SAR images.At the same time,for images that are heavily polluted by noise,an adaptive binary weighting matrix is used to eliminate pixels that are too different from the central pixels in the neighborhood,thereby avoiding unreasonable similarity calculation results between neighborhood gray-scale matrices when performing similarity measurement.In addition,when selecting the filter parameters of the Gaussian kernel function,the method of selecting parameters based on the similarity between the pixels in the search domain and the central pixel is studied to avoid unreasonably setting the filter parameters.Finally,in order to more accurately illustrate the filtering effect for multiplicative noise,Equivalent Number of looks,Edge Protect Index,and Peak Signal to Noise Ratio were used to evaluate the suppression effect of the proposed method on the speckle noise.Experimental results prove the effectiveness of the proposed algorithm.In order to improve the performance of SAR image segmentation,the traditional fuzzy C-means clustering(FCM)algorithm is first studied.It is found that the standard FCM algorithm requires manual input to determine the clustering center and the number of clusters,but the accuracy of the number of clusters cannot be guaranteed manually.Aiming at the problem that the segmentation performance and the real-time performance caused by manual participation are difficult to meet the requirements,a peak in the smooth histogram is used to characterize a consistent region of the SAR image to determine the clustering center and the number of clusters.First calculate the smooth histogram of the SAR image to reduce the histogram glitch.Using the statistical information of the SAR image,the cluster center and the number of clusters can be obtained by detecting the peak value of the gray smooth histogram of the SAR image,which can reduce the number of iterations of the algorithm and improve the clustering accuracy.In order to further improve the clustering accuracy,an improved FCM algorithm using image redundancy information and inter-class dispersion information,that is,a fuzzy compact separation algorithm based on non-local maximum ratio distance is studied.In order to verify the experimental results of the proposed algorithm for SAR image segmentation,simulated images and real SAR images were used for segmentation processing.In addition,a comparative experiment of different clustering centers is set up,and the experimental results prove that the proposed algorithm has more accurate image segmentation results.