Research On Despeckling,Segmentation Algorithm And Related Technology Of Synthetic Aperture Radar Image

Synthetic Aperture Radar(SAR) has became a hot topic in modern remote sensing research because of its capability of providing all weather,full-time information about the earth.As a coherent imaging system,Synthetic Aperture Radar(SAR) images are contaminated by multiplicative noise that will influence imaging quality and go against identifying and classifying images because of its backscatters.Therefore the disposal of SAR image has became one important subject in the technique of image processing.(1) Analyzed the traditional filter methods in different window sizes on suppressing the speckle noise on SAR Image.Appling Lee filter,enhanced Lee filter,Frost filter,enhanced Frost filter,Gamma MAP filter,Kuan filter,Sigma filter,Wiener filter to deal with SAR Image.The result shows that with enlargement of window size,the standard deviation tended to increase.The changes speed of standard deviation were more rapid between 5×5 size and 7×7size,but traditional method can' t keep edge information and detail of SAR Image.(2) In order to overcome deficiency of traditional filter method,this paper puts forward an improvedàtrous wavelet algorithm based on undecimated discrete wavelet transform—SAR Image compression in wavelet domain method.This method smoothed processing high-pass sub band of wavelet coefficient of logarithmically transformed SAR images,then performed mean filtering with low-frequency signal.The method integrated used advantages of wavelet transformation and smoothing processing based on median theory in speckle suppression field,utilizes not only the correlation in wavelet transform domain,but also the protection of the image edge that base on median smoothing processing.This paper analyzed influence on filtering from natural visual feature of images,different window size,different decomposition algorithm,different wavelet base and different scale,then quantitative evaluation on filtering result by edge keeping index and smoothness index.The results show that method in this paper is more satisfactory under Coif wavelet base. Compared with other traditional methods,SAR Image compression in wavelet domain method is more simple and brief in algorithm,easy to implement in engineering,effectively eliminated fuzzy of marginal noise,realized edge-preserving and beneficial to inhibited SAR Image Despeckling.(3) Take image segmentation algorithm as an example,first mainly summarized some typical and classical segmentation algorithms,then give an improved SAR image segmentation algorithm based on genetic algorithm.It is an effective and parallesolution for globe optimization with the features of robustness parallelism and self-adaptation,especially for large volume image data such as SAR image,we can easy search the threshold value and it is advantage for the subsequent treating.With the increase of the solution space scale, genetic algorithm will be more effective.At present,methods of Image Segmentation mainly is based on gray and texture.Because of SAR image's inherent multiplicative Noise,we can't use gray threshold method.Method in this paper segmenting recovering SAR image, and determine optimal threshold for the image.First encoded SAR image into 8 bit-binary codes to express gray levels of image,then keep the same probability to generating initial population and calculating each individual's fitness,in order to operating of selection, crossover and mutation to produce new group.Take largest fitness individual as optimum result and reverse its coding,that's optimal segmenting threshold we need.Using Morphological Filter Disposal Methods to get segmentation graph.This paper discussed influence from visual effect,filtering,iteration number,selection of parameters on segmentation result,then make a quantitative evaluation by region consistency,region contrast and shape measurement,the results show the genetic algorithm can achieves the right image segmentation after setting suitable genetic operator.In order to prove effectiveness of algorithm in this paper,we making a comparison between the different segmentation method's effect on same SAR image.The experiment shows that method in this paper is superior to other methods no matter accuracy of segmentation or anti-noise.