Research On Methods Of SAR Image Denoising, Segmentation And Target Detection

Synthetic aperture radar(SAR) has wide development prospect and practical value in civilian andmilitary fields. The key techniques of SAR image processing become domestic and foreign researchfocus. Therefore, studying the key techniques of SAR image processing has theoretical and practicalsignificance. On the basis of previous research results, researches on SAR image denoising,segmentation and target detection techniques have been done in this thesis, and are described asfollows:Firstly, research a SAR image denoising method based on complex contourlet transform andhidden Markov tree(HMT) model. Using multi-scale, multi-directional, and translational invariancefeatures of complex contourlet transform, this method is combined with HMT model and canaccurately describe the correlation of complex contourlet domain coefficients between different scales.According to the quantitative analysis of experimental results, the method has achieved gooddenoising effect.Then, a SAR image thresholding method based on2-D fuzzy Tsallis entropy and chaotic particleswarm optimization is proposed. The fuzzy set theory is introduced and the membership functions aredefined to consider the neighborhood gray information of pixels. At the same time, by using chaoticparticle swarm optimization, this method achieves optimization of the fuzzy parameters and reducesrunning time. The experimental results show that the segmentation results and processing speed areboth improved.And then, a SAR image ship detection method based on2-D minimum Tsallis cross entropy isimplemented. This method uses the thresholding method based on2-D minimum Tsallis cross entropyto extract region of interest which may contain ship targets from the background. At last,mathematical morphology is used to remove false targets in region of interest. From the experimentalresults, this method can extract ship targets from the complicated background, and effectively reducethe false alarm rate.Next, on the basis of the research of kernel fuzzy c-means(KFCM) algorithm and Chan-Vese(CV)model, a SAR image marine spill oil detection method based on KFCM algorithm and CV model isproposed. This method uses KFCM algorithm to map images to high dimensional feature space, thusthe data which belong to different classes can be divided more easily. To improve the convergencespeed of CV model, KFCM regional information is coupled into CV model. By the comparison in variance and normalized logarithmic likelihood ratio of images, the results show that this method issuperior.Finally, a SAR image river detection method based on contourlet domain modulus maximum andimproved mathematical morphology is given. According to the directional information and gradientinformation of contourlet transform directional subband, this method uses modulus maximum andimproved mathematical morphology edge detection operator to detect high frequency sub-imageedges and low frequency sub-image edges. The experimental results show that the river edge imagesare clear and accurate.