Research On Methods Of Marine SAR Image Spill Oil And Ship Target Detection

Synthetic aperture radar (SAR) has wide development prospect and practical value in civilian andmilitary fields. In recent years, spill oil and ship detection and surveillance using SAR images hasreceived considerable attentions in the area of marine remote sensing. Therefore, studying the keytechniques of SAR image processing has theoretical and practical significance. On the basis ofprevious research results, researches on SAR image denoising, segmentation, feature extractionclassification and target detection techniques have been done in this thesis, and are described asfollows:Firstly, a segmentation method of marine spill oil images based on Tsallis entropy and improvedCV model with motion factor is researched. Multi-threshold selection algorithm based on Tsallisentropy is used to extract rough spill oil region and coarse contour which can provide local region andinitial contour for CV model, respectively, to reduce the scene complexity of CV model and itssensitivity to initial situation. In addition, the motion factor which can represent the local informationof the image is introduced into the CV model. It makes the model adapt to local characteristics of theimage to a certain extent and achieves more favorable segmental results than Tsallis entropy thresholdmethod, CV model method and improved CV model method.Secondly, a segmentation method of marine spill oil images based on Tsallis cross entropy and CVmodel with edge instruction function is proposed. Make a course segmentation based on Tsallis crossentropy considering the relationship between target and background. Meanwhile, make use of thechaotic particle swarm algorithm to optimize the recursive process which can improve the speed andaccuracy for searching the optimal value. Besides that, make use of image edge strength instead of theDirac items to imorove the partial differential equation of CV model. It makes the segmentationalgorithm better adapt to the SAR image and improves the convergence speed at the same time. Theresults show that the proposed method is better than Tsallis cross entropy multi-level thresholdmethod, CV model method and CV model with edge instruction function method.Thirdly, a spill oil image texture segmentation algorithm based on complex contourlet transform,Krawtchouk moment and FCM is realized. Complex contourlet transform has the characteristics ofmulti-scale, multi-directional and translational invariance features. Extract the mean and variance ofsub-bands in different directions on each decomposition level which can compose the texture featureof the image. Use the Krawtchouk moment invariant to describe the shape characteristic of the image.FCM clustering is used to realize the texture segmentation and the obtained texture segmentation results are better than those of wavelet transform and FCM method, the combination method of NSCT,Krawtchouk moment and FCM.Fourthly, a marine spill oil classification method based on Gabor transformation, Krawtchoukmoment and support vector machine is put forward. First of all, extract the characteristic parametersin different directions of the image after the Gabor filter. The Krawtchouk moment invariant is used todescribe the shape feature of the image. Make classification of the feature vector based on SVM.Through the quantitative analysis of the results, the proposed method can gain high accuracy constrastto the minimum classification model method and the maximum likelihood distribution model method.Fifthly，a SAR ship image denosing method based on dual tree complex wavelet transform andanisotropic diffusion is carried on. The image is decomposed by dual tree complex wavelet transform.Then improved TV diffusion is used in low frequency part. And the high frequency part contains mostof the noise and image detail information, so PM diffusion model is carried on this part. Finally theimage is synthesized. The experimental results show that compared with the wavelet thresholddenosing method, WPMTV, WSTV and CTND algorithm, the proposed method has better denosingeffect, can reserve more complete details and keep more clear texture.Finally, a marine ship SAR image target detection method based on KFCM and improved twoparameter CFAR is proposed. This method uses KFCM algorithm to map images to high dimensionalfeature space, thus the data belong to target and background can be divided more easily. Then detectthe ship target based on improved two parameter CFAR method. The experimental results show thatcompared with two parameter CFAR method and improved two parameter CFAR method, theproposed method has better denosing effect, can reserve more complete details and keep more cleartexture.