High Resolution SAR Image Classification Based On The Mixture Model And Level Set

Synthetic Aperture Radar(SAR) can work with all-weather, day to detect the surface of the earth and the characteristics of high spatial resolution SAR images. It has been used in military investigation, agriculture application, medicine detection, and so on. Image classification is an important problem in SAR image interpretation. It is also one of the key technologies in the property of SAR image automatic interpretation. However, the existence of the multiplicative coherent spot noise in SAR image, SAR image classification has become more difficult. The traditional image classification method for additive noise is no longer suitable for high resolution SAR image.According to the above problem, this paper carried out the comprehensive research on the SAR image classification method based on the level set and mixture model and put forward the suitable level set classification method for high resolution SAR image. The main work is following.(1) Due to the coherent noise in the SAR image, traditional distribution cannot accurately model the areas in the high resolution SAR images. This paper proposed a dictionary-training mixture model using K-SVD algorithm.The mixture model based on Log-normal model and Weibull model. In view of the traditional EM algorithm, the process of modeling mixture model of SAR image is complex and has many faults.Therefore, K-SVD algorithm is proposed to train the dictionary. This paper chooses the SKS parameter estimate method. The experimental results show that the dictionary-training mixture model can obtain more accurate fitting results on both the homogeneous features and the heterogeneous features.(2) Because the level set method based on the single model can’t get better classification result for high resolution SAR image, this paper proposes the mixture model for fitting the high resolution SAR image in the level set. The Chen- Vase(CV)model assumed that each target area of the image is homogeneous, but there are many heterogeneous features in the high resolution SAR image. Therefore, this paper proposed the level set method with mixture model for modeling of high resolution SAR image.(3) Because the edge information is very important in SAR image classification, this paper put forward the combination of the improved regional information and the edge information in the level set method for high resolution SAR images. In the regional energy function, the high resolution SAR image can be well statistical modeled using the mixture model instead of the model in Chen- Vase(CV) model. The boundary information in the high resolution SAR image is in combination with the region information in the level set method. Experiments on synthetic and real SAR images demonstrate that the proposed method can obtain good classification results.