The Research Of Intelligence Algorithm For The Classification Of SAR Sea Ice Images

There is great discrepancy in the surface roughness degree between seawater and sea ice. In addition, both the Coefficient of dielectrical loss and the roughness degree on the surface of sea ice can be affected by the thickness, distribution and figuration time of sea ice. According to this principle, Synthetic Aperture Radar (SAR) images are extensively used for the determination of distribution and floating direction of brash ice. At present, in order to ensure the security of sea area, many developed countries using SAR images supply the forecast of sea ice for conveyance and project which are acted on the sea. Meanwhile, our country has had hardware foundation to some extent, but has no related software in the intelligent processing on ice information. This paper is based on the work of building a pattern recognition system for the sea ice images which are caught by SAR.In the aspect of imagery processing, the author propose a method of using GLCM(Gray-Level Co-occurrence Matrix) to calculate the texture characteristic of SAR images. And the SAR image is classified by using the feature vector that is composed of the Gray Level Co-occurrence matrix features and gray of pixels.In order to improve the classification accuracy of intelligence system, a combined basis (RBF-BP) neural network model is investigated for the detection of sea ice based on the texture analysis of SAR images. The model is trained and tested by the sample data set of the feature vectors. The classification results from a sea ice image show that the classification accuracy for sea ice is 93.70% by this model. As is shown from lots of experiments, the model based on the combined basis neural network is more effective than that based on an common one such as BP neural network and RBF neural network. Meanwhile, this model keeps the prime proximity and has a good performance of classification.