Automatic Classifier Construction And Algorithm For SAR Image

SAR(Synthetic Aperture Radar) is a kind of active microwave sensor. It plays an important role in imaging radar and has already been widely applied in scientific exploration,national security and economy, etc. At present, SAR image processing covers image speckle suppression, edge detection, feature extraction, image classification and so on. What’s more,image classification is an important research content of recognition and interpretation a ground scene based on SAR image data. In this paper, in-depth research on SAR image classification has been done. The main research contents are shown as follows:1. Owing to speckle noise, it is a challenge to study high-precision classification of SAR images. According to the imaging mechanism and the statistical properties of SAR images, a classification algorithm based on improved Ada Boost is proposed in this paper to improve the classification performance of SAR images. In this classification algorithm, the gray level co-occurrence matrix is used to extract the features and error correcting output code is introduced. Experimental results show that the proposed classification algorithm can obtain a better classification result and the accuracy is significantly improved.2. Considering poor stability and excessive training of SAR image classification algorithm based on Ada Boost, a classification algorithm of SAR images is presented using Ada Boost with stump functions as base classifiers. The bootstrap method is used to improve stability and accuracy and to prevent overtraining. A data set is randomly split into two subsets: one for training and the other one for validation. Subsampling and training validation steps are repeated to derive the binary classifier by the majority vote of the classifiers. Then,combining with error correcting output code, the multi-class classifier is constructed. Variable relevance applied to classification can be estimated by this method And, some measures are adopted to estimate prior probabilities of the variables for random combinations. In numerical experiments with SAR images, the proposed method performs extremely well and shows that it is superior to support vector machines, artificial neural networks, and other well-known classification methods.