Skip to Main Content
The boosting algorithm is a powerful tool to combine a set of classifiers for improving classification performance. There are two issues to be considered for classification in polarimetric SAR image analysis: (1) how to choose the measurement with discriminability on various categories with different scattering characteristics; (2) how to combine the classifier sets with minimum loss of performance. For solving both the problems, the authors propose a supervised classification method. Firstly, the result of the optimization of polarimetric contrast enhancement (OPCE) is employed as the distance measurement. Then the measurement is translated to a confidence rate, which describes the decision of a "weak" classifier on a refined scale. These classifiers are generated and combined to a powerful classifier under the boosting framework. This proposed method takes advantages of both the specific characteristics of polarimetric SAR image and of the powerful machine learning algorithm. Using polarimetric SAR image, the authors demonstrate the effectiveness of the proposed method.