Local incidence angle referenced classification on polarimetric synthetic aperture radar images in mountain glacier areas

The incidence angle is a crucial parameter for synthetic aperture radar (SAR). The local incidence angle differs greatly in mountainous areas, and it may lead to quite different backscatter properties for the same ground targets for different incidence angles. As a result, it has to be taken into consideration in image classification. This work demonstrates the importance of the local incidence angle and the necessity of considering this angle in SAR image classification in mountainous areas. A local incidence angle referenced method based on support vector machines is developed to perform classification. In this method, the datasets are divided into different zones according to local incidence angle, and the training and predicting are performed within these zones to eliminate the influence of the local incidence angles. The experiments are performed around the Dongkemadi Glacier in the central Qinghai-Tibetan plateau using two RADARSAT-2 polarimetric SAR images. This paper concentrates on the effect of the local incidence angles in SAR image classification. Compared to the method using only backscatter coefficients, the proposed method improves the overall classification accuracy by 6 to 8%. More important, it is verified that this method is more helpful for the ground cover types where the terrain changes sharply. (C) 2016 Society of Photo-Optical Instrumentation Engineers (SPIE)