PolSAR (Polarimetric Synthetic Aperture Radar) imagery classification plays an essential role in monitoring remote sensing data. Such classification is a difficult task due to the speckle noise which appears in these kind of data. Therefore, there is a need to design new efficient methods to classify PolSAR images. In this work, a new approach to classify PolSAR data is proposed. The method relies on simultaneously filtering and classifying pixels within the image through embedding the problem into a diffusion-reaction partial differential equation system. The diffusion term smooths the patches within the image, and the reaction term tends to move the pixel PolSAR values towards the closest (in some sense) representative class. In particular, the method inherits the benefits of speckle filtering reduction by diffusion-like methods. An iterative schema is stated and, by properly selecting the algorithm control parameters, the user may force the classification to evolve according to her/his requirements to account for other image post-processing tasks (i.e. quantitative analysis to monitor deforestation, drought or urban areas growing). Results on real PolSAR data show the performance of the method, which is evaluated both visually and by means of the confusion matrix, showing an average classification rate 87.56%.
展开▼
