An Adaptive Pixon Extraction Technique for Multispectral/Hyperspectral Image Classification

Hyperspectral imaging has gained significant interest in the past few decades, particularly in remote sensing applications. The considerably high spatial and spectral resolution of modern remotely sensed data often provides more accurate information about the scene. However, the complexity and dimensionality of such data, as well as potentially unwanted details embedded in the images, may act as a degrading factor in some applications such as classification. One solution to this issue is to utilize the spatial-spectral features to extract segments before the classification step. This preprocessing often leads to better classification results and a considerable decrease in computational time. In this letter, we propose a Pixon-based image segmentation method, which benefits from a preprocessing step based on partial differential equation to extract more homogenous segments. Moreover, a fast algorithm has been presented to adaptively tune the required parameters used in our Pixon-based schema. The acquired segments are then fed into the support vector machine classifier, and the final thematic class maps are produced. Experimental results on multi/hyperspectral data are encouraging to apply the proposed Pixons for classification.