Adaptive thematic object extraction from remote sensing image based on spectral matching

Thematic object extraction is of great significance to remote sensing applications. Its procedure is always complicated while the result is not so precise, especially for object with various subtypes. An adaptive extraction method based on spectral matching, considering both spectral and spatial information, is proposed to extract thematic object completely and accurately from remote sensing image. This method considers the essential spectral representation of thematic object through endmember selection, and then achieves complete extraction via “whole–local” scale spectral matching. Experiments on ETM+ images to extract water and bare land are employed, and the results demonstrate the effectiveness and universality of this method through comparison with maximum likelihood classification and support vector machine (SVM) classification.

► We propose an adaptive method to extract the thematic object comprehensively and precisely. ► The thematic object's endmember is selected to pursue intrinsic spectral matching. ► “Whole–local” scale transformation is utilized to focus on thematic object's local areas with various representations caused by different surroundings. ► Proceeding fine consideration within each local area according to its special characteristics would be more rational.