I. Introduction
The mixed-pixel problem is a ubiquitous phenomenon in remotely sensed imagery, which means that one sampled pixel consists of multiple land-cover classes on the Earth’s surface [1], due to the limitation of the frequency of the instantaneous field of view (IFOV) of the sensor. This phenomenon has seriously impeded the development of land-use/land-cover (LULC) information extraction [2], since image classification is implemented pixel-by-pixel, and one pixel should be assigned only one label of a land-cover class. When encountering a mixed pixel, the one dominant land-cover class within the mixed pixel can conquer the whole mixed pixel and efface the other possible content. This means that the mixed pixel can be mistakenly identified as a pure pixel, which is counter-intuitive and will compromise the classification result.