Multispectral subpixel detection using least square unmixing

Many approaches have been developed for subpixel target detection in the past, and least square unmixing is one of the most widely used methods. It can detect subpixel target by estimating its abundance fraction resident in each pixel. This method has been successfully applied in hyperspectral remotely sensed images, but in order for this approach to be effective, the number of bands must be no less than that of signatures to be classified. This constraint is known as Band Number Constraint (BNC). Such inherent constraint is not an issue for hyperspectral images since they generally have hundreds of bands, which is more than the number of signatures resident within images. However, this may not be true for multispectral images where the number of signatures to be classified is greater than the number of bands. In this paper, instead of increasing the number of bands, we decrease the number of signatures by selecting part of materials applied to least square approach, and then those detection results are nonlinearly combined for endmember detection. It can be viewed as an extension of the least square approach and the experimental results showed it can successfully detect all endmembers.