Subpixel Land Covers Detection and Classification for Hyperspectral Imagery

Hyperspectral imaging has recently received considerable interest in land-cover classification. With the improvement of spectral resolution, hyperspectral images can be used to detect and classify subtle land cover types which cannot be resolved by multispectral data. Unfortunately, most of techniques for land cover classification are developed based on pattern classification rather than target classification. The chief difference between these two is that patter classification. is performed by classifying all image pixel vectors into different types of pattern classes, including image background, whereas target classification is conducted based on targets of interest regardless of what image background is. This paper presents hyperspectral land-cover classification techniques based on targets of interest. Experiments are conducted using DAIS data acquired by GER for applications in agriculture and environmental monitoring.