An Automatic Bad Band Pre-Removal Method for Hyperspectral Imagery

For most hyperspectral remote sensing applications, removing bad bands, such as low signal-to-noise ratio bands, is a required preprocessing step. Currently, the commonly used methods are by visual inspection and the sensor setting. The former is very time-consuming, and the latter is easy to either delete bands with tolerable quality or overlook some noisy bands. In this article, an inherent connection between bad band removal and target detection has been found. As we know, the result of target detection is the linear combination of all bands, and the weight coefficient of each band, i.e., the component of the filter vector, can be considered as the contribution of each band for the detection of targets of interest. Based on this fact, we develop an automatic bad band pre-removal method by using the matched filter (MF) weights of multiple targets within a scene, named the normalized MF weight method (NMFW). Experiments with three well-known hyperspectral data sets show that NMFW cannot only identify most reference bad bands but also extract other corrupted bands that are not labeled as reference bands.