Optimal Band Selection of Multispectral Sensors for Wildfire Detection

The selection and definition of spectral bands are of great important in the design of multispectral sensors. In this paper, an optimal band selection strategy is proposed, considering the characteristics of the Earth surface, atmosphere, and targets as well as the influence of multispectral sensor system. Based on the spectral characteristics of the Earth surface and atmosphere, the spectral irradiances at sensor aperture from backgrounds and targets are calculated using an atmospheric radiative transfer model. Spectral response functions with different central wavelengths and bandwidths are used to calculate the intensity of the background and target radiations in specific bands. The contrasts of the targets to the backgrounds in these bands are analyzed to generate a two dimensional matrix. For a single target, the local maxima of the contrast matrix are selected as the optimal bands. When an optimal band is selected for multiple targets with different spectral characteristics, a scoring matrix is obtained for each target via grading the contrast matrix according to a specified threshold and the global maximum. The scoring matrices of all targets are integrated to generate an evaluation matrix where the local maxima correspond to the optimal central wavelengths and bandwidths. Optimal band selection of space-based multispectral sensors for wildfire detection is taken as an example to demonstrate the usefulness of the proposed strategy.