Cloud Detection for Multispectral Images of Space-based Earth Observational Sensors Using Imagery Simulation Model

In this work, a cloud detection scheme is proposed to process the multispectral images of space-based Earth observational sensors. With the assumption that the spectral and the spatial characteristics of the ground covers are invariant through a relatively long period, physically-based imagery simulation model is adopted to generate clear sky images for the specified sensor under the similar observational geometry with the same scene parameters. As the spectral bands of the selected sensor locate in atmospheric windows, the atmospheric condition are arbitrarily set as typical values to simulate the clear sky images. The structure similarity (SSIM) of the measured and the simulated images are calculated in the pixel by pixel manner to generate the SSIM image, in which the pixels with smaller SSIM values indicate the higher possibility of cloudy region. The cloud mask image can be obtained via selecting a suitable SSIM threshold for binary detection. A set of data measured by the Fengyun(FY)-4B geostationary satellite is used to demonstrate the usefulness of the proposed scheme. The images of the spectral bands NO.5 (1.58~1.64μm) and NO.7 (3.50~4.00μm) are selected as examples to implement cloud detection using monochromatic image alone as well as color ratio data. The results of the cloud detection validate the usefulness and the interpretability of the proposed scheme.