GAL: A stepwise model for automated cloud shadow detection in HICO oceanic imagery utilizing guided filter, pixel assignment, and geometric linking.

摘要

Detection of cloud shadow pixels is an important step in image processing in several remote sensing ocean-color application domains, such as obtaining chlorophyll content. While shadow detection algorithms do exist, the vast majority are for over land which leaves few options for detection over water.;The detection of cloud shadow over water in HICO imagery is a unique problem. As its name implies, HICO (Hyperspectral Imager for the Coastal Ocean) imagery is produced for coastal and oceanic regions. Since land based algorithms remove water before processing, these approaches would not be applicable. The only currently published HICO shadow pixel detection algorithm produces good results for predominantly homogeneous regions. It also involves hand-tuning of the parameters, which is not suitable for automation.;GAL is a fully automated stepwise model that starts by using satellite imagery and navigational data. The next step is applying the guided filter algorithm proposed by He, Sun, and Tang to these images in order to filter and enhance the images before shadow detection. The third step classifies pixels into water, land, and clouds. The fourth step uses cloud shadow geometry to indicate possible shadow pixels. The final step is to reduce the amount of possible shadow pixels to the most probable shadow pixels.;This research combines the past techniques of cloud shadow geometry, edge detection, and thresholding, along with the new techniques of guided image filtering, in such a way that has never been done before. GAL works best with well-defined cloud shadows that contain a large contrast between water and shadow. Water type, coastal or deep ocean, does not affect GAL. Shadows with a large gradient may be under-detected. GAL can be applied to HICO data immediately, with the potential of being applied to all global high resolution ocean-color satellite imagery.