Gradable Cloud Detection in Four-Band Remote Sensing Images

Abstract

Cloud detection is one of the major techniques in remote sensing image processing. Many cloud detection algorithms have been developed recently. According to the type of remote sensing images that are used to detect cloud, they can be divided into two major categories: visible image-based methods and multispectral image-based methods. The first category mainly uses structure and texture characteristics for thick cloud detection, while the second category often uses the specific spectral bands for good results. In general, the existing methods above deal with cloud detection as a binary classification problem, cloud or non-cloud. However, as cloud has various forms and types, it is inappropriate to simply classify detection results into cloud or non-cloud. In this paper, we present a novel cloud detection method using orthogonal subspace projection (OSP), which can yield gradable cloud detection results. This detailed detection result not only conforms to the characteristics of cloud, but also brings more valuable guidance to subsequent interpretation of remote sensing images. Additionally, the proposed method only uses four universal bands including red, green, blue and near-infrared bands for detection, and has no requirement for special spectral bands, which make it more practical. Experiment results indicate that the proposed method has excellent results with high speed and accuracy.