ABSTRACT
The cloud and cloud shadow are difficult to captureaccuratelyin optical imagery because of insufficient spectral information. In this paper, an automatic multiple features combined (MFC) method is proposed for cloud and cloud shadow detection in GF–1 WFVimagery which includes three visible and one near–infrared bands. The local optimization strategy with guided filtering, and the proposed object–based filter combining geometry and texture featuresareused in the proposedmethod to refineclouddetectionresults and exclude non–cloud bright objects. The experimental results indicate that MFC performs well under different conditions. Automatic Cloud and Cloud Shadow Detection
As a kind of common contaminant in optical remote sensing data, cloud cover impedes optical satellite from obtaining clear views of the earth surface. Cloud and its shadow make negative influences on the use of imagery such as vegetation fraction estimation and land cover monitoring. Extracting the accurate distribution of clouds and cloud shadows in imagery to ignore or eliminate is an essential and important pre–processing step for precise application.The wide field of view (WFV) imaging system onboard the Chinese GF–1 optical satellite has a 16–m resolution and four–day revisit cycle for large–scale Earth observation. Each WFV camera hasfour bandswhich are similar to Landsat ETM+ sensors in the first four bands spectral setting. The advantages of high temporal–spatial resolution and wide observation filed make GF–1WFVimagery widely used in environment, agriculture, land resources, and emergency disasters etc. Cloud detection in GF–1 WFV imagery is a challenging work due tounfixed radiometric calibration parameters and insufficient spectral information
In recent years, scholars have undertaken a great deal of research into cloud and cloud shadow detection for different types of remote sensing data such as AVHRR, MODIS and Landsat series imagery. ACCA , Fmask and MSScvm are typical algorithms designed for Landsat imagery which mainly rely on spectral features. Haze optimized transformation (HOT) which requires priori knowledge of the imagery is developed for the detection of haze/cloud and cloud shadow distributions in Landsat scenes . Methods based on machine learning include SVM and neural network are also applied in automatic cloud detection. Tmask and MTCD are multi–temporal methods used for cloud detection in multi–temporal imagery of the same area. Besides, cloud shadow detection is usually along with cloud detection. Cloud shadow location in imagery can be predicted by geometrical calculations after cloud detection.In this paper, an automatic multiple features combined method named MFCis proposed for cloud and cloud shadow detection in GF–1 WFV imagery. Experimental results suggestthatour method works well in different ground cover condition, and it can also capturethinclouds around cloud boundary accurately merely relying on four optical bands
The input data for the MFCalgorithm is TOA reflectance of all four bands in GF-1 WFV imagery. MFCfirstimplements threshold segmentation and local optimization strategy with guided filtering based on spectral features to generate preliminary cloud mask. Then, an object-based cloud filter combining geometry and texture features is constructed to improve the cloud detection results and produce the final cloud mask. Finally, cloud shadow mask can be acquired by means of cloud and cloud shadow match and correction. Fig. 1 is the brief flow of MFCalgorithm.