Classified Linear Regression Based Landsat Image Cloud Removal Method

An approach for cloud removal based on linear regression after image classification is proposed in this article.First of all,the clouds in a remote sensing image and its referenced data to be processed are detected,from which two cloud masks are built.Then,an ISODATA classification is applied to the referenced image with the cloud mask.Next,the masked part of the contaminated image is classified with the existing clusters of the referenced data using the minimum distance method.Last,the digital numbers of the cloudy areas of the contaminated image are replaced with by the prediction value of the referenced data calculated by the linear relationships determined between clusters on the referenced image and the corresponding contaminates done according to the pixel location.This algorithm is programmed to automatically detect and remove the clouds areas in Landsat images.The accuracy of cloud detection and the prediction of original values of the cloud cover are evaluated.Results show that the proposed method is effective.