Cloud Recognition for Four Bands Cameras of High Spatial Resolution Combined with the Regularized Least Squares Algorithm

This article presents a new cloud detection method combining regularized least squares algorithm and threshold method based on the characteristics of Chinese ZY-3 multispectral imagses. In the process of the new method, second extraction of clouds using a regularized least squares algorithm is done based on a first extraction of clouds using the threshold method, which overcomes confusion of clouds, roads, and buildings. Compared to existing cloud detection methods, the accuracy of the new method is subjectively visibly higher than the threshold method and the K-means clustering combined with threshold method, achieving the same level of accuracy as a support vector machine combined with the threshold method for higher efficiency. Using the new method on different scenes collected at different time, the overall accuracy of the proposed cloud detection method is higher than 97% and the Kappa coefficient is higher than 0.9. These results show that the new method can detect cloud effectively in the case of different underlying surfaces. It is anticipated that this method will be popularized and further applied to imagery from other satellite systems.