Adaptive Threshold Cloud Detection Method based on Gaofen-1 Remote Sensing image

Abstract

Aiming at the problem of fewer bands of Gaofen-1 satellite sensors, which makes it difficult to realize accurate cloud localization under the restriction of limited band information, an adaptive cloud detection method based on four bands is proposed. The method combines the physical properties of clouds, adopts the improved Haze Optimized Transform (HOT) index for the radiometrically calibrated Gaofen-1 remote sensing images, selects three sub-positional values of the histogram for threshold segmentation, and introduces the maximum interclass variance method based on this method, which takes the maximum variance value as the threshold of the average reflectance index for image segmentation, and combines the visible band ratio and the cloud in the blue wave with the cloud in the blue wave. Based on this method, the maximum inter-class variance method is introduced, and the maximum variance value is used as the threshold value of the average reflectance index for image segmentation, and at the same time, combining with the visible band ratio and the optical properties of the cloud in the blue band, the final cloud mask is obtained by multiple corrections through the cascade of multi-class indicators. The experimental results show that the proposed method makes the average accuracy of cloud recognition as high as 94.52% and the average precision rate as high as 98.97%, which improves the traditional threshold cloud detection accuracy and at the same time, it can be used to prepare labeled data to provide objective validation sets for remote sensing imagery in the process of exploring the field of deep learning.