Spatio-temporal Suitability Evaluation And Fusion Algorithm Implementation Of Cloud Detection Algorithm Supported By Prior Remote Sensing Products

Clouds are ubiquitous in remote sensing images.The occlusion of the ground objects weakens the satellite sensor’s acquisition of the true radiation characteristics of the surface,thus affecting the validity of the spectral information of the features.Therefore,cloud detection has become an important pre-processing link for remote sensing images,which can effectively improve image utilization.Due to the mixed pixels of the underlying surface,various types of surface types,and the influence of remote sensing sensor channels,the traditional threshold method cloud detection is difficult to achieve high detection accuracy.The recently proposed Universal Dynamic Threshold Cloud Detection Algorithm(UDTCDA),the Cloud Detection Algorithm(LCCD)supported by surface type products,and the Automatic Generation Cloud Detection Algorithm(CDAG)based on hyperspectral data pixel library can effectively solve the above problems,but at the same time,each algorithm is also limited by its own principles,and there are problems that are difficult to avoid.In order to solve the problems of each algorithm and obtain a cloud detection algorithm with high stability,this paper proposes a priori based on the in-depth analysis of each algorithm.A fusion cloud detection algorithm supported by surface data.The main contents of this paper mainly include as followings:Spatio-temporal applicability evaluation of cloud detection algorithm supported by prior remote sensing data.Through the detection principle,prior knowledge and main steps of UDTCDA,CDAG and LCCD algorithms,a large number of cloud detection results corresponding to various table types are obtained,and the surface is divided into vegetation,water body,artificial land surface and high-bright bare land.Analyze the detection characteristics of each algorithm,UDTCDA algorithm has the phenomenon of high-brightness surface pixel misextraction,LCCD algorithm is prone to the problems of incorrect lifting of inland water body and poor detection of thin fragmentary clouds.At the same time,the CDAG algorithm underestimates the cloud information seriously and has poor recognition accuracy for thin fragmentary clouds.Fusion mode determination and cloud probability image generation.Through qualitative analysis and quantitative evaluation of the results of the algorithms,the best cloud detection algorithm for underlying surface of different surface types is determined by using the numerical variation law of cloud pixel accuracy rate and clear sky pixel error rate,and the cloud detection algorithm is combined by weight method to obtain cloud probability image.The results of cloud detection and pseudo-color synthetic images are visually interpreted and verified.The results show that the method can effectively improve the error detection problem of each algorithm,and has good recognition accuracy in various areas,especially for thin and fragmented clouds.Then,the experimental results were quantitatively compared and validated in 40 sample areas randomly selected from the whole world according to the surface types.The results show that the recognition accuracy of fused cloud detection results over different surface types is better,the user accuracy and producer accuracy are also better than the results of the algorithms,the phenomenon of misjudgement is significantly reduced,and the spatial consistency with visual interpretation results is increased.The most accurate surface type is vegetation and water body.