Research On Cloud Detection Algorithm For High Resolution Remote Sensing Image

Currently,high spatial resolution remote sensing data has become the main data source for the recognition and classification of ground objects,and its processing tends to be automated and intelligent by computers.As a normal natural phenomenon,cloud is difficult to avoid completely when acquiring images.The existence of the cloud itself and its shadow affects the imaging,obstructs the information of the ground features,and results in the loss of part of the imaging information,which results in image matching,image mosaic and ground feature recognition,etc.Produce great interference.High-resolution images have high spatial resolution and cloud images are more obvious.Therefore,the automatic detection of clouds in the application of high-resolution images has always been an important process of data processing.Scholars have done a lot of work on cloud detection in remote sensing images.However,most of the current algorithms formulate corresponding algorithms based on specific pop characteristics,which can only process multi-band(more than four)images,which is not applicable to low-band images.And it is difficult to effectively extract cloud transition areas and remove cloud-like features.In response to the above problems,this paper proposes three cloud detection algorithms with strong universality.(1)Cloud ontology detection algorithm based on multi-threshold jointThrough analysis,it is found that according to the difference of the histogram of clouded and non-clouded images,it is possible to distinguish whether there is cloud in the image,generate a gradient image and add it with the gray image according to a certain weight to enhance the edge gray value of the cloud area.And the combination of Otsu method and polynomial fitting threshold method can not only improve the detection effect of cloud edge,but also can effectively detect cloud area when the image contains a small amount of clouds.Finally,the spectral characteristics of the cloud in the red,green,and blue bands are used to further discriminate the cloud detection results and improve the cloud detection accuracy.(2)Cloud transition zone detection algorithm based on adaptive zone growthBy analyzing the gray characteristics of the cloud transition area,it is found that the gray value of the cloud transition area shows a slow decreasing trend from the inside to the outside.Therefore,this feature can be used to construct boundary discrimination criteria,and the area growth algorithm is used to grow the initial detection area of each cloud from the inside to the outside,so that the cloud area can be detected more completely,and the automatic detection of the cloud transition area can be realized.(3)Cloud-like object removal algorithm based on gradient meanBy analyzing the difference between cloud and cloud-like features,such as ice,snow,highgloss buildings,color characteristics and texture characteristics,it is found that the gray value of the transition area between cloud and cloud-like features is obviously different,and the gray value of the cloud transition area Gradually decrease while the cloud-like objects become smaller sharply.Therefore,the gray value of the transition area between cloud and cloud-like features is quite different.Based on this feature,a threshold criterion can be constructed to distinguish between clouds and cloud-like features and reduce the false detection rate of cloud detection.In this paper,the high-resolution remote sensing image data of Gaofen No.1(GF1),Ziyuan No.3(ZY3)and Tianhui No.1(TH1)are selected for experiments.The selected images involve a variety of surface types,including flat bare land and large Waters,snow and mountains,etc.,have different texture characteristics at the same time,including low-cloud images and cloudy images,with great lighting differences and scene changes.By comparing the algorithm in this paper with the classic Otsu method(Otsu)threshold algorithm,treestructured cloud detection algorithm and manual experience extraction results,the experimental results show that the algorithm in this paper can quickly and accurately detect cloud areas,and it can detect clouds.Transition zone and remove some cloud-like objects.