Research On Cloud Detection Method Based On FY-3D And Landsat Satellite

With the development of remote sensing technology,satellite remote sensing has more and more application scenarios,such as change monitoring,atmospheric parameter retrieval,crop monitoring,etc.Remote sensing satellites can be blocked by clouds when observing the earth,resulting in missing information on the obtained remote sensing images,which affects the data quality and thus directly affects these remote sensing applications.Therefore,cloud detection is a key step in the pre-processing of satellite remote sensing images.Compared with ground objects,clouds have higher reflectance and lower brightness temperature in visible band,and are different in texture characteristics,so this can be used to achieve cloud detection.However,the complexity of the surface and the diversity of cloud types directly affect the accurate cloud detection in remote sensing images.To solve the above problems,two different cloud detection methods are proposed in this thesis:(1)Aiming at the difficulty in detecting and correctly labeling clouds on complex surface types,a cloud detection method by combining a multilayer perceptron with a radiative transfer model is proposed,taking the medium resolution spectral imager(MERSI II)on the FY-3D as an example.The datasets of reflectance values of various land surfaces are obtained through the radiative transfer model and used to train the multilayer perceptron.It is then used to detect clouds in FY-3D MERSI II images.(2)Aiming at the problem that cloud shadow objects are difficult to detect,a cloud detection method based on semantic segmentation network structure is proposed by using multispectral images such as Landsat 7 and Landsat 8.4-band or 5-band images are obtained by band selection and synthesis and used as the data set of this method.In addition,considering the particularity of the 9th band of Landsat 8,the effect of this band on the cloud detection is studied.In this thesis,two different methods are used for cloud detection of FY-3D and Landsat satellites.For data sources whose labels may be lacking,the cloud detection method of FY-3D can be adopted.For data sources with abundant images and delineated cloud shadows,the cloud detection method of Landsat can be adopted.After experiments,the accuracy of the cloud detection method based on FY-3D in surface near the equator can reach 91.74%,while the accuracy of the cloud detection method based on Landsat is more than 90%,which can provide some ideas and references for remote sensing satellite image cloud detection.