Research On Cloud Detection Method For Remote Sensing Satellite Imagery

The rapid development of satellite technology has made remote sensing technology play a very important role in meteorological research and applications.During the process of remote sensing image acquisition,cloud interference causes spectral distortion of the original object,which affects the interpretation of remote sensing products and images,and has a bad influence on information extraction.In the remote sensing image processing,cloud pixels and cloudless pixels are effectively distinguished,which have positive effects on weather forecasting,ecological environment monitoring,and prevention of meteorological disasters.Therefore,cloud detection of remote sensing images becomes more and more important.This paper presents some cloud detection methods.The main contributions include the following:(1)A remote sensing image cloud detection method based on integrated threshold and random forest was proposed.The binarization is firstly performed using ten threshold methods of the first infrared band and visible channel of the image,and the binarized images are obtained by the voting strategy.Secondly,the binarized images of the two channels are combined to form an ensemble threshold image.Then the middle part of the ensemble threshold image and the upper and lower margins of NSMC's cloud detection result are used as the sample collection source data for the random forest.The training sample only depends on the source image data at one time,and the trained random forest model is applied to images of other times to obtain the final cloud detection results.This method performs well on FY-2G images and can effectively detect incorrect areas of the cloud detection products of the NSMC.The accuracy of the algorithm is evaluated by manually labeled ground truth using different methods and objective evaluation indices including Probability of Detection(POD),False Alarm Rate(FAR),Critical Success Index(CSI)and the average and standard deviation of all indices.The accuracy results show that the proposed method performs better than the other methods with less incorrect detection regions.(2)In this part,the cloud detection algorithm based on connected region matting is proposed.The binarization is firstly performed using ten threshold methods of the grayscale image.and the binarized images are obtained by the voting strategy.Secondly,this paper finds the connected region on the binarization,eliminates the relatively small connected region,and uses the center of gravity of the connected region as the seed point to generate a connected region ternary image;Then,theweight factor is introduced in the paper,and the local linear relationship is solved by moving least squares method to obtain the Laplacian matrix.Finally,the conjugate gradient method is used to solve the alpha value.The method is compared with a learning-based matting algorithm and a robust matting algorithm.The comparison results show that the proposed method performs well in the cloud detection results of FY-2G,Landsat8 and sentinal-2 images,and has better performance than the learning-based matting algorithm and robust matting algorithm.(3)The cloud detection image dataset is established,and the dataset plays an important role in the research of deep learning algorithms.In this paper,the image data of FY-2G is used as the original data,and the data set of Y-2G image semantic segmentation is established.The image consists of the NSMC cloud detection result edge part and the middle part of the ensamble threshold random forest cloud detection result.This paper prepares samples of four different band combinations based on the multi-band characteristics of the FY-2G image.That is,sample 1: IR1,IR2,and IR4bands;sample 2: IR1,IR3,and IR5 bands;Sample 3: The first three bands are preselected using the PCA for the IR1,IR2,IR3,and IR4 bands.Sample 4: The first three bands are preselected using the PCA for the IR1,IR2,IR3,IR4,and VIS bands.This method uses the Deeplab V3+ network to train different band combinations to compare the differences in cloud detection results.The experimental results show that the cloud detection result obtained by Deeplab V3+ has the highest correct rate of cloud detection in sample 3,reaching 95.1%-97.5%.The variance fluctuated between0.17% and 0.51%,indicating that the cloud detection results using the sample 3 in this method are relatively stable.