Cloud And Shadow Detection From Landsat 8 Satellite Data Based On The Combined Using CDAG And Fmask Algorithm

Compared with the previous Landsat satellites data,the two main loads carried on the Landsat 8 satellite-OLI and TIRS have been significantly improved in terms of band setting and spatial resolution design,and they have been therefore more widely used.However,the existence of clouds and cloud shadows seriously constrains the appliance of this data.Clouds and cloud shadows severely decreasing the accuracy of both land surface variation monitoring and the quantitative parameter extraction in remote sensed images.Effective clouds and cloud shadows identification is critical to the application of Landsat 8 data.In order to improve the cloud detection accuracy of Landsat 8 data,the CDAG(A Cloud Detection Algorithm-Generating Method for Remote Sensing Data at Visible to Short Wave Infrared Wavelengths)algorithm and the Fmask(Function of Mask)algorithm were developed.The CDAG algorithm is based on hyperspectral data of AVIRIS,and which fully exploiting the spectral difference information between clouds and typical surfaces of Landsat 8 data from visible to short-wave infrared bands.The Fmask algorithm is based on the spectral difference and brightness temperature analysis of clouds and typical surfaces,which comprehensively utilizes the reflection information from visible to short-wave infrared bands and the thermal temperature information in the thermal infrared band of Landsat 8 data.Both the two algorithms have advantages and disadvantages in cloud detection.In order to make full use of two cloud and cloud shadow detection algorithms’ advantages,this paper proposes a cloud and cloud shadow detection algorithm combining and improving the two algorithms in view of the problems in the two algorithms.The main works include the following aspects:1)Cloud detection algorithm combined with CDAG and Fmask algorithm(COMCF)The CDAG algorithm fully exploits the spectral information of Landsat 8 satellite data from visible to short-wave infrared,therefore,based on the cloud detection algorithm automatically generated by CDAG algorithm from visible to short-wave infrared bands,the cloud detection algorithm proposed in this paper analyzes the relationship between simulated data and real Landsat 8 data,then randomly selects samples in Landsat 8 images to calculate cloud pixels correct rates to test each algorithm threshold,and the appropriate weight values are given according to the correct rate of cloud pixels to realize the normalized cloud probability of visible to short-wave infrared bands.Then,by analyzing the influence of different underlying surface on the reflectance of cloud detection results,the reasonable thresholds are set to realize the preliminary detection of clouds.The detection method of the thermal infrared band in Fmask algorithm is introduced to identify cloud and cloud shadow pixels.Analyzing the characteristics of thermal infrared band and calculating the cloud probabilities in images to realize the second step detection of clouds.Combine the two detection results to obtain high-precision cloud detection results.2)Improvement of cloud shadow detection algorithmThis paper improves the cloud shadow detection method based on the cloud detection results combined with CDAG and Fmask algorithm.The traditional cloud shadow detection algorithms use the radiation characteristics of thermal infrared band to determine cloud heights to further identify cloud shadows.Due to the large deviation in cloud heights estimation,the cloud shadow detection has low accuracy.To achieve high-precision cloud shadow detection,this paper proposes a cloud shadow detection algorithm combining with geometric and spectral percentage method(GSPM).The cloud detection results are divided into independent cloud blocks according to the eight connected regions,and treats each cloud block as a planar cloud model.Then,the height iteration is performed in two pixels from the assumed minimum cloud height to the highest cloud height.According to the relevant parameters of sun and sensors,the cloud shadows positions of cloud models are calculated by the geometric relationship of clouds and cloud shadows,and the projection ribbons are generated.It sets some reasonable thresholds in each projection region,and realizes the initial judgment of cloud shadow pixels position based on the percentage of the pixels satisfying the threshold.Then the image-based dynamic thresholds are used to obtain the final cloud shadow pixels.3)Accuracy verification of cloud and cloud shadow detectionThe Landsat 8 Biome Cloud Validation Mask released by the U.S.Geological Survey are selected for visual interpretation and quantitative analysis.The results show that for different cloud types in different regions,the proposed algorithm can detect clouds and cloud shadows more completely,and the overall accuracy is higher than 90%,and the algorithm can effectively reduce the influence of underlying surface environment.Comparing the results of proposed algorithm,CDAG algorithm and Fmask algorithm,it is found that the proposed algorithm has a higher accuracy and lower commission error and omission error.