Research On Cloud Removal Method Of Remote Sensing Image Based On Penalty Weight TRPCA Model

With the development of aerospace and remote sensing technology,the quality of remote sensing images is getting higher and higher,and there are more and more applications in meteorological observation,map surveying and mapping,military reconnaissance and other fields.However,because optical sensors are extremely susceptible to the influence of the atmosphere,more than half of the remote sensing data collected each year are data contaminated by clouds.These data cannot meet the needs of subsequent actual application scenarios,resulting in a huge waste of remote sensing resources.Therefore,remote sensing image cloud removal is of great significance in improving the utilization of remote sensing data.The current existing remote sensing image cloud removal methods differ in their use scenarios and reconstruction accuracy according to their use of spatial information,time information,and spectral information.Robust Principal Component Analysis（RPCA）is a kind of low-rank matrix reconstruction model,which is widely used in the field of image restoration.The RPCA model can effectively use the time-related information of remote sensing images.However,remote sensing images are different from ordinary images.The same area and different time of remote sensing images will be affected by the lighting conditions and produce greater differences.Only the introduction of time-related information is more prone to overfitting.On this basis,this thesis introduces spatial information,and applies the Tensor Robust Principal Component Analysis（TRPCA）model to remote sensing image cloud removal.The TRPCA model has a higher reconstruction accuracy when the cloud layer distribution is relatively scattered,but the effect is poor when the cloud layer is relatively concentrated and the spatial information is severely damaged.Aiming at the shortcomings of the TRPCA model,this thesis adds a penalty weight coefficient term to the original model.Before substituting the model into the model,first perform cloud detection to obtain the initial cloud mask.The cloud mask distinguishes the cloud coverage area in the target image and the uncovered cloud.Set different thresholds for these two areas.Experiments have verified that the improved penalty weight TRPCA model in this thesis has better reconstruction accuracy under such land cover types as mountains,towns,and lakes.In order to improve the computational efficiency of the tensor model,the preprocessing operation of band reconstruction is performed before the data is input to the model,and only the data with high correlation with the reconstruction area is retained,and the data of the same band at different times is reconstructed into a band tensor.In the input model,the calculation efficiency is improved while the accuracy is guaranteed.And on this basis,a block operation is added,which improves the reconstruction accuracy and operation efficiency under the composite land cover type.Finally,the model in this thesis is integrated into the remote sensing image automatic cloud removal system,which is convenient to apply to actual engineering scenarios.