Study on cloud removal method of remote sensing image based on penalty Weight Tensor RPCA

• 1、Chongqing University of Posts and Telecommunications,College of Computer Science and Technology,City Chong Qing , 400065

Abstract：With the development of aeronautics and astronautics and remote sensing technology, remote sensing technology has been widely used in various fields of society.However, due to the inherent defects of sensors or the interference of clouds, many remote sensing data cannot be used every year.Therefore, remote-sensing image cloud removal plays an important role in improving the utilization rate of remote sensing data and has attracted extensive attention.With the proposed tensor singular value decomposition (T-SVD), the tensor robust principal component analysis (PRCA) model has been widely used in image recovery and background extraction.In this paper, the tensor RPCA model is applied to the cloud removal of remote sensing images. Based on the characteristics of remote sensing images, the coefficient term of penalty weight is added. Firstly, cloud detection is carried out to obtain the initial cloud mask, through which different soft thresholds are set for the cloud-containing and cloud-free regions.In this paper, the improved penalty weight tensor RPCA model can be applied to more effective spatial information by constraining low-rank components.Moreover, waveband reconstruction is carried out before data is input into the model, which improves the computing efficiency and still ensures the accuracy.Experiments show that this method has good effect on both simulated and real clouds and can remove remote sensing image clouds under various complicated geographical conditions.

Keywords： Remote sensing image processing Cloud removal Penalty weight Tensor robust principal component analysis Low rank