Research On Complex Stripe Noise Removal Method For Remote Sensing Images Based On Optimization Model

Remote sensing images can truly and vividly reflect the current distribution of ground objects,the interrelationships between ground objects or phenomena,and the evolution of the surface.However,the quality of remote sensing images is often affected by stripe noise.The current algorithms perform well in removing more regular stripe.But,when removing complex stripe,there may be a phenomenon of stripe residue or excessive smoothing.In remote sensing applications such as geological exploration or environmental monitoring where high quality images are required,the effectiveness of the algorithm is a primary consideration.However,when remote sensing images are used in scenarios such as natural disaster monitoring or military reconnaissance,it is vital that the processed image information is available quickly.In order to improve the effectiveness and efficiency of removing complex stripe in remote sensing images,the following two explorations have been conducted in this paper.In order to improve the effectiveness of removing complex stripe,this paper proposes a destriping model based on variable weight and group sparse regularization.Firstly,different region weights are set for different regions in the image,and different stripe estimation methods are used.Secondly,in order to better remove stripe noise with different intensities,variable weights are set for different stripe rows.And an adaptive weight estimation method is proposed to effectively prevent texture damage and stripe residuals caused by destriping to achieve better stripe estimation results.Meanwhile,the overall stripe noise is constrained by group sparsity.The alternating direction multiplier method(ADMM)is used to solve the proposed model in an alternating minimization manner.In order to improve the efficiency of complex stripe removal,this paper proposes a destriping optimization model based on a nonconvex regular constraint.The model uses a nonconvex normalized ε-penalty function constrains the sparsity of the stripe.Unlike most algorithms that use fixed smoothing weight,this paper uses adaptive vector weights to constrain the degree of smoothing of each row in the image.At the same time,the model also introduces regional weights to process extreme regions.When solving the model,this paper adopts the idea of domain decomposition to transform the proposed model into one-dimensional weighted L1 problems in different directions.Then,an iterative updating method is used to quickly converge to the optimal solution.Simulation experiments and real data experiments are carried out to compare these methods with several state-of-the-art destriping methods.The experimental results show that both methods have better image detail retention and better complex stripe removal effectiveness,and the nonconvex regular constrained destriping optimization model has excellent efficiency.