Research On Infrared Image Super-resolution Method Of Power Equipment Based On Compressed Sensing

With the advent of the Internet of Things era,building the power Internet of Things is an important measure to ensure the reliable,safe,economical and high-quality operation of the power system.The effective and extensive installation of various online monitoring sensors is a key link in the construction of the power Internet of Things.Among them,infrared monitoring technology has been more widely used.However,due to cost constraints,it is obviously difficult for large-scale installed online monitoring infrared sensors to achieve the accuracy of the current mainstream infrared imagers.In order to solve the problem of insufficient imaging clarity of low-precision infrared sensors,this paper studies the super-resolution technology.First,by classifying the existing super-resolution methods,the advantages and disadvantages of their methods are summarized.According to the characteristics of infrared sensors in the power industry,the acquisition frequency is low but the accuracy of image information is high,and the single-frame image reconstruction super-resolution method is selected for research.Since the compressed sensing theory is suitable for signal recovery and reconstruction,it is selected as the basic principle basis and improved to achieve the goal of improving the quality of images collected by low-precision sensors.Secondly,since the accurate establishment of the image degradation model is the main factor affecting the performance of the super-resolution algorithm.Therefore,based on the requirement of accurate and high-quality reconstruction of actual captured images,a compressive sensing blind super-resolution method is studied and proposed.According to the statistics of the internal information change rules in the process of infrared image degradation of power equipment,the prior knowledge of infrared image brightness channel extreme value distribution and infrared image gradient norm ratio prior information are introduced as the basic signal recovery model of compressed sensing.,in order to achieve an accurate estimation of the image degradation model.And an effective solution algorithm is designed for the proposed model.Improves the algorithm’s ability to process the actual captured images.Thirdly,due to the fact that in the actual operation,the blur kernel and the reconstructed image are estimated at the same time in the blind super-resolution algorithm,and it is easy to obtain sub-optimal reconstruction results.The algorithm of non-blind image reconstruction has been improved.A compressed sensing non-blind image reconstruction model is proposed.The two-prior quadratic estimation method is used to solve the penalty parameter of the regular term,and the adaptive intensity control of the constraint term is realized.In the solution process,the blurred image is initially reconstructed by deconvolution through Gaussian prior,and the significant edges of the reconstructed image are separated by threshold shrinkage to generate a label image.Then,according to the different semantics of pixels in the image,the intensity of the super-Laplace prior regularization term is controlled to improve the clarity of the reconstructed image and avoid the generation of artifact ringing.And an effective solution algorithm is designed for the proposed model.The image recognition experiment carried out confirms the practical application value of the method in this paper.This paper provides a complete solution for improving the quality of images collected by low-precision infrared sensors,and lays a foundation for further realizing unmanned monitoring and intelligent operation and maintenance of power equipment.