Research On Algorithm Of Improved Generative Adversarial Network For Deblurring Power Inspection Images

At present,with the rapid development of UAV technology,the use of UAVs to complete various inspection and reconnaissance tasks is widely used in various fields.The electric power inspection is a typical application of UAV inspection,but during the inspection,it is often affected by unfavorable factors such as bad weather,relative motion,and imaging equipment jitter,which makes the acquired inspection image blurred.The quality of the inspection image directly affects the timely grasp and accurate judgment of the power inspection information.How to restore high-quality clear images from the fuzzy power inspection images is of great significance and practical application value,and it is also the research focus of this article.Based on the Generative Adversarial Networks,this paper studies the key technologies in image deblurring,proposes an improved deblurring algorithml,and applies this model to the electric power inspection image deblurring experiment..The main work of this article can be summarized into the following two aspects:(1)Proposed RRDB-based deblurring algorithm for Generative Adversarial Networks.In the process of image deblurring and reconstruction,problems such as loss of details,edge distortion and excessive sharpening are prone to occur.First,on the basis of the Deblur GAN network,a multi-layer residual network and densely connected RRDB network units are substituted for the RB network units in the generator to improve the learning and generating ability of the generator.Second,considering the consistency of pixel content and spatial features before and after image deblurring,pixel error constraints and spatial feature constraints are used to ensure the convergence of the pixel content and spatial features of the image,respectively.Then,carry on the experimental verification to the improvement part separately.In the end,the model is compared with the benchmark model Debulr GAN and the deblurring model DMCNN on Go Pro and Hand-held Camera data sets respectively.Experimental results show that the deblurring model proposed in this paper improves the peak signal-to-noise ratio by 2.5% and 4.9%,and the structural similarity by 7.5% and 4.1%.(2)The model of this paper is used to de-blur simulation of power inspection image.First,the patrol image is preprocessed uniformly in size,format and bit depth.Secondly,according to the imaging link model during drone inspection,the point diffusion function is used to perform fuzzy degradation processing on the drone power inspection image to make the power inspection image data set.Thirdly,the RRDB-based Generative Adversarial Networks deblurring algorithm is applied to the deblurring process of power inspection images.Finally,we use the deblurring algorithm in this paper,Deblur GAN and DMCNN to carry out the simulation of electric patrol image deblurring experiments on the self-made data set.The experimental results show that the deblurring effect of the algorithm in this paper is better than the other two models.The patrol image after deblurring can better retain the detailed texture of high-voltage towers,transmission lines,high-voltage fittings and other devices.It can be seen that the deblurring method proposed in this paper can provide effective help for the assessment of potential safety hazards in the power field.