Research On Image Enhancement Technology Of UAV Cable Tunnel Inspection Based On Deep Learning

With the country’s active promotion of urban infrastructure construction,the construction of power pipe gallery has become an important project of urban infrastructure construction in China.How to efficiently inspect and maintain the pipe gallery has become a key issue.In order to eliminate hidden dangers as early as possible,improve reliability and reduce the intensity of manual inspection,online monitoring system and cable tunnel inspection unmanned aerial vehicle(UAV)began to assist manual inspection of cable tunnel.In the patrol inspection,the UAV is used for image acquisition and analysis.In present,the difficulty is that the UAV has serious distortion of the observation data due to the limitations of the shooting environment and equipment.In this paper,cable tunnel inspection is taken as the application scenario,and image enhancement requirements are taken as the goal.The main work and contributions include:(1)This paper analyzes the current situation of cable tunnel inspection,discusses the advantages and disadvantages of various inspection methods,analyzes the motion blur,high noise,atomization and other problems existing in the image collected by UAV during inspection,then puts forward specific image enhancement requirements.(2)In view of the atomization problem caused by the humidity of the tunnel environment during the inspection of the cable tunnel,this paper first analyzes the specific causes and principles of the atomization image.On this basis,the contrast limited adaptive histogram equalization(CLAHE)algorithm is selected to process the atomized image in cable tunnel inspection,and the dark channel prior and CLAHE algorithm are used to contrast the enhancement effect of the image.The experimental results show that the CLAHE method can effectively process the atomized image.(3)Aiming at the problem of motion blur caused by camera jitter in cable tunnel inspection,this paper first analyzes the causes of motion blur and establishes its degradation mathematical model.On this basis,the R-L algorithm and the depth learning method are used to deblur the inspection image.The results show that the method based on deep learning can effectively remove the motion blur.(4)Aiming at the high noise problem of low illumination image caused by the limitation of light source.First of all,BM3 D denoising method is used for image,its effect can not meet the needs of patrol inspection,and the noise is still too much.On this basis,an improved UNET algorithm is proposed to train the neural network through a large number of data sets based on real scenes.Finally,the experimental results show that the improved algorithm has better denoising effect than the traditional method,and has more advantages in highlighting,and can meet the needs of patrol inspection.