Research On Damage Assessment Technology Of Cylindrical Silencing Wall Based On Image Processing

Well launches of large aircraft are done in silos.The firing of the aircraft is accompanied by loud sound,hot tail-flame and high-pressure jet flow.These processes are brief,but can cause damage to the well facilities and walls.In order to reduce the impact of these factors,in the well wall inlay thousands of pieces of silencer board,composed of silencer wall,it can attenuate the sound energy to the weakest,but the silencer wall will also produce middle bulge,edge warping phenomenon because of the ablation of the tail flame,serious will fall off,causing serious consequences.Therefore,periodic damage assessment of the silencing wall is required.At present,the damage of the silencer wall in China is mainly assessed by the naked eye,but the depth of the shaft is tens of meters,so it is difficult to evaluate by the naked eye.To solve the above problems,combined with image algorithm,deep learning network and other related technologies,a set of image processing based cylindrical silencing wall damage assessment technology is proposed.In the future,binocular vision scanning can be combined with unmanned aerial vehicle platform to realize unmanned efficient detection of silencing wall.The algorithm of line feature detection and point feature detection is studied,a Hough-Harris based vertex detection algorithm of the silence-board was proposed to locate the vertex of the silence-board inaccurately.By establishing regression model equation,the least square method was used to calculate the minimum error of vertex coordinates.To locate the apex coordinates of the silencer board.Affected by the lighting environment in the shaft,the acquired data set of the silencer board will appear overall brighter or darker.Therefore,a local histogram enhancement method based on YCr Cb was proposed to improve the contrast between the damaged part and the background part of the silencer board,and solve the problem that the contrast is too large to produce a lot of noise.The common image spatial denoising methods are compared and analyzed,and bilateral filtering is adopted and improved on this basis.When the spatial kernel and value kernel are weighted,the pseudo-pixels in the value kernel are removed.The experiment shows that the improved bilateral filter can effectively improve the image quality of the sound cancellation board.Three network models with good adaptability to image segmentation are introduced and compared: U-Net,Mask R-CNN and Deeep Labv3+.Combined with the analysis of experimental data,Deeep Labv3+ network is selected for segmentation detection of the damaged area of the acoustic board.An improved Deep Labv3+ network model was proposed to improve the accuracy of network detection by integrating Neck layer and CBAM attention mechanism with Deep Labv3+ network model.Mish replaced Re LU activation function and added BN normalized layer to improve the robustness of the network model.Finally,through the geometric and shape features of the image,the damage parts detected by the acoustic board image were assessed.Through experimental verification,the research method proposed in this paper can effectively and accurately detect the damage area of the silencer wall,which provides theoretical support for the damage assessment of the silencer wall.