Crack Detection And Measurement Of Concrete Dams Based On Deep Learning And Binocular Vision

The total amount of dam reservoirs in China is the first in the world.As an integral part of water conservancy construction in China,dam safety detection has become a measure to avoid severe water calamities.The safety of the dam is subject to various factors,such as topographic factors,construction quality problems,environmental factors,natural factors,etc.Among them,the aging rate of the dam caused by natural factors is particularly serious,and consequently,the aging rate causes a lot of cracks in the dam.Thousands of miles of embankment is worthy of ant cave,even small cracks can not be ignored,the slight damage of cracks will lead to reservoir water release,dam shutdown for overhaul,and heavy damage will lead to dam breakdown,causing downstream flood disasters,posing a threat to the safety and property of those living downstream.Therefore,dam safety detection is very important.Dam safety monitoring has always been a challenge.Due to the complex internal structure and diverse situations of dams,manual detection has drawbacks such as high risk,difficult operation,time-consuming and laborious,imprecise detection data,and lack of intuition.Due to the advancement of computing,deep learning and binocular stereo vision technology,this has become a trending technology in the fields of industrial detection,industrial measurement,scientific research and so on.To address these issues,the integration of deep learning and binocular stereo vision technology is applied in this article to study an intelligent detection and measurement method for dam cracks.Its main research work is as follows:(1)Due to uneven material and illumination of dam caused by rainwater scouring,the obtained image distortion degrades.Therefore,image preprocessing is carried out for the obtained dam crack image at first.This paper discusses three algorithms for image enhancement:linear mapping,histogram equalization and logarithmic transformation,and two image processing method for filtering and denoising: spatial filtering method and frequency filtering method.The paper compares various denoising and enhancement algorithms by utilizing mean square error(MSE)and peak signal-to-noise ratio(PSNR)as evaluation metrics.Finally,a 3*3template median filtering algorithm is proposed for image denoising.Then,Otsu method is applied to segment the crack image,and the skeleton line and edge are extracted.(2)In terms of dam crack detection,based on the original YOLOv5,this crticle suggests an enhanced compact YOLOv5 s object recognition model.Substituting the feature extraction network of YOLOv5 s with the agile Ghost Net and fusing the efficient CA(Coordinate Attention)attention mechanism at the predictor side.First,the pre-processed dam crack image is labeled with Label Img to make the data set.Then train and evaluate the network architecture.Finally,the refined model is assessed using four metrics: Precision,Recall,m AP and FPS,and the outcomes are contrasted with various prevalent algorithms.The experimental findings demonstrate that the enhanced algorithm enhances the precision and pace of dam crack detection to some degree,while also fulfilling the real-time requirement of industrial detection.(3)In terms of crack measurement,this paper combines binocular stereo technology with dam crack measurement application.Binocular vision is used to calculate 3-D coordinates of edge(long)pixels at both ends,and Euro distance formula is used to calculate crack width(length).A series of measurement experiments are designed to verify the feasibility of dam crack measurement in this paper.The enhanced algorithm proposed in this paper can better detect dam cracks,and the proposed measurement method can effectively measure dam cracks,which can provide reference and practical value for future intelligent safety monitoring of dams.