Research On Intelligent Recognition Of Drainage Pipe Defects Based On Image Segmentation And Deep Learning

The rapid development of urbanization has increased the burden of urban drainage pipe network.In addition,pipe defects caused by aging of pipe network will not only cause the decline of drainage function,but also cause road collapse and environmental pollution.However,the internal environment of the drainage pipeline is complex,and the types of defects are various and difficult to identify.At present,the detection of pipeline defects mainly depends on manual detection,which not only consumes energy,but also is very easy to make mistakes under the subjective influence of the inspectors.Therefore,it is of great significance to realize the intelligent identification of drainage pipeline defects.With the rapid development of deep learning,image recognition technology has made great progress.In order to quickly detect the defects of the drainage pipeline and restore the drainage function of the drainage pipeline.This paper presents an intelligent defect detection method based on deep learning,which can not only extract the geometric features of defects,but also evaluate the impact of pipeline defects on flood risk based on hydrodynamic model.Firstly,the deeplabv3+ semantic segmentation neural network model is constructed,and the network is used to automatically identify,segment and label defects（obstacles,broken walls,staggered openings,tree roots and cracks）.Secondly,a feature extraction algorithm is established in MATLAB to quantitatively evaluate the severity of defects;Finally,based on SWMM hydrodynamic calculation,the influence of defects on the drainage capacity of the pipe network is analyzed.The main studies are as follows:（1）This paper analyzes and studies the current drainage pipeline detection technologies and methods at home and abroad.According to the detection requirements of drainage pipeline defects,based on the development status of deep learning neural networks,the semantic segmentation network that is most suitable for intelligent detection of drainage pipeline defects is compared and analyzed.（2）Based on Matlab,the data set label creation and sample data enhancement of drainage pipe defects are carried out to obtain high-quality data sets.The semantic segmentation model of Deep Labv3+ is constructed,and the five types of feature extraction skeletons,Res Net-18,Res Net-50,Mobilenet＿v2,Xception and Inception Resnet＿v2 used in the semantic segmentation model network are used for research and comparative analysis.Based on the actual case data of five types of defects such as wrong mouth,broken wall,obstacle,tree root and crack,the skeleton is screened according to the evaluation indicators such as PA,MPA,MIOU and FWIOU.The results show that the network models of the first four types of skeletons have the best effect on the actual segmentation of broken walls,all reaching more than 90%.The network model based on the Res Net-50 skeleton has the best effect on defect segmentation and annotation,with an accuracy rate of 89.75%.（3）Based on the pixel level segmentation results,the geometric feature extraction algorithm is constructed to calculate the area,length,center of circle and other geometric features of different defects.According to the drainage system defect evaluation procedures,the corresponding level of quantitative calculation and risk assessment are carried out.By comparing the actual defects with the prediction results of the semantic segmentation model,the model is modified and optimized.The results show that more than 87% of the predicted results are consistent with the actual defect grades,while the evaluation grades of other predicted defects are higher as a whole.（4）Carry out GIS terrain model construction and data calculation,build a SWMM pipe network model,and simulate them into the pipeline system according to the prediction levels of four types of defects: leakage,obstacles,falling off and misalignment,and compare and analyze no defects,actual defects and predictions.The impact of flooding in the three scenarios of the defect was analyzed,and the overflow situation in different return periods was analyzed.The results show that the pipeline defects increase the flood risk as a whole,including the number of overflow nodes and the overflow volume.The overflow volume of the predicted defect increases significantly faster than the actual defect as the return period increases.