Program Study On The Integration Detection Of Pipelines Defect

Defect detection and interpretation of underground drainage pipelines is the basis for efficient management of underground pipelines,and it is also a key issue to achieve ’smart city’.With the rapid development of urbanization and the deepening of urbanization,underground drainage pipelines are becoming more and more intensive.Various accidents caused by defects such as pipeline blockage,corrosion,rupture and even collapse have seriously affected people ’s social life.Therefore,regular inspection of the defects of underground drainage pipelines can effectively prevent accidents caused by pipelines.At present,in the application field of drainage pipeline engineering,the mainstream detection method of underground drainage pipeline is CCTV detection system.The underground drainage pipeline robot is used to shoot the inside of the pipeline,and then the video image is interpreted by professional inspectors to determine the type and location of the defect.However,manual identification of drainage pipeline defects is time-consuming and laborious,and the subjective error is large,so it is impossible to identify the defects in time and accurately.In order to improve the detection efficiency,this paper studies the integrated scheme of defect detection based on deep learning.This paper summarizes the current deep learning algorithm and pipeline defect recognition technology,and analyzes the significance and feasibility of this study.We use video key frame extraction and convolutional neural network algorithm to intelligently detect video images of underground drainage pipelines,and obtain the accurate location of pipeline defects based on image text recognition.An integrated scheme of intelligent detection of underground drainage pipeline defects is proposed.In this scheme,the drainage pipeline video defect detection process is divided into three stages : video preprocessing stage before detection,drainage pipeline defect detection model construction stage and detection result optimization output stage.(1)The video preprocessing stage before detection is divided into key frame extraction and positive and abnormal frame classification.Firstly,the inter-frame difference algorithm is used to process the pipeline video frame extraction and screen out the key frames.Then,the Efficient Net network is used to classify the key frames into positive and abnormal frames,extract the interest detection frames,and reduce the amount of data to be detected.(2)Defect detection model construction stage : select YOLOv3 as the main framework of the network,replace the original backbone network with a lightweight and efficient Efficient Net structure;an efficient pipeline defect detection model is established by using transfer learning strategy and training with self-built data set Pipe-DATA.(3)The optimization stage of defect detection results : the optimization strategy of two outputs is used to prevent the missed detection of defects when detecting the frame output detection results.Text recognition is performed on the detected defect frame image,and the defect detection form is automatically generated by deduplication optimization combined with defect detection results and text recognition results.Finally,we designed and developed the FEDDR(Frame Extracting Detection Duplicate Removal)system based on the integrated scheme of defect detection of drainage pipeline network.The system was used to process the video data of nearly27170.9 meters of pipeline in practical engineering.A total of 656 defects were detected.Compared with the manual discrimination results,the accuracy rate was94.3 % and the recall rate was 98.7 %.The whole process was integrated,which greatly reduced the labor cost and improved the detection efficiency of drainage pipeline defects.It has engineering practicability.