Research On Defect Detection Method Of Drainage Pipeline Based On Gragh Convolutional Neural Network

With the development of society and the improvement of urbanization level,it is getting higher and higher to the requirements of urban infrastructure.Urban underground drainage pipe network is an important infrastructure for the discharge and treatment of storm water.The underground drainage network is prone to a series of defects,such as fracture,deformation and leakage,due to the perennial erosion of sewage and the accumulation of debris caused by the rising water level.Buried deep underground drainage pipe network,which is unfavorable for detection and artificial detection not only vulnerable to environmental restrictions,and subjective factors,it is difficult to form a unified standard,can’t meet the needs of modern development,the rise of artificial intelligence to promote the development of all walks of life,the use of artificial intelligence technology for drainage pipeline defect detection is the development trend of the future.Therefore,in view of the above problems,this paper conducts in-depth research on the drainage pipe defect detection method based on deep learning model,in order to improve the accuracy of drainage pipe defect detection,mainly including the following aspects:(1)Aiming at the problem of low quality of drainage pipeline image due to the limitation of shooting conditions and the need to model label correlations to improve the accuracy of defect detection,an improved label graph was proposed.Firstly,the adaptive symbiosis probability is constructed to adjust the attention degree of the label to itself and the attention degree of the symbiosis label,so as to adjust the aggregation degree of the label and its symbiosis label.Then,an adaptive symbiotic intensity alternative average allocation is constructed for each type of symbiotic label,so that the label with high symbiotic intensity has a high degree of aggregation,and the label with low symbiotic intensity has a low degree of aggregation.Experiment to research object of drainage pipeline data sets collected,theoretical analysis and experimental results show that the proposed method of m AP value reached 95.6%,compared with the existing drainage pipeline defect detection network,m AP value increased by 8%,the proposed method is effective to solve the correlations degree between ignoring label lead to the problem of low detection accuracy.(2)To solve the problem that the defect detection accuracy of drainage pipelines is low due to global label correlations in existing networks,a local label correlations GCN(LLCGCN)model was proposed to improve the global label correlations into local label correlations.Firstly,K-means clustering algorithm is used to cluster the samples and group the samples.Then,graph convolution network is performed for each group of samples to obtain node information with different local label correlations.Finally,according to the node information of the group to which the detection samples belong and the characteristics of the samples,the prediction is made to improve the accuracy of defect detection.Experiments to research object of drainage pipeline data sets collected,theoretical analysis and experimental results show that using the local model of drainage pipeline defect detection of label m AP of 95.5%,compared with the existing drainage pipeline defect detection network,m AP value increased by 7.9%,the proposed method is effective to solve the correlations between the use of global label lead to the problem of low detection accuracy.