Application Of Machine Learning To Image Classification Of Defects In Underground Drainage Pipes

With the development of modern cities,the scale of cities continues to expand,and the early drainage pipes buried deep in the city have been overwhelmed.At present,in the field of engineering applications,the defects of drainage pipelines are mainly recognized by human eyes,which is time-consuming and laborious,and subjective errors are large.In order to help engineers quickly,effectively and adaptively classify drainage pipe defects,this paper introduces and constructs two classification models based on supervised learning according to the classification complexity of machine learning,which are traditional machine learning with SVM-KNN hybrid model and deep learning classification model based on VGG16 backbone.(1)The SVM-KNN hybrid model is aimed at the problem of low classification accuracy near the optimal hyperplane in the traditional Support Vector Machines(SVM)algorithm.The K-Nearest Neighbor(KNN)algorithm is introduced to optimize The data in the area near the hyperplane is classified to improve the classification accuracy of the support vector machine.(2)In view of the long training time and easy overfitting of the VGG16 convolutional neural network model,this paper uses the stochastic gradient descent algorithm,the Dropout method,the Batch Normalization method,the Early Stopping method and the method of transforming the activation function to optimize the model and improve the model's performance.Efficiency and accuracy,and reduce model overfitting.The above two types of methods are applied to the defect image data of underground drainage pipes,and SVM-KNN and the optimized VGG16 model are evaluated according to multiple evaluation indicators such as accuracy,F1-score and AUC(Area Under Curve).The experimental results prove that SVM-KNN model and optimized VGG16 convolutional neural network model can improve the classification accuracy of SVM and traditional VGG16 model,and can be applied to pipeline network inspection projects.Among them,the classification accuracy rate of the SVM-KNN model is 83.7%,which is 6.8% higher than that of the SVM model;the classification accuracy rate of the optimized VGG16 model is 92.3%,which is 1.8% higher than that of the traditional VGG16 model.Finally,it is determined that the optimized VGG16 convolutional neural network model is most suitable for the classification of defects in underground drainage pipes.Its classification accuracy is significantly higher than the recognition accuracy of human eyes.It can accurately and quickly distinguish defective images,and has certain scalability and promotion value.