Research On Detection Algorithm Of Pavement Crack Based On Deep Belief Network

With the increase of highway mileages,the task of road maintenance is more and more heavy.Rapid crack detection,efficient identification and timely maintenance is an effective way to improve road service quality and prolong road life.Crack is the earliest and the most common representation of disease on pavement.The adaptability of pavement crack detection algorithm based on digital image processing technology is limited,and it can not acquire good recognition effect when dealing with high resolution gray image data of pavement.In view of the problems of low efficiency and poor precision existing at home and abroad,this paper discusses the principle of road image data acquisition,analyzes the characteristics of the original image data,summarizes the characterization of the geometric distribution of the cracks,and studies the application of the depth learning model,and designs the pavement based on the depth confidence network.The crack identification algorithm solves the contradiction between high efficient collection and inefficient processing,and has completed the high reliability and high precision automatic recognition of the asphalt pavement cracks.The main contents are as follows:(1)through the analysis of the principle of road image data collection,the basic source of noise in the original pavement image is obtained from the hardware level,and the corresponding denoising method is designed.The geometric distribution of the crack disease is summarized,and the segmentation method of the two-dimensional pavement gray image is studied,and a reliable sample data set of fracture recognition is established.It provides a reliable basis for subsequent processing.(2)Analyzes the basic principle of DBN and its feature extraction and classification ability,studied the feature extraction ability from the angle of a single Restricted Boltzmann Machine(RBM),The RBM feature extraction capability is validated by using the pavement crack image data set.On this basis,the RBM feature extraction ability of different iterative times and different hidden layer nodes is validated.(3)From the angle of pattern recognition,aiming at the problem that the traditional pavement crack recognition algorithm has low recognition rate for the cracks with lower background contrast and incomplete expression form,using TensorFlow depth learning frame to improves the efficiency of this research algorithm to feature extraction of complex pavement.The effectiveness and superiority of the proposed algorithm are verified by the example and experiment.This paper puts forward a crack recognition method based on DBN,and uses TensorFlow,a deep learning framework to realize crack detection.In this paper,the algorithm is validated in the test of nearly hundred kilometers asphalt pavement.This scheme provides a new idea for pavement crack detection,which has great significance for highway maintenance management decision,pavement performance evaluation and analysis and prediction.