Pavement Crack Detection And Identification Classification Research And FPGA Implementation

The development of China's expressways has attracted worldwide attention.At the same time,the pressure of road maintenance is increasing day by day.This makes the traditional road surface damage manual visual inspection unable to meet the demand.The manual method has problems such as low efficiency and low safety.Therefore,there is an urgent need for a kind of efficient and highly intelligent road surface inspection device for detecting road surface damage detection.With the development of computer vision technology,many researchers have proposed intelligent detection methods for pavement damage and systematically developed them.According to the deficiencies of the existing methods and systems,the pavement cracks in pavement damage are taken as an example,and the cracks are divided into four types: horizontal,vertical,block and mesh crack.A real-time online pavement crack and position detection system is proposed.The system uses image processing and recognition technology to initially determine whether the captured road image is damaged.The Beidou navigation system is used for cracks location,and the crack picture and position information are transmitted in real time through the mobile communication system.Finally,combined with the current very popular artificial intelligence theory,the cracks are accurately identified and classified by convolutional neural networks.The whole system is divided into two parts: the upper computer and the lower computer.The lower computer is mainly realized by FPGA.The crack picture information and positioning information are transmitted to the upper computer through the 4G transmission module.The system upper computer software displays the crack picture,damage type,position and map label,etc.information.The main tasks completed in this paper include:(1)Pre-processing the acquired pavement image on the FPGA and improving the Sobel edge detection operator.According to the existing image preprocessing algorithm,image filtering and noise reduction are performed on the acquired crack image,and the improved Sobel edge detection algorithm is used to segment the image to obtain the binarized edge information of the image,Then the morphological expansion corrosion algorithm is further processed to obtain the ideal crack edge contour information.Finally,by calculating the projection features of the image,the cracks are initially detected and classified according to different thresholds.(2)Control the whole lower computer system through FPGA,obtain the Beidou positioning information,control the 4G transmission module to realize the image data transmission and drive the LCD to realize the local display of the detection result,and write the upper computer to realize the saving and receiving of the upload data of the lower computer.(3)In-depth study of image classification algorithms based on machine learning,comparing the performance of various classification algorithms,customizing the crack image dataset,modifying the structure of VGG convolutional neural network for the requirements of this paper,and finally convolving nerves The network further accurately identifies and classifies the received images,which improves the accuracy of identifying the classification.