Road Crack Detection And Recognition Based On Deep Learning

With the increase of China’s highway coverage area and the development of the transportation industry,pavement maintenance has become an important research topic in current traffic management.The prominent factors that affect the quality of the road surface are different types of road cracks.Detecting and repairing road cracks in time can effectively prevent traffic accidents and ensure the safety of pedestrians.At present,research on road cracks mostly staying at the stage of a digital image,susceptible to a variety of factors outside interference,poor detection effect,but also in the actual road repair cracks during different types of road crack repair methods are different.Based on the above problems,this paper combines the research of road cracks with deep learning methods,aiming at the characteristics of road cracks,and proposes an SSD(Single Shot Multibox Detector)road crack detection method based on multi-scale feature extraction and an improved convolution neural network road crack recognition classification method,the main research contents are as follows:(1)Since the current road crack detection methods are mostly based on image processing and segmentation methods,the method is complicated and cumbersome and is greatly affected by environmental factors such as noise.This study draws on the target detection algorithm in deep learning and applies the SSD algorithm with better detection effect to road crack detection.The SSD algorithm extracts multi-scale features of the image through multiple convolution operations and locates the road crack frame by generating proir box and matching strategies.The algorithm is realized by establishing a road crack image data set and using SSD300 and SSD512 models to detect cracks.The experimental results show that the mean average precision of SSD300 and SSD512 is more than 90%,and the detection speed of SSD300 can reach 40.18fps,which is an effective method for detecting road cracks.(2)In view of the problem that different types of crack images need to be repaired in practical applications,this study proposes an improved convolution neural network road crack recognition classification algorithm.The original LeNet-5 network is used as the basic framework for network construction,adding BN and Droupt method to prevent overfitting and gradient disappearance during training.The network changes the activation function of the LeNet-5 network to the ReLU function to improve the network convergence performance.At the same time,the network parameters are optimally adjusted to achieve a high-precision classification and recognition effect.The experimental results show that the improved convolutional neural network model has a recognition rate of 100%,98%,and 90%for lateral cracks,longitudinal cracks,and massive cracks.The average recognition rate is 96%,which is higher than the original LeNet-5 The model is improved by 6%,which has a good effect of identifying and classifying road cracks.