Research On Pavement Crack Detection Method Based On Deep Convolutional Neural Network

As one of the most important infrastructures,road is the foundation of transportation.With the increase of using years,road was affected by many natural conditions and human factors,which can cause cracks in the road surface.Cracks on the road surface indicate that the inside structure of road has been damaged.This will have a serious impact on traffic safety and social economy.Therefore,the pavement crack detection has become an important task.However,automatic crack detection is challenging due to the diversity of crack morphology and complex pavement background.To deal with the insufficient stability and low detection accuracy of existing crack detection algorithms in complex road environments,an end-to-end crack segmentation network based on convolutional neural network is proposed in this thesis to predict the crack pixels in pavement image.The feature maps extracted at each convolution stage are retained in order to detect more crack information,and a multi-dilated convolution module is used to extract multi-scale image information in parallel.In order to augment the key crack contextual information in the retained feature maps,the hierarchical feature augmentation module is proposed to introduce the contextual information of feature maps in deep layers to shallow layers,then the network utilizes a bottom-up path augmentation structure to introduce the detail informations of feature maps in shallow layers to deep layers.In our method,side networks are introduced in each convolution level to enhance the detection effect of the model through side supervision.The experimental results show that the proposed method can effectively predict the crack pixels in pavement image.In Deep Crack dataset,the F1-score and mean IOU reached0.8559 and 0.8675,respectively.The detection accuracy is improved compared with the other methods.A binary image can be obtained from detection network,then a method based on image processing is used to process the binary image to get the crack skeleton image.Based on the study and analysis of the crack skeleton,a crack classification algorithm was designed to obtain the crack types,and then different calculation methods were used to get the length and width information of crack.In order to display the research work,a simple crack detection system is implemented,which can receive pavement images and extract crack skeleton from crack areas,then the length,width and types of cracks are given.