Road Pavement Crack Detection Research Using Remote Sensing Data Based On Deep Neural Network

With the rapid development of China’s road traffic,road traffic becomes particularly important for people’s travel.However,pavement disasters,especially road cracks,have been seriously affecting road safety,and the need for road maintenance and accurate positioning of diseased pavement has become more urgent.At present,in the actual detection of road cracks,manual detection is still the main method.There are many problems,such as high cost,long time-consuming,low efficiency,poor detection effect and so on.In recent years,with the development of high-resolution remote sensing technology,high-resolution,hyperspectral and high-temporal remote sensing image products gradually affect people’s lives.Using high-resolution remote sensing images,we can clearly see the road conditions.In target detection,with the development of deep learning,the accuracy of target detection is gradually improved,and deep learning avoids the drawbacks of artificial design features,and the efficiency of target detection is greatly improved Therefore,aiming at the problems existing in the process of road crack detection,this paper constructs a multi-perspective convolution network(Multi-PerNet)to extract the features of remote sensing images.Based on Faster R-CNN framework,the network is used to train the detection model of remote sensing images for vehicle targets.In the process of model training,the image is enhanced firstly,and the sample set is established.Then the image and the corresponding label of the training sample are input.Firstly,the feature map of the input image is extracted by Multi-PerNet3.Then,the area and aspect ratio of the vehicle area in the training sample are obtained by K-means clustering.And then based on the clustering results,the candidate window is generated by region generating network in Faster R-CNN and mapped to Feature map to get the features of the candidate window.Finally,the candidate window and its features are input into the classifier for training,and the detection model is obtained.The experimental results show that the detection accuracy of the model trained with Multi-PerNet3 as feature extraction network is 6.2% higher than that of ZF-net model,1.8% higher than that of PVANet model,and 26.0% smaller than that of PVANet model.The detection speed of a single image is 0.06s/sheet,which meets the needs of real-time detection.The proposed road crack detection method based on multi-view convolution neural network in remote sensing images can achieve fast and accurate road crack location detection,which provides a basis for judging road crack dangerous grade,and also provides a basis for road maintenance and road safety.