Research On Intelligent Detection Algorithm Of Concrete Structure Crack Based On Mobile Carrier

With the continuous strengthening of the national economy,the state’s investment in infrastructure such as roads,bridges and buildings is also increasing.The maintenance of infrastructure has become an important problem perplexing people.Affected by temperature changes,construction materials,improper construction and other factors,concrete facilities often produce cracks,which brings many dangers to people.In order to avoid the occurrence of danger,concrete crack detection has become an essential work.The traditional concrete crack detection technology will not only consume a lot of manpower and material resources,but also be easily affected by noise and other factors,so it can not accurately detect the cracks.Therefore,this paper adopts deep learning technology and image enhancement algorithm to realize the detection of concrete cracks.Firstly,the concrete crack detection technology based on deep learning needs a large number of labeled images to complete the training of neural network model,and the data set is expanded by using rotation,translation and other technologies.Four defogging methods,Retinex,histogram equalization,dark channel a priori defogging algorithm and optimized dark channel a priori defogging algorithm,are compared.The optimized dark channel a priori defogging algorithm is applied to the defogging of concrete crack image.Secondly,this paper uses Canny algorithm,region growth algorithm and u-net semantic segmentation algorithm to detect concrete crack images.Aiming at the problems of insufficient feature extraction of u-net network and poor effect of individual crack detection,a residual module is added to the four up sampling structures of u-net decoder to increase the network depth,and a CBAM attention module is added after the output feature map of the decoder to extract key information,Se module and residual module are added to the connecting part of encoder and decoder to increase the correlation of feature channel and enhance the feature extraction ability of the algorithm.The improved u-net network has better segmentation effect and higher detection accuracy for concrete cracks.Thirdly,YOLACT++algorithm is used to detect concrete cracks.From the experimental results,it can be seen that YOLACT++algorithm has high detection accuracy and the detection speed is faster than that of u-net algorithm.Finally,the trained YOLACT++model is implanted into the configured Jetson TX2 mobile platform for image and video detection.Jetson TX2 mobile platform has high performance and small volume,which can be easily installed in unmanned vehicles and other carriers.