Road Cracks Based On Optimized Convolutional Neural Network Identification Method Research

With the active construction of our country’s highway system and the increase in the number of airports,the road surface will inevitably appear to have varying degrees of disease.Among these diseases,cracks are the most common and the most serious.At this stage,the main task of road crack detection is manual inspection,and the inspection process is labor-intensive and the inspection effect is poor.Therefore,achieving rapid and accurate control of key road information and gradually realizing automated inspection of road defects are important tasks that need to be resolved at this stage.With the continuous development of deep learning technology,the recognition accuracy gradually surpasses traditional recognition algorithms.Compared with traditional recognition algorithms,deep learning technology can automatically extract original image features for abstract expression.In summary,this article aims at the identification of pavement cracks,based on the image classification technology of deep learning,and proposes two practical methods that can automatically identify pavement cracks.The main contents of this article are:First,the current research status of road crack recognition at home and abroad is described,and the main principles and theories of deep learning convolutional neural networks are introduced in detail,that is,the entire framework,structural characteristics of convolutional neural networks,and the entire process of neural network training.Secondly,this paper proposes a new convolutional neural network model that combines Res Net50 and CBAM to improve the network’s ability to extract pavement crack features,thereby effectively identifying pavement crack images.Experiments show that the network model can effectively reduce network over-fitting,and achieved a recognition rate of 96.31% in the experiment.Compared with the traditional SVM method and other typical network models,the network has improved accuracy,recall rate and F1 score.The results confirm the effectiveness of the recognition method proposed in this paper.Finally,this paper makes targeted adjustments and optimizations to the structure and hyperparameters of the initial CNN network model,and proposes another method based on the optimized convolutional neural network model to identify pavement crack images,and puts it on the new public data set After verification,a good recognition rate was achieved.Then the two road crack recognition methods proposed in this paper are compared with the current mainstream model algorithms.The experimental results show that the model has achieved better training accuracy and its training efficiency has also been improved.Through the research of this paper,it is beneficial to further realize the automatic identification of pavement cracks and promote the intelligent development of engineering detection technology.