Research On Apparent Disease Detection Method Of Concrete Bridge Based On Deep Learning

In recent years,China’s highway bridge engineering construction has entered a stage of rapid development,bridges,as an important part of transportation facilities,their detection and custody demand is getting higher and higher.However,traditional human inspection methods have many problems in terms of efficiency,accuracy and safety,and are difficult to meet today’s requirements for fast and accurate inspection.With the continuous promotion of the concept of "smart city" and the rapid development of artificial intelligence technology,the realization of comprehensive and efficient automatic inspection of infrastructure has become a future trend in the field of civil engineering.Based on deep learning and digital image processing technology,this paper proposes an algorithm for apparent disease detection of concrete bridges.The algorithm fully considers the needs of rapid detection and refined detection of bridges,realizes feature recognition through digital image processing technology,and uses deep learning technology to realize object detection and instance segmentation.The conclusions of this paper are as follows:(1)Research on crack feature recognition based on image processing.In view of the particularity of the bridge site environment and the characteristics of crack diseases,the crack image was preprocessed,and the image processing technology was used to identify the key characteristics such as the width,length and direction of the crack.(2)Apparent disease recognition of concrete bridges based on object detection algorithm.In this paper,a method for identifying apparent diseases of concrete bridges based on the improved YOLOv5 neural network model is proposed,which can effectively accurately locate and qualitatively detect three types of apparent diseases of concrete bridges: cracks,falling off and exposed ribs in the form of detection frames.In order to verify the effectiveness of the model,more than 2000 concrete apparent disease image datasets were produced,and the default parameters were obtained as initial parameters for training using the COCO dataset hyperparameter evolution.In the training process,the CA attention mechanism and the CBAM attention mechanism were added respectively,and it was found that the model with the CA attention mechanism could effectively capture smaller disease features,so as to achieve better detection results.(3)Apparent crack identification of concrete bridges based on instance segmentation algorithm.Based on the object detection technology,this paper uses the YOLOv7 deep convolutional neural network model to segment the apparent cracks of concrete bridges in a pixel-level manner,and trains by making and labeling more than 3000 fracture image datasets,and using the default parameters obtained from the evolution of hyperparameters of COCO datasets.After training the model,the automatic detection of crack diseases is completed,which can better meet the detection needs of apparent cracks in concrete bridges.(4)Software development and case testing.Based on the training results of the YOLOv5 deep convolutional neural network model in the identification of apparent diseases of concrete bridges,a system called "Bridge Apparent Disease Detection Software" was developed,which completely realized the detection process of "collection-transmission-storage-visualization".The system is tested on an actual bridge,and the results show that the system has excellent detection performance and has a wide range of application prospects.