Research On Bridge Crack Classification And Measurement Based On Convolutional Neural Networks

With the rapid development of national infrastructure,China’s highway bridges have been extensively built and popularized.During the operation and use of the bridge,there were bridge collapse accidents,resulting in a large number of economic losses and casualties.Therefore,the maintenance and supervision of bridges are highly valued.The traditional method of manual detection of diseases is inefficient and cannot meet the huge workload of daily bridge inspection.It is an inevitable trend to use automatic technology to automatically detect bridges.Deep learning excels in the field of machine vision and surpasses traditional image processing methods in many aspects such as target recognition and image detection.Unlike traditional methods,deep learning does not require manual extraction of features,but rather simulates the human visual system and extracts features based on the characteristics of the original image for target recognition and detection.In view of the complexity and diversity of the environment bridge located,this paper studies the results of bridge crack detection based on the UAV equipped with a high-definition camera and GPS positioning system.The main research contents of this paper are as follows:(1)Data acquisition and calibration: In this paper,UAV is used to collect bridge crack images.The method of using MATLAB to calibrate crack data is proposed to reduce the workload of marking.(2)A new CNN-based bridge cracks identification algorithm: A new network framework is introduced,and the identified images are integrated with morphology to remove the crack texture caused by special conditions,which promote the accuracy of bridge cracks.(3)A CNN-based bridge crack detection algorithm: The target is used to locate the crack.The image processing method is used to process the crack after the positioning,and the specific value of the bridge crack feature is obtained.The target detection algorithm is improved by means of reducing the influence of irrelevant information for the characteristics of the bridge crack so that the locating of cracks is faster and more accurate.(4)Package the proposed bridge identification and measurement algorithm.Crack detection and output results are automatically performed,eliminating the need for personnel operation,which makes the bridge crack detection more automated and intelligent.The experimental results show that the proposed bridge crack detection algorithm shows higher accuracy and less measurement error.