Research On Crack Detection And Segmentation Algorithm Of Bridge Pavement Under Complex Background

Transportation is the basic needs and prerequisite of economic development,and it is also the basis for survival and the symbol of development in modern society.The rapid development of transportation benefits from the high-speed construction of bridges and roads.However,the problem of "emphasizing construction but ignoring conservation" in the development of bridges and roads in China will bring huge hidden dangers to the safety of bridges and roads.It is well known that bridge cracks are not only an external manifestation of potential safety hazards in bridge structures,but may have unpredictable effects on traffic safety.Therefore,the key to implementing bridge and road maintenance work is to implement efficient and accurate inspection of bridge cracks.It is a key step for China to change from a country with many bridges to a country with strong bridge construction technology by adopting effective methods to detect and divide bridge pavement cracks and adopting technological innovation to realize bridge road maintenance and management.At present,many scholars at home and abroad have carried out extensive and in-depth researches on this issue,but there are still several realistic problems that have been ignored.The first problem is that there is no published dataset for bridge crack detection in complex background.However,using deep learning algorithms to extract and detect bridge cracks required a large number of bridge crack images as samples.The second problem is that the existing algorithms are all aimed at the researches of crack images with simple background,but the background of the actually collected images often contains many obstacles.The obstacles in the background will seriously affect the detection results of existing algorithms.The third problem is that the current bridge crack detection algorithms do not meet the real-time requirements.Because the existing detection algorithms ignore speed in order to improve accuracy.In response to these problems,this paper starts from the following three aspects:(1)Aiming at the problem that the lack of open bridge crack image dataset and bridge crack image collection work is too dangerous and heavy,this paper proposes a bridge crack image amplification algorithm based on deep convolutional generative adversarial network.The core idea of the algorithm is:First,collect a small number of bridge pavement crack images through drones.Then,the first image is amplified by three types of image processing methods.Finally,these images are second expanded by the proposed bridge crack image generation model to form the dataset.(2)Aiming at the problem that the existing bridge crack detection algorithms ignore the complexity of the background of crack images,this paper proposes a bridge crack image segmentation model that is suitable for complex background.The network model is based on image semantic segmentation.It consists of 74 convolutional layers,including a downsampling path consisting of DenseBlock and Transition Down,an upsampling path consisting of DenseBlock and Transition Up,and a Softmax.(3)Aiming at the problem of unsatisfactory real-time performance of existing bridge crack detection algorithms,this paper first proposes the application of bidirectional segmentation network to the crack detection of bridge pavement,which realizes the real-time semantic segmentation of bridge crack images.The bilateral segmentation network considers the both of location information and the receptive field.Its specific structure is as follows:Feature extraction is first performed through Spatial Path and Context Path.The output features of the two components are then fused by the feature fusion module.Finally,the crack segmentation result map is obtained by bilinear interpolation.