Research On Crack Identification Technology Of Concrete Bridge Based On CNN Deep Learning

In 2019,the Central Committee of the Communist Party of China issued the outline for the construction of a powerful transportation country,which proposed to "promote the transformation of transportation development from pursuing speed and scale to paying more attention to quality and efficiency,build a safe,convenient,efficient,green and economic modern comprehensive transportation system,and provide strong support for building a modern socialist power in an all-round way and realizing the Chinese dream of the great rejuvenation of the Chinese nation." The transportation power strategy proposed under the background of the times,the bridge that plays a link role in the modern three-dimensional and comprehensive transportation system will usher in a new round of construction and development period.At this stage,for China’s huge bridge projects,"construction and maintenance" is the basic policy that should be implemented.However,many concrete bridges often produce a variety of diseases due to their own material aging,environmental factors(temperature and humidity change,earthquake action,flood impact),human factors(load action,Vehicle overload)and other factors.Such as honeycomb,pitted surface,damage,crack,exposed reinforcement,water seepage and other diseases.Among them,crack disease is the most common and urgent disease to be solved,which has a significant impact on the healthy service performance of the bridge."Detection" is the basis of "maintenance",How to detect bridge cracks quickly,efficiently and intelligently,so as to achieve accurate positioning and identification,so as to provide preconditions for the timely repair of bridges in the later stage.Ensuring the healthy traffic capacity of bridges and improving the safety and durability of bridge structures are very important for the steady and sustainable development of China’s transportation industry.Combined with the Jilin provincial transportation innovation and development support project "Research on key technologies and application of highway bridge robot inspection",this paper uses the bridge intelligent inspection technology,that is,UAV carries high-definition industrial camera to scan the image of the bridge,and takes the image scanning results as the data set,The research on concrete bridge crack recognition technology is based on CNN(revolutionary neural networks).CNN is a nonlinear representation and highly parallel depth learning method,which can realize hierarchical automatic extraction of image features.It is of great significance to realize the intelligent detection of bridge cracks.The specific research contents are as follows:(1)The first is the acquisition and calibration of data sets.The image set selected in this paper is all the pictures taken by the intelligent inspection of UAV bridge on Changchun rail transit line 3,and the images used are labeled by labelme labeling software.Due to the diversity and complexity of its environment,the bridge will show similar crack characteristics in different degrees in the apparent detection part,such as template mark,dark black water stain,black brown line handwriting,black linear spider silk,etc.,which need to be distinguished from the real crack.Combined with yolov3 target detection model,bridge cracks and pseudo cracks are distinguished.(2)Based on the distinction between real cracks and pseudo cracks,this paper proposes an improved deep labv3 + semantic segmentation model,which can recognize the crack image at the pixel level,not only complete the classification and location of cracks,but also generate the crack area with detailed boundary to realize crack segmentation.On the same bridge crack data set,the algorithm proposed in this paper is compared with the traditional crack image processing method--threshold segmentation method,In order to verify the effectiveness of the improved algorithm in this paper,Further,the experimental results are compared with the current representative image segmentation models such as maskr CNN and deeplabv3 +,which show that the crack segmentation accuracy of each algorithm.(3)Complete the non-contact measurement of crack length and width.Open CV is used to measure the length and width of the split crack image in the pixel sense of the crack by combining with the maximum connected area method,and the scale coefficient is determined by combining with the corresponding physical information during camera calibration,which corresponds to the actual physical size of the crack and realizes the automatic measurement of the crack,which has practical application significance in engineering.