Research On Intelligent Detection Algorithm Of Bridge Cracks

Bridgesare important nodes and key projects for transportation interconnection and communication.They are an important guarantee for national economic development and the safety of people’s lives.The safety and durability of the bridge is related to the national economy and people’s livelihood,so it is necessary to monitor the bridge regularly.At present,the bridge monitoring has been highly valued by the state,especially the detection of bridge cracks,so the research on the bridge cracksintelligent detection algorithm is of great significance.With the development of computer vision,domestic and foreign scholars have applied a variety of technologies to the detection of bridge cracks,and achieved corresponding research results.However,the current mainstream bridge cracks detection algorithms mainly face the following problems:(1)Traditional bridge cracks detection algorithms are based on digital image processing.Most of them involve the processing of various thresholds.Although the method of cracks recognition based on threshold segmentation is easy to implement,such algorithms do not take into account the changes in the image surface environment and the influence of light,noise,texture and other factors on cracks recognition.Therefore,the obtained results are extremely unstable and it is difficult to determine an optimal threshold to process different types of cracks pictures;(2)When the bridge cracks is not a simple form or the bridge cracks image has a complex background and large noise interference,the traditional digital image processing algorithms and shallow machine learning algorithms cannot detect bridge cracks well.In view of the above problems,the specific research contents of this paper are as follows:(1)For the traditional bridge cracks detection algorithms based on digital image processing methods,it is difficult to determine an accurate threshold and parameters,and the adaptability is not strong,and it is difficult to accurately extract and detect bridge cracks.In this paper,a bridge cracks detection method based on image regeneration is proposed.Firstly,the collected real bridge cracks image is subtracted from an equal size pure white image to eliminate some background noise.Secondly,in order to increase the pixel difference between the background,the image enhancement is carried out based on the algorithm of gamma transformation.Thirdly,the histogram of image is used to remove the background of the image,retain the cracks,and regenerate the cracks image with the black background.Finally,the contour find function is used to extract and locate the cracks.The experimental results show that the bridge cracks detected by this method accord with the real area of the cracks,almost no noise interference,and achieve the accurate detection of cracks.(2)The traditional digital image processing algorithms and shallow machine learning algorithms cannot detect the cracks well,however;the deep visual structure features of the neural network based on deep learning are helpful to deal with these problems.Therefore,this paper proposes a bridge cracks instance segmentation network based on Faster R-CNN.Firstly,a crack mask branch is added to the Fast R-CNN network to predict the crack segmentation mask on each region of interest-Then in order to complete the pixel level alignmentwhen the region of interest of the feature map is mapped back to the corresponding region of the original map,the region of interest alignment layer is used instead of the region of interest pooling layer in Faster R-CNN.Finally,in order to generate high-quality crack masks,a prediction network called prediction crack mask intersection over union is implemented.The network module takes the crack instance feature and the crack prediction mask as the input,and learns to predict the quality of the crack instance segmentation mask,so as to improve the segmentation results of the bridge cracks.