Intelligent Detection Method For Concrete Structure Surface Cracks Considering Underwater Environment

Concrete cracking problem is common in the main structure of buildings,which will accelerate the weathering,erosion and damage of the main structure and seriously affect the service life.Concrete crack detection plays an important role in structural safety assessment.Traditional crack detection methods mostly refer to manual inspection,which is a time-consuming,inefficient,subjective and laborintensive task.In particularly,it is more difficult to identify and detect cracks in wading buildings.Therefore,considering the factors of water environment,this paper studies the intelligent detection method of air and underwater concrete cracks based on digital image,puts forward the full convolution neural network R-FPANet crack intelligent segmentation model and the double convolution neural network underwater crack classification method,and realizes the automatic detection of air and underwater concrete surface cracks.The main research results of this paper are as follows:(1)The full convolution neural network R-FPANet crack intelligent segmentation model of concrete surface cracks in air is established.Using the idea of deep learning,an intelligent segmentation model of crack is established from the crack image of concrete surface.The model integrates self-attention mechanism and feature pyramid network.The former strengthens the interdependence between features and highlights important feature through attention map,while the latter emphasizes feature reuse and enhancement the fusion of shallow features and highlevel features.Based on the crack binary image,combined with digital image technology,the length,area,mean width and maximum width of the crack are quantified at the pixel level.Experiments show that the R-FPANet proposed in this paper has strong generalization ability and more prominent segmentation details.(2)An underwater concrete surface crack enhancement and classification method based on double convolution neural network is proposed.Due to the scattering and absorption of light by water and suspended solids in water,problems such as uneven illumination,blur and low contrast appear in underwater imaging,which increases the difficulty of crack identification.Firstly,contrast stretching,gamma color correction and contrast enhancement algorithms are used to enhance the underwater image to make the difference between the crack and the background greater;Then the convolution neural network is designed to enhance the underwater image to achieve the enhancement effect of the traditional method.Finally,the enhanced images are input into a new convolutional neural network for classification training to realize the intelligent detection of underwater cracks.Experiments show that the double convolution neural network can realize the endto-end intelligent detection of underwater concrete cracks,which can not only achieve high recognition accuracy,but also improve the efficiency of underwater crack detection.(3)An intelligent identification system on concrete surface crack considering water environment based on Android is developed.The system is mainly composed of server and mobile client.The server is built using the framework of Flask,which is mainly used to deploy the deep learning model and return the final identification results.The mobile client is mainly composed of storage and display module,identification module and user setting module.It has the functions of consulting the identified and saved results,selecting pictures to upload and identify,user center and so on.The system can effectively detect and identify cracks,provide a new tool for building structure safety detection.