Research And Implementation Of Building Surface Crack Detection Telchnology Based On Deep Learning

Concrete plays an important role in the construction of transportation infrastructure such as dams,bridges and tunnels.However,the concrete members of these building often have different degrees of cracks due to the bad weather,used time,material quality and construction techniques,which results in certain damage to the concrete members.This situation can have a serious impact on the life of the infrastructure,pedestrian safety and social economy.Therefore,the detection of defects such as cracks on concrete surfaces has become a key research issue.This thesis presents a crack image detection and segmentation method based on deep learning.The method is mainly divided into two stages.In the first stage,we extract abstract features through convolutional neural networks and preserve the feature maps for each layer.In the second stage,we use the transposed convolution to enlarge the feature map and fuse it with the features of the low-dimensional features to obtain a network model with higher segmentation accuracy.In addition,we propose a boundary weighted loss function suitable for the segmentation network of small objects,which gives different weights according to the distance from the pixel to the boundary,so that the fine boundary target is more easily judged as a positive sample.This can reduce the discontinuous prediction results and improve the detection accuracy to some extent.In order to meet the actual application scenarios,we also construct a real crack data set for training and verification of the model.After that,we analyze the segmentation results of the crack based on image processing techniques.The images of cracks are processed through basic morphological methods such as corrosion,expansion,opening and closing operations to obtain segmented marker maps,skeleton maps,and geometric parameter information.These can provide a better way for professional engineers to evaluate and repair the defects.Finally,we develop a web-based intelligent defect detection system of prototype that integrates the crack detection and analysis methods.Through the platform,we can import images that need to be detected in batches,and then get a list of image information with cracks.We can also export test results for offline viewing with a high practical value.