Research On Building Crack Detection And Measurement Method Based On Deep Learning And Image Processing Technology

Crack is a common and important disease in housing construction.The detection of crack is of great significance for structural detection,identification and maintenance of dangerous buildings.With the help of deep learning and image processing technology,this paper studies the recognition algorithm of building crack image,the optimization method of recognition results and the calculation method of crack parameters.This paper also realizes the construction of crack detection and measurement system,which can provide a lot of reference data for housing quality inspection and evaluation.This paper mainly carried out the following research :(1)The deep learning model of Mask R-CNN is applied to the crack recognition of buildings.A large number of training samples are obtained through data enhancement of data set.The final crack recognition effect is better than that of traditional image processing technology.The crack recognition rate and recognition accuracy are improved.(2)Based on the deep learning model of Mask R-CNN to identify cracks,this paper proposes a method to modify the boundary of cracks obtained by deep learning recognition through RGB model of image,so as to obtain more accurate crack shape.Through the RGB three-dimensional scatter map of the pixels in the crack recognition frame,the concepts of the crack point cloud and the background point cloud are proposed,which can more intuitively show the effect of the optimization of the deep learning recognition results.(3)The two-dimensional code is used as the image mark.The two-dimensional code is pasted next to the crack when the image is taken.The cracks are numbered by two-dimensional code.At the same time,the geometric features of the two-dimensional code are used to calibrate the crack image.The relationship between the actual length and the pixel is established.An improved skeleton extraction algorithm based on K3 M algorithm is proposed to extract the skeleton from the fracture pattern after deep learning recognition and optimization.Through image sub-pixel processing,the problem of skeleton fracture in fracture skeleton extraction is solved;through the distribution of skeleton pixels in the 3 × 3 domain of skeleton pixels,the situation of skeleton endpoint is judged,and the skeleton is corroded from the endpoint to solve the multi bifurcation problem in fracture skeleton extraction.(4)A method is proposed to determine the width direction of the fracture points to be measured through the 5 × 5 neighborhood pixels of the skeleton points to be measured,and to calculate the fracture width,length and area parameters through the fracture skeleton.The evaluation system of fracture width measurement accuracy is proposed,and the validity and stability of fracture parameter calculation method are analyzed and verified through simulation fracture test.(5)By integrating deep learning crack image recognition technology,deep learning recognition result optimization technology,two-dimensional code recognition calibration technology and crack parameter calculation method,a crack detection and management system is built.The system algorithm consists of two-dimensional code identification and calibration module,crack identification module,crack identification result optimization module and crack parameter calculation module.The website can interact in real time,which improves the operability of using deep learning and image processing technology to recognize and measure building cracks.