Research On Bridge Defect Recognition Method Based On Support Vector Machine

With the rapid development of my country’s transportation industry,although various bridges with good tensile and crack resistance,strong rigidity and low weight have been constructed or are being constructed,many of the previously constructed bridges have been overloaded and brought great safety to the society.Hidden dangers,bridges need to be inspected regularly to maintain normal operation and maintenance.Traditional manual detection methods are obviously difficult to meet today’s needs due to their high cost and low efficiency.In modern detection technology,bridge disease detection methods through image processing are emerging day by day,but there are also many problems.For example,due to factors such as light and water surface reflection in the natural environment,the detection effect is greatly affected;and the existing bridge disease detection research objects are mostly cracks,while other diseases(honeycomb,hemp surface,exposed tendons,etc.)are rarely studied in depth In addition,most of the current real-time detection systems for bridge diseases collect bridge images and send them to the ground station for processing.Due to the limitation of the communication channel,the real-time performance is low.Therefore,this paper proposes the research of bridge disease identification method based on support vector machine algorithm,and uses FPGA chip as the hardware experiment platform to design a bridge disease detection system that can collect,detect and identify diseases in real time at the front end of the video system.In this paper,the data of bridge disease images collected by the experimental platform after image preprocessing,image segmentation and feature extraction are used as training samples,and the recognition method of support vector machine is used to classify and compare the recognition accuracy.Finally,the best performance is selected.The classifier verifies the feasibility and portability of the algorithm,and transplants the design ideas to the FPGA to implement the HDL language of each algorithm module to complete disease-free images and disease-free images(honeycombs,cracks,exposed tendons,and pitted surfaces)Disease image)real-time recognition.The main work of this paper is as follows:(1)After the UAV’s onboard camera collects the bridge image,it is imported into the Matlab software to preprocess the image and perform median filtering and noise reduction processing.And comparing different image segmentation methods,due to poor results,the Sobel operator is improved to obtain satisfactory segmentation results.Finally,the gray-level co-occurrence matrix method is used to extract the four texture features of energy,entropy,contrast and autocorrelation.(2)In the stage of bridge disease classification,the basic principles and applications of support vector machines are introduced.By comparing three support vector machine multi-classification design methods,the "one-to-one" multi-classification method is selected and combined with the grid search algorithm.A multi-classification model of support vector machines based on parameter optimization for bridge diseases was established,and the recognition accuracy of multi-classifiers using different kernel functions was compared through experimental analysis.Finally,the radial basis kernel function classification model was used,and the bridge disease recognition rate reached95%.(3)After completing the algorithm comparison and verification in Matlab,use Altera’s FPGA development board with Cyclone IV series chips as the control core to design and implement the hardware system.The HDL language is used to complete the module design of grayscale,median filtering and edge detection,and the classifier design of the kernel function is implemented in the NIOSⅡ kernel.This recognition system utilizes the advantages of FPGA parallel processing and multi-stage pipeline design to improve the speed of the system;in the feature extraction and classification recognition module,SOPC technology is used,and the realization of complex algorithms is solved through the method of combining software and hardware.Finally,the disease information is displayed in real time on the TFT screen,which meets the real-time requirements of disease identification.