Research On New Technology Of Bridge Block Detection Based On Image Processing And Machine Learning

Since the foundation of PRC,especially after the Reform and Opening Up,China has experienced a leaping development in the transportation field.As the length of highway and railway increases,the number of bridge also increases.While the construction of new bridges is still on-going,the maintenance of in-service bridge is becoming more and more important and urgent.To maintain or repair bridges with diseases,first we need to detect the diseases.Currently this step still relies on manpower,which costs amounts of human resources and money,along with lots of danger and limitation.However,we can find some useful technology and tools from other industry to simplify our work.One example is the civil drone.Civil drones can go to positions where human find it hard to access.Besides,civil drones can take photos with large-scale pixels.If we can detect diseases from photos,that would be a new method to solve our problems.Based on the background above,this paper does some researches on algorithms of stitching images to get a panorama of bridges and auto-detecting diseases from digital images taken by civil drones.Firstly,researching on stitching images.Bridges are often too large to be photoed in one clear picture.In that case,we need to find a way to stitch multi photos and generate a panorama,which will help professionals to fully understand the situation of bridge surface and give reasonable assessment on the health conditions of bridges.Secondly,trying to extract the disease region from the image.Diseases often have special patterns and to make them distinguished,some image processing procedure need to be done,so that further analysis can be made.The next step is to classify different diseases.There are many kinds of diseases with different patterns.After the image processing,some data that indicate the patterns of disease region has been recorded,which can be applied to machine learning algorithms to auto classify the diseases.If the civil drone has taken amounts of photos,this step can save the time for human to manually classify the photos.At last,a relatively complete software that can systematically do all the processing steps above is developed,and a demo is given to show the results.The research in this paper is based on some formal researchers and has been inspired by some other industries.Compared with others,different algorithms are tested to measure the performance and efficiency to finally choose a better algorithm.About the software,it works well and more features can be expected.