Large Crane Surface Crack Detection Based On Convolutional Neural Network

In the 21st century,the rapid development of economy is based on the efforts of countless people.Therefore,it is very important to ensure the life safety of the workers who create these economic developments in real life.There are tens of thousands of construction sites in China,and there are numerous defects on the surface of countless production parts on these construction sites.For example,the large crane often used in the construction site,after using for a period of time,due to the influence of external factors such as stress,cracks and other defects will appear.Once large crane cracks,it will be scrapped and can not be used.Therefore,in order to avoid potential safety hazards,it must be tested before each use.Traditional detection methods not only cost a lot of human and material resources,but also affected by many external factors.Therefore,the research of automatic detection technology based on defects can not only control the production cost to some extent,but also promote the improvement of production efficiency and the guarantee of product quality.At the same time,it can lay a solid foundation for the intelligent transformation of manufacturing industry.Firstly,the research background and status of defect detection technology based on convolution neural network are described,and the advantages and disadvantages of various defect detection algorithms are compared.It also introduces the background knowledge of statistical learning theory and neural network theory based on defect detection,which lays the foundation for the subsequent research of defect detection based on convolution neural network.Secondly,the data set is established for the research of defect detection technology based on convolutional neural network.Based on the fact that it is difficult to obtain the training set of crack defects on the surface of large cranes,the UAV is used to obtain the image data of the surface of large cranes on the actual construction site.Through data cleaning,the data set suitable for the actual training is selected.Then,based on one-step method and two-step method,the crack detection algorithms based on Faster R-CNN and SSD are proposed respectively,and the algorithm structure and process,network training and deployment,performance analysis and other aspects are described.The final experimental results show that the algorithm based on Faster R-CNN is feasible,R-CNN’s crack detection algorithm is better in accuracy,while SSD based crack detection algorithm has the advantage of speed.Finally,a feasible scheme of crack detection system based on convolutional neural network and HTTP transmission is proposed.In this scheme,the images captured by UAV are transmitted based on HTTP,and the images are transmitted to the server(computer).The server detects and analyzes the incoming images through the crack defect detection algorithm based on convolution neural network,and sends the detected results back to the client(whether there are defects,if so,the type and specific coordinates of the defects).Experiments show that the scheme is feasible.