An efficient neural architecture search based deep learning algorithm for surface defects detection of T-beam

The surface defect detection of T-beam of highway bridge in the traditional quality control process is mainly based on naked eye observation. The detection results are too dependent on personal experience and require a significant amount of work. This paper applies the convolutional neural network constructed by differentiable neural architecture search method to the automatic detection of surface defect on T-beams. And a reduction cell of the modified network is designed to search for the normal cell to improve the search efficiency. The results show that the search efficiency of the network is improved and the search time is reduced by about 31%.