YNU-Industrial Fasteners: a benchmark dataset for industrial fasteners classification

Industrial automation and computer vision are developing quickly and, while labor costs have significantly increased as a result of the aging population, it is urgent to improve the level of automation and information technology in industrial production. Automatic assembly based on an industrial robotic arm has obvious advantages over the traditional manual assembly method in various aspects such as efficiency, accuracy, and flexibility. But there are still many shortcomings in the application of industrial robotic arms in automatic assembly. The main technical bottleneck is the problem of classification and identification of industrial fasteners during automatic grasping. The existing methods for fastener classification and recognition are not accurate and lack public targeted industrial fastener datasets. In this paper, a new industrial fasteners dataset -- the YNU-Industrial Fasteners Dataset (YNU-IFD) -- is proposed and a Transformer-based classification neural network-RtNet is constructed. By comparing the experiments of various deep learning models on the self-built dataset (YNU-IFD), the results show that this dataset has better performance in application to standardized mechanical sorting.