A Design and Implementation of Stationery Product Recognition Method Using Two-Stage YOLO v8 Model

A company typically maintains a lot of stationery products, such as ball-pens, glue-sticks, and erasers, for daily use in the operations. However, their accurate managements will cost a lot or often be impossible. Currently, YOLO is a very popular model for object recognition using images due to high capabilities. In this paper, we present a design and implementation of a stationery product recognition method using the latest YOLO v8 model at two stages. The first-stage model recognizes the category of the target object from the given image. The second-stage model selects the specific product in the category. For evaluations, we prepare 795 images of 45 stationery products in 9 categories and train the two-stage YOLO models using NVIDIA's RTX-3060 GPU. Then, we measure the training time and the recognition accuracy of the proposal. The results show the effectiveness of the proposal.