GOODS IDENTIFICATION BASED ON IMAGES ON MOBILE DEVICES

Abstract

The retail market in Vietnam has been assessed as developing quite strongly in recent years. Retail markets in cities and urban areas continue to grow with sophisticated and modern forms, and even the rural retail market has been receiving attention and diversifying various types. The increasing number of retail stores has made shopping more convenient. However, managing goods and prices in small and medium-sized retail stores is always challenging when there are too many items. In addition, employees also need time to learn and become familiar with the store's products. This unintentionally leads to poor experiences for customers, affecting their shopping satisfaction and the store's reputation. Furthermore, the cost of supporting equipment such as barcode scanners and management software is a barrier for many stores, especially small stores or those in rural areas. They often face difficulties when trying to adapt to modern equipment. This research presents a method to support product identification through images on mobile devices using deep learning architectures such as You Look Only Once (YOLO) - one of the best object groups of detection models with modern performances. The method to get mAP (Mean Mean Accuracy) is 55.6 with 17GFlops. The result is an application that includes vocabulary, meanings and sentences with the search word through object recognition and detection, and the objects that we used to identify in this study are the output, popular objects in Can Tho city. To do this, our method is to select the model that best fits and has good accuracy with the model trained from 2890 images belonging to 100 different classes using the application on the device for object detection is appropriate.