Improving Accuracy in Image-Recognition Systems with Techniques for Handling Similar-Looking Items in Inventory Management

Abstract

Inventory management is crucial for businesses dealing with many visually similar items. This paper presents the development of an automated board game detection system utilizing computer vision and deep learning to enhance the accuracy and efficiency of tracking visually similar inventory. The dataset used to train the model was collected using a Pi Camera Module 3 mounted on a Raspberry Pi 5, ensuring consistency in captured images. To improve the model's performance, images were annotated using CVAT, and dataset augmentation was applied using Augmentations, incorporating transformations such as flipping, brightness adjustment, blurring, and noise addition. These augmentations allowed the model to learn and differentiate board games under varying environmental conditions accurately. The selected model was YOLOv8m, chosen for its balance between accuracy and computational efficiency. The model was trained for 20 epochs using a learning rate of 0.0005, a batch size of 12, and augmentation techniques such as Mosaic and MixUp. Additionally, overfitting prevention techniques including early stopping, dropout, and weight decay were applied. After training, the model was converted into the ONNX format to optimize it for deployment on embedded devices and then uploaded to the Raspberry Pi 5 for realworld testing. The system was evaluated on 10 visually similar or small-sized board games, with 10 tests per game, totaling 100 trials. The results showed that the model successfully classified all board games without a single error. Moreover, these findings demonstrate that the proposed system effectively identify board games with high accuracy while maintaining real-time processing capability. The use of data augmentation and model optimization significantly improved object classification performance, making it a promising approach for inventory management in board game cafes and similar environments.