This study focused on image classification using deep learning and investigated methods for improving classification accuracy.
Image classification is a critical technology in various application fields, such as automated checkout systems, and there is a growing need to improve its accuracy.
In this study, we focused specifically on images of Western-style confectionery. We trained the model using 14 food classes, including cupcakes and macarons, after adding new image data, and evaluated the classification accuracy using the macro-average, a metric for multi-class classification tasks.
The results showed that accuracy improved with each addition of cupcake and macaron images, indicating that training the model with newly added image data is effective for improving classification accuracy. We also found that accuracy tends to improve as the amount of training data increases. Furthermore, we developed a web system to make image classification accessible to anyone, allowing users to perform image classification by selecting or dragging and dropping image files to upload them.
In the future, we aim to improve accuracy not only by increasing the amount of training data but also by optimizing hyperparameters and balancing the data.
Figure 1. Example Screen of the Western Confectionery Image Classification System
