|
การจำแนกหมวดหมู่สินค้าบนระบบอีคอมเมิร์ซโดยใช้เทคนิคการเรียนรู้เชิงลึกวิเคราะห์ภาพสินค้า |
|---|---|
| รหัสดีโอไอ | |
| Creator | ฐิติรัตน์ ศิริบวรรัตนกุุล |
| Title | การจำแนกหมวดหมู่สินค้าบนระบบอีคอมเมิร์ซโดยใช้เทคนิคการเรียนรู้เชิงลึกวิเคราะห์ภาพสินค้า |
| Contributor | ภาณุพันธ์ เมฆสุวรรณ์ |
| Publisher | King Mongkut’s University of Technology Thonburi |
| Publication Year | 2566 |
| Journal Title | KMUTT Research and Development Journal |
| Journal Vol. | 46 |
| Journal No. | 2 |
| Page no. | 153-172 |
| Keyword | Deep Learning, Image Classification, Categorizing products, E-commerce |
| URL Website | https://journal.kmutt.ac.th/ |
| Website title | เว็บไซต์วารสารวิจัยและพัฒนา มจธ. |
| ISSN | 2697-5521 |
| Abstract | This research aimed to create a deep learning model to classify categories of products that are being sold on an e-commerce platform by analyzing images of the products. Different techniques were comparatively assessed and their efficiencies were compared to arrive at the best model. The public image dataset of the Shopee e-commerce system, consisting of 38 product categories, with a total of 106,309 images, was used. As these images belong to actual product images in the Shopee e-commerce system, they are highly complex in terms of computer vision and image analytics. Six models were tested along with 2 different loss functions. The results revealed that EfficientNetB5 was the best model in terms of accuracy, considering the size of the model and the time used in classifying the images; the accuracy on the test set was noted to be 84% while the value was 92% on the additional test set. The inference time used was 0.068 seconds per image with the model size of 141.9 MB. When the model was tested for the task of image classification in comparison with the actual Shopee application, the EfficientNetB5 model was 31.5% more accurate. When compared to four human participants, although our model was 7.16% less accurate, the model was 25.5 times faster in classifying each image. |
