表題番号：2022C-635
日付：2023/04/06
研究課題化学実験画像の物体検出のためのデータセットの構築
| 研究者所属（当時） | 資格 | 氏名 | |
|---|---|---|---|
| （代表者） | 理工学術院 先進理工学部 | 助教 | 藤波 美起登 |
| （連携研究者） | 先進理工学部 化学・生命化学科 | 教授 | 中井 浩巳 |
- 研究成果概要
- In this research, we aim to develop a system that automatically recognizes videos of chemical experiments using image recognition technology. The system is expected to be applied to create electronic laboratory notebooks and to warn against dangerous actions automatically. Image recognition using machine learning requires a dataset with coordinates and names of objects in images. We have constructed a dataset of about 2300 images and 5000 objects. However, further expansion of the dataset is necessary to improve the prediction accuracy. In this project, we expanded the chemical experiment image data set. Documentation was prepared so that anyone could create a dataset. The dataset was expanded with the research assistants' cooperation. We extended the dataset to about 5000 images and 16000 objects. The prediction accuracy of test data was evaluated. Mean Average Precision (mAP) was used as an evaluation index. The mAP takes values in the range from 0 to 1, and the closer to 1, the higher the prediction performance. Learning on the expanded dataset improved mAP to above 0.8. This study was reported in two conference presentations, including an international conference.