Construction of Image Datasets for Chemical Experiments and Numerical Assessment of Object Detection Methods

Abstract

In recent years, the remarkable advances in artificial intelligence technology have led to digital transformation (DX) in various fields. The automated construction of laboratory notebook through filming experiments is a promising application of image recognition for chemistry． In this study, we created an image dataset of chemical experiment, which contains 2376 images and consists of 7 classes of objects. Object detection methods and a multiple object tracking method were implemented and assessed using the dataset toward to develop automated laboratory notebook system.