Computer Vision for Recognition of Materials and Vessels in Chemistry Lab Settings and the Vector-LabPics Data Set

Area of Science:

• Computer Vision
• Machine Learning
• Chemistry Automation

Background:

• Visual recognition of materials in transparent containers is crucial for laboratory tasks performed by humans and robots.
• Existing machine vision methods rely on large, annotated image datasets for training.
• Accurate identification of vessels and their contents is essential for laboratory workflows.

Purpose of the Study:

• To develop a machine learning approach for computer vision-based recognition of materials within vessels.
• To create and release the Vector-LabPics dataset for training and advancing AI models in chemistry.
• To enable automated identification of material phases, regions, and boundaries in laboratory settings.

Main Methods:

• Development of a machine learning model for image-based material recognition.
• Creation of the Vector-LabPics dataset, comprising 2187 annotated images of materials in transparent vessels.
• Training of convolutional neural networks for semantic and instance segmentation tasks.

Main Results:

• The trained neural networks demonstrated high accuracy in detecting and segmenting vessels and material phases.
• Accurate classification of liquids and solids was achieved.
• Relatively lower accuracy was observed in segmenting multiphase systems, such as phase-separating liquids.

Conclusions:

• The developed machine learning approach and the Vector-LabPics dataset show promise for automating material recognition in chemistry labs.
• Further research is needed to improve the segmentation of complex multiphase systems.
• This work contributes to the advancement of AI applications in scientific research and laboratory automation.