U-Net Convolutional Network for Recognition of Vessels and Materials in Chemistry Lab
This work addresses a domain-specific problem for chemistry lab automation, but it is incremental as it applies an existing method to a new dataset with hyperparameter tuning.
The paper tackled the problem of recognizing vessels and materials in chemistry labs using a U-Net convolutional network on the Vector-LabPics dataset, achieving 87% accuracy in vessel recognition and improvements of over 18% in IoU and 19% in accuracy for small datasets.
Convolutional networks have been widely applied for computer vision system. Encouraged by these results, a U-Net convolutional network was applied to recognition of vessels and materials in chemistry lab using the recent Vector-LabPics dataset, which contains 2187 images of materials within mostly transparent vessels in a chemistry lab and other general settings, labeled with 13 classes. By optimizing hyperparameters including learning rates and learning rate decays, 87% accuracy in vessel recognition was achieved. In the case of relatively small training and test sets (relatively rare materials states, the number of training set samples less than 500 and the number of test set samples less than 100), a comprehensive improvement over 18% in IoU and 19% in accuracy for the best model were achieved. Further improvements may be achievable by incorporating improved convolutional network structure into our models.