Classification of gas dispersion states via deep learning based on images obtained from a bubble sampler

•Bubble images were classified into five classes based on the gas dispersion states.•A new deep CNN was built and trained with a newly created bubble image dataset.•The CNN model achieved a high accuracy with a short execution time.•The CNN model could be used for online gas dispersion monitoring and alarming tool.•The pretrained model and newly created bubble image dataset are open-sourced. The gas dispersion state within the bubble columns and reactors greatly affects their performance. A bubble sampler is a useful device to measure gas dispersion in the operating bubble columns and reactors as it has an adjustable sampling part that allows the device to obtain bubble samples from desired regions. However, implementing the bubble sampler as a real-time gas dispersion monitoring tool is difficult owing to the lack of reliable and automated image processing approach. In the present study, we developed a new convolutional neural network model to classify the gas dispersion states in a bubble column based on the images obtained with a bubble sampler. The model was trained with a labeled bubble image dataset comprising five different classes, which corresponded to five different gas dispersion states in the column. The average classification accuracy of the model was 97.5%. It was demonstrated that the trained model can accurately identify the change in gas dispersion state in real-time. The dataset created in this investigation and the pretrained BubbleNet can be found at http://dx.doi.org/10.17632/m3zjf8z286.1. [Display omitted]