<p>The performance of flotation is closely related to the froth structure and its morphological features. Traditional segmentation methods, such as the watershed algorithm, struggle to accurately segment flotation froth, making it difficult to extract morphological features. This study proposes a new method for instance segmentation of coal flotation froth images using a Mask Region-based Convolutional Neural Network (Mask R-CNN). By comparing different combinations of upsampling structures and Convolutional Neural Network (CNN) architectures, the optimal network structure is selected as the instance segmentation method. The segmentation results are compared with those of the watershed algorithm. Finally, the average flotation bubble area is extracted from the flotation froth images using the optimal instance segmentation model, and the relationship between the extracted flotation bubble area and ash content is analyzed. The least squares method achieved a correlation coefficient of 0.78, providing a strong theoretical foundation for developing an ash prediction model based on the morphological features of the froth.</p>