Flotation froth image segmentation using Mask R-CNN

Abstract

Automatic control of the flotation circuits needs online information from froth indicators, such as froth texture, motion speed, shape, and number and size distribution of bubbles. In principle, these indicators may be extracted with a machine vision system. This paper presents a real-time image analysis system based on Mask R-CNN for flotation froth segmentation and bubble size measurement. The main objectives were detection of bubbles in flotation froth, measurement of the size distribution of the bubbles, and detection of non-loaded bubbles in the froth. Application of the classical image segmentation methods in an industrial copper flotation plant showed considerable errors in bubble identification and sizing. The accuracy of the proposed method in bubble detection and sizing was evaluated using manually segmented images. The proposed method could detect bubbles with an accuracy of 92.82%, which is a considerable improvement to classical image segmentation methods. The proposed system is installed, tested, and verified in an industrial copper flotation process.