Image restoration and analysis with application to quality variable prediction in flotation process

Flotation process is commonly used to separate valuable minerals from undesirable ones. Measurement of flotation froth concentration is essential for control applications. The concentration of froth is often evaluated visually by operators and through the laboratory analysis of collected samples over a long period of time. However, the latter suffers from significant time delays and the former is not very reliable. Computer vision is widely used in industry to analyze the froth concentration based on various froth image features. In this study, we propose a computer vision model to estimate the froth concentration. Important practical challenges including contaminated images with bright lighting spots, camera noise and outliers are addressed. First, the contaminated images are restored using a modified Kalman filter. The color, texture, and deep features of froth images are then extracted using gray-level co-occurrence matrix (GLCM) and a transfer learning-based convolutional neural network (CNN), respectively. Afterward, a regression model is built to estimate the froth concentration based on the integrated visual features. An expectation–maximization (EM) algorithm is then applied to determine the unknown parameters of the model, which is also robust against outliers by using a t -distribution to model the noise. Experiments conducted on a batch flotation column for tailings bitumen extraction show that the proposed algorithm is able to estimate the bitumen concentration with reasonable accuracy. The proposed algorithm is a time-saving alternative to time-consuming and costly laboratory analysis and is essential for advanced process control applications.