A Computer Vision System for Bitumen Content Estimation in Flotation Froth with Degraded Images*

Abstract

Listen

Translate article

Oil sand is one of the main oil resources in Canada. In oil sands operation, primary separation cell, where bitumen is separated from the sand using water-based gravity separation, generates froth, middling and tailing. Tailing contains solids, water, and residual bitumen. The presence of bitumen in the tailing leads to loss of bitumen and increases environmental footprints. In order to remove residual bitumen, tailing is sent to a flotation cell. Determining the bitumen content in the froth of the flotation cell is essential for ensuring the efficiency of the process operation and improving environmental performance. Thus, samples of the flotation froth are often collected over a relatively long period of time and then sent to a laboratory for analysis. While laboratory analysis can provide a more accurate measurement of cumulative bitumen content over the sample collection period, it is costly and time-consuming. The long processing time in the laboratory will hinder real-time applications. In recent years, computer vision technology has been widely used to monitor and control flotation processes. This article proposes a computer vision-based solution for estimating cumulative bitumen content in flotation froth that usually contains a relatively small amount of bitumen. In spite of the frequent recording of images of the froth, the corresponding bitumen content label (laboratory analysis result) is not commonly available. The froth is collected batch by batch, and there is only one laboratory analysis of bitumen content in each batch, which reflects the cumulative bitumen content collected over the entire batch. Furthermore, images of the froth are often degraded by noise and varying lighting conditions. To solve the problem, a Kalman filter-based algorithm is proposed to restore the contaminated images. Gray-level co-occurrence matrix (GLCM) is then used to extract color and textural features from the restored images. Using these features, a linear regression model is built for predicting cumulative bitumen content over each batch. Validation results in a laboratory-scale flotation process demonstrate the proposed algorithm is promising in estimating bitumen content.