Multi-information online detection of coal quality based on machine vision

Abstract

Online multi-information detection of mineral properties and composition plays a vital role in the realization of digital mining and digital concentrating mill, and the way of machine vision technology is put forward as a cost-effective and safe approach at present. This paper presents an exploratory study employing a bench-scale approach to detect the multi-information of coal quality online by machine vision simultaneously, including particle size distribution, density distribution, the ash content of each density fraction, and the total ash content. Firstly, we adopt a Finite-Erosion-and-Exact-Dilation (FEED) algorithm and a particle-on-edge region segmentation algorithm to segment overlapped particles and ensure the full analysis of target regions. Moreover, twenty-nine features are extracted and optimized to enable the particle mass estimation model, particle size characterization, classification model of density fraction, and prediction model of ash content to be implemented. Finally, an experimental study shows the merits of the proposed approach, and the average prediction errors of size distribution, density distribution, and ash content of each density fraction are 1.85%, 2.57%, 3.36%, respectively. The total ash content error is 2.54%. Results derived using the proposed approach reveal that it has the potential to be applied to the coal processing industry.