A Two-Stage Framework With Ore-Detect and Segment Anything Model for Ore Particle Segmentation and Size Measurement

This article proposes a novel detection-driven method for ore particle (OP) segmentation, specifically designed to enable precise measurement of particle size distribution (PSD), a critical factor in mineral processing. Accurate segmentation is essential for calculating particle size, which directly contributes to the evaluation of PSD and ensures effective ore quality control and processing efficiency. Conventional segmentation methods often struggle to generalize across diverse ore types, irregular particle geometries, and complex spatial arrangements, limiting their robustness and adaptability in real-world applications. The proposed method overcomes these challenges by first using Ore-Detect, a detection module that efficiently localizes OPs, followed by a refined segmentation process using the segment anything model (SAM) to precisely delineate particle boundaries. This two-stage approach ensures high accuracy in particle segmentation and size measurement, even in the presence of occlusions and significant size variations. Experimental results demonstrate the effectiveness of the proposed method, achieving state-of-the-art performance across key metrics: mean intersection over union (mIoU) of 85.31%, precision of 88.81%, recall of 87.12%, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$F1$ </tex-math></inline-formula>-score of 87.95%, average precision (AP) of 94.06%, boundary displacement error (BDE) of 6.98, and object-level accuracy (OLA) of 89.51%. Furthermore, the framework achieves a PSD measurement error within 5%. In terms of computational efficiency, the proposed method processes 100 ore images of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$512\times 512$ </tex-math></inline-formula> resolution in 11.27 s, demonstrating its ability to meet real-time industrial requirements.