Efficient Rock Fragmentation Estimation on Conveyor Belts Using YOLOv8-Based Oriented Bounding Box Method

This paper presents computationally efficient method for estimating the size of rock fragments on conveyor belts in mining operations. Leveraging convolutional neural networks, specifically the YOLOv8 architecture, our approach utilizes oriented bounding boxes and key points to determine the maximum size of rock fragments. Unlike traditional methods, our proposed technique offers a faster and more efficient alternative for real-time processing. Experiments conducted on a dataset collected at a mining and processing plant demonstrate the effectiveness of the oriented bounding box method compared to key points and bounding box. The Rosenblatt-Parzen method was used to estimate the distribution density function for the fragment size distribution. The YOLOv8 model exhibits reliable performance in accurately detecting rock fragments and estimating their sizes. The YOLOv8-based method, particularly with oriented bounding boxes, emerges as a promising technique for accurately estimating the size distribution of rock fragments on conveyor belts, contributing to advancements in mining technology and energy cost reduction.