当前位置：
X-MOL 学术
›
Miner. Eng.
›
论文详情
Our official English website, www.x-mol.net, welcomes your
feedback! (Note: you will need to create a separate account there.)
Enhanced pre-selection of spodumene via ultraviolet-induced fluorescence and improved YOLOv8 deep learning algorithm
Minerals Engineering ( IF 4.9 ) Pub Date : 2025-05-16 , DOI: 10.1016/j.mineng.2025.109432
Tianyou Yu, Yimin Zhu, Jie Liu, Yuexin Han, Yanjun Li
Minerals Engineering ( IF 4.9 ) Pub Date : 2025-05-16 , DOI: 10.1016/j.mineng.2025.109432
Tianyou Yu, Yimin Zhu, Jie Liu, Yuexin Han, Yanjun Li
With the rapid development of the new energy industry, the demand for lithium resources has significantly increased. Spodumene, a key lithium mineral, often mixes with gangue during mining, leading to high energy consumption and carbon emissions during beneficiation. To enhance the Li2 O grade and reduce production costs, intelligent pre-selection experiments using deep learning algorithms were conducted. We found that spodumene ore emits red light under 365 nm ultraviolet illumination. In the pre-selection process of spodumene, traditional sorting methods cannot effectively distinguish spodumene from gangue minerals such as lepidolite and quartz due to their similar colors, leading to energy waste. However, utilizing ultraviolet fluorescence sorting technology enables efficient pre-selection of spodumene ores that emit red light under 365 nm ultraviolet illumination, thereby improving pre-selection efficiency. This study focused on spodumene from Dahongliutan mine, Xinjiang, and proposed an ore recognition method based on an improved YOLOv8 model, termed ME-YOLOv8. Experimental results show that ME-YOLOv8 achieved an accuracy of 90.2 %, a recall rate of 75.2 %, and an average precision of 85.1 %. Compared to the original YOLOv8s, the model’s weight was reduced by 48.94 %, with improvements in accuracy, recall, and precision by 2.4, 2.2, and 1.7 percentage points, respectively. The algorithm was tested on intelligent photoelectric pre-selection equipment under various ore particle sizes, feeding speeds, and conveying speeds. Results indicate ME-YOLOv8′s superiority in improving the grade of pre-selected concentrate and waste rejection rate, maintaining high recognition accuracy under larger particle sizes and high-speed conveying. Compared to traditional X-ray methods, the deep learning-based ultraviolet fluorescence pre-selection method shows higher effectiveness and adaptability. This technology can achieve intelligent separation of spodumene and gangue, reduce grinding and flotation processing volumes, enhance the Li2 O grade of feed ore, lower production costs, and mitigate the environmental impact of flotation reagents. The open-source code and dataset will be released at: https://github.com/98123111/98123111.
更新日期：2025-05-16
京公网安备 11010802027423号