Minima-YOLO: A Lightweight Identification Method for Lithium Mineral Components Under a Microscope Based on YOLOv8

Area of Science:

• Geological Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Smart mining relies on efficient mineral identification for ore sorting.
• Current methods may not be suitable for deployment on resource-constrained edge computing devices.

Purpose of the Study:

• To develop a lightweight and efficient lithium mineral identification method for edge devices.
• To enable rapid mineral sorting in smart mining applications.

Main Methods:

• Proposed Minima-YOLO, a lightweight model based on YOLOv8.
• Introduced YOLOv8-tiny, Faster-EMA module with PConv and EMA attention, GhostConv, and Slim-Neck structure.
• Utilized visible light microscopy for lithium mineral image data.

Main Results:

• Achieved 99.4% mAP50 on a self-constructed lithium mineral dataset.
• Reduced FLOPs to 2.3 G, parameters to 0.72 M, and model size to 1.63 MB.
• Maintained a high inference speed of 103 FPS.

Conclusions:

• Minima-YOLO offers a highly efficient, lightweight solution for lithium mineral identification.
• The algorithm demonstrates superior performance compared to other object detection methods.
• Provides an intelligent, eco-friendly computer vision method for rapid lithium mineral sorting.