PS-YOLO-seg: A Lightweight Instance Segmentation Method for Lithium Mineral Microscopic Images Based on Improved...

Area of Science:

• Geoscience
• Computer Science
• Materials Science

Background:

• Microscopic image recognition is vital for mineral analysis and smart mining.
• Traditional methods face challenges in deployment on edge devices.

Purpose of the Study:

• To develop a lightweight segmentation model for lithium ore analysis under visible light microscopy.
• To address limitations of traditional methods and enable edge device deployment.

Main Methods:

• Proposed PS-YOLO-seg, a compressed lightweight segmentation model.
• Utilized width factor adjustment, PSConv-based downsampling, C3k2-PS module, and Segment-Efficient head.
• Evaluated on a self-constructed lithium ore microscopic dataset.

Main Results:

• PS-YOLO-seg achieved slightly improved segmentation performance over YOLOv12n.
• Reduced FLOPs by 20%, parameter count by 33%, and model size by 32%.
• Demonstrated faster inference speed, suitable for real-time deployment.

Conclusions:

• Architectural optimization significantly enhances instance segmentation for lithium minerals.
• The model is compact, fast, and suitable for industrial and edge computing scenarios.