Ore image segmentation method using U-Net and Res_Unet convolutional networks.

Area of Science:

• Mining Engineering
• Computer Vision
• Image Processing

Background:

• Accurate particle size distribution is crucial for crushed ore analysis.
• Existing image segmentation methods struggle with ore particle adhesion and image quality issues.
• Challenges include dark areas in blast heap and conveyor belt images, reducing segmentation accuracy.

Purpose of the Study:

• To propose a novel image segmentation method (UR) for enhanced crushed ore particle analysis.
• To improve segmentation accuracy by addressing issues of particle adhesion and dark image regions.
• To validate the efficiency and accuracy of the UR method against existing techniques.

Main Methods:

• Image preprocessing using gray-scale, median filter, and adaptive histogram equalization.
• Deep learning models (U-Net and Res_Unet) for ore contour detection and optimization.
• Image segmentation results visualized using OpenCV.

Main Results:

• The UR method demonstrates improved accuracy in segmenting crushed ore images.
• Preprocessing techniques effectively reduce noise and extract target regions.
• Comparative analysis validates the superior performance of the UR method.

Conclusions:

• The UR deep learning-based image segmentation method offers a significant advancement for particle size distribution analysis.
• The method successfully overcomes common challenges in ore image segmentation.
• The UR method provides a more accurate and efficient solution for open pit mine applications.