Intelligent Coal-Gangue Sorting Method Based on CCD and DE-XRT Image Fusion

Area of Science:

• Materials Science
• Computer Science
• Engineering

Background:

• Single-modal ore sorting faces challenges like missed detection, false positives, and low accuracy in complex environments.
• Existing methods struggle with accuracy and efficiency in distinguishing coal from gangue.

Purpose of the Study:

• To develop an intelligent coal-gangue sorting method that overcomes limitations of single-modal approaches.
• To improve accuracy, reduce false detections, and enhance efficiency in complex sorting environments.

Main Methods:

• Collected industrial Charge Coupled Device (CCD) and Dual Energy X-ray Transmission (DE-XRT) images of coal and gangue.
• Applied adaptive weighted fusion for preprocessing X-ray images and constructed a dual-branch MobileNetV4 network.
• Integrated the SimAM attention mechanism to boost feature representation and recognition accuracy.

Main Results:

• The fused image approach significantly reduced computational complexity (FLOPs by 64.9%) and model parameters (by 61.7%).
• Recognition accuracy improved by 7.6% (vs. CCD) and 28.3% (vs. XRT).
• Achieved 100% recognition accuracy with no missed or false detections, demonstrating real-time sorting capability.

Conclusions:

• The proposed fusion method effectively combines surface and internal feature information for superior coal-gangue sorting.
• This intelligent sorting system offers a robust solution for complex environments, enhancing both performance and efficiency.
• The lightweight network architecture ensures practical applicability for real-time industrial sorting.