Multi-Scale Feature Fusion for Coal-Rock Recognition Based on Completed Local Binary Pattern and Convolution Neural...

Area of Science:

• Geosciences
• Computer Science
• Artificial Intelligence

Background:

• Intelligent coal mining requires accurate automatic coal-rock recognition.
• Existing methods often suffer from poor performance and low robustness.
• Distinctive visual features of coal and rock are crucial for effective recognition.

Purpose of the Study:

• To propose a robust multi-scale feature fusion coal-rock recognition (MFFCRR) model.
• To enhance the accuracy and performance of coal-rock identification in mining applications.
• To address the limitations of current coal-rock recognition techniques.

Main Methods:

• A multi-scale Completed Local Binary Pattern (CLBP) was used for texture feature extraction (TFE).
• A Convolutional Neural Network (CNN) was employed for deep feature extraction (DFE).
• Fusion of multi-scale CLBP and deep features, followed by nearest neighbor classification with chi-square distance.

Main Results:

• The proposed MFFCRR model achieved a coal-rock image recognition accuracy of 97.9167%.
• This accuracy represents a 2%-3% improvement over existing state-of-the-art methods.
• The model effectively integrates texture and macroscopic information for improved recognition.

Conclusions:

• The MFFCRR model demonstrates superior performance and robustness for automatic coal-rock recognition.
• The fusion of multi-scale texture and deep features is effective for intelligent coal mining.
• This approach offers a significant advancement in coal-rock identification technology.