Gabor-GLCM-Based Texture Feature Extraction Using Flame Image to Predict the O₂ Content and NO .

Area of Science:

• Combustion engineering
• Image processing
• Boiler monitoring

Background:

• Industrial boiler flame images differ significantly from laboratory settings.
• Existing methods lack accuracy due to limitations in capturing industrial flame characteristics.
• Accurate flame image analysis is crucial for effective boiler control.

Purpose of the Study:

• To propose a novel flame image texture feature extraction algorithm specifically for industrial boilers.
• To enhance the characterization of industrial boiler flame images for improved combustion monitoring.
• To develop a more accurate combustion process regression model for predicting flue gas emissions.

Main Methods:

• Texture features were enhanced using Gabor filters on RGB channels.
• Gray-Level Co-occurrence Matrix (GLCM) was used to scalarize texture statistics.
• Data compression involved Gaussian-weighted mean and Principal Component Analysis (PCA) to yield eight key variables.
• A Gated Recurrent Unit (GRU) model was employed for combustion process regression.

Main Results:

• The extracted eight variables effectively characterized O2 and NO contents in flue gas.
• The GRU model achieved a Mean Absolute Percentage Error (MAPE) of 7.5% for O2 and 10.2% for NO.
• The proposed Gabor-GLCM method significantly reduced prediction errors compared to conventional PCA and GLCM+PCA methods.

Conclusions:

• The developed Gabor-GLCM based flame feature extraction is suitable for industrial combustion systems.
• This approach offers improved accuracy for monitoring and controlling boiler combustion.
• The method provides a foundation for advanced analysis and real-time control of industrial boilers.