A new classification algorithm for low concentration slurry based on machine vision

Area of Science:

• Materials Science
• Computer Vision
• Chemical Engineering

Background:

• Accurate classification of low concentration coal slurry is crucial for efficient resource utilization and process control.
• Traditional methods for slurry analysis can be time-consuming and may lack precision.
• Advancements in machine vision offer potential for rapid and accurate slurry characterization.

Purpose of the Study:

• To develop and evaluate a machine vision-based system for accurate classification of low concentration coal slurry.
• To optimize image acquisition parameters for coal slurry collection.
• To create a robust classification model using deep learning techniques.

Main Methods:

• Utilized machine vision and orthogonal experiments (L9(3^4)) to determine optimal image acquisition parameters (exposure, thickness, light intensity).
• Developed a low concentration classification model involving image acquisition, data augmentation, and dataset partitioning.
• Employed Deep Convolutional Generative Adversarial Network (DCGAN) for image generation and Convolutional Neural Network (CNN) for classification.

Main Results:

• Optimal image parameters identified: exposure 10, slurry thickness 7 cm, light intensity 5×10^4 lux.
• DCGAN achieved high similarity (SSIM 0.9381, pHash 0.9375) in image generation.
• CNN model demonstrated over 95% accuracy on training and validation sets for coal slurry concentration prediction.
• CNN model achieved high performance metrics on the test set, with accuracy reaching 94% and F1 scores up to 1.000.

Conclusions:

• The developed machine vision system effectively classifies low concentration coal slurry.
• The optimized image acquisition and CNN-based classification model show excellent performance and accuracy.
• This approach offers a promising solution for real-time, precise detection of coal slurry concentration.