Illumination-Invariant Flotation Froth Color Measuring via Wasserstein Distance-Based CycleGAN With...

Area of Science:

• * Mineral Processing and Materials Science
• * Computer Vision and Machine Learning
• * Industrial Process Monitoring

Background:

• * Froth color is a critical indicator for flotation concentrate grade.
• * Inconsistent illumination in industrial settings hinders accurate froth color measurement.
• * Existing methods struggle with time-varying, multi-source lighting interferences.

Purpose of the Study:

• * To develop an illumination-invariant method for robust froth color measurement.
• * To maintain spatial structure and texture integrity during color translation.
• * To enable reliable online monitoring of flotation concentrate grade.

Main Methods:

• * Proposed a Wasserstein distance-based structure-preserving CycleGAN (WDSPCGAN).
• * Employed two generative adversarial networks (GANs) with shared generators and individual discriminators.
• * Utilized an improved U-net-like full convolution network for spatial structure-preserved color translation.

Main Results:

• * WDSPCGAN successfully achieved illumination-invariant froth color features across diverse lighting conditions.
• * The method preserved the structural and textural integrity of froth images.
• * Validated on benchmark datasets and an industrial bauxite flotation process.

Conclusions:

• * WDSPCGAN offers a robust solution for illumination-invariant froth color analysis.
• * The model's online adaptability enhances its potential for real-time flotation monitoring.
• * This technology can significantly improve the accuracy of online concentrate grade estimation.