Image and infrared backscattering feature fusion for high-accuracy intelligent detection of flotation tailings coal ash content

The stability and optimization of coal flotation processes are critically constrained by the low accuracy of tailings ash content detection, which represents a persistent bottleneck for efficient closed-loop control. To overcome the limitations of laborious, delayed off-line assays and the insufficient accuracy of current online methods, this study introduces a novel intelligent framework for high-precision, real-time ash content monitoring. We first designed and constructed a custom multi-sensor data acquisition system capable of simultaneously capturing visible light images and infrared backscattering signals from the tailings slurry. We elucidated the quantitative relationship between multi-dimensional visual features extracted from the images and the actual ash content, rigorously quantifying the significant impact of slurry concentration on these features – a crucial factor often overlooked in prior visual detection approaches. Leveraging these insights, we innovatively propose a multi-source data-driven detection methodology. This method employs image feature extraction and convolutional neural networks for deep image feature extraction, integrates complementary Infrared Backscattering Light signals, and utilizes a dedicated intelligent feature fusion framework to achieve accurate ash content prediction. The outcomes of this research provide a robust solution for achieving high-precision in situ monitoring and enhanced quality control of ash content within flotation circuits.