Image Deep Learning-Driven Granularity Effect Correction: A Novel Approach to Improve the Accuracy of NIRS-XRF Coal Quality Analysis

Abstract

As a vital global energy resource, coal quality directly impacts the efficiency of power generation and environmental sustainability. Consequently, fast and accurate coal quality analysis is essential to promote the clean and efficient use of coal. However, the combined spectral analysis of coal quality using near-infrared spectroscopy (NIRS) and X-ray fluorescence (XRF) often suffers from granularity effects, leading to instability and inaccuracies in the analysis results. This study proposes an innovative granularity effect correction method driven by deep learning and image processing. By integrating the segment anything model (SAM) for image segmentation with convolutional neural networks (CNNs), this approach significantly enhances the accuracy of NIRS-XRF coal quality analysis, providing a novel solution to mitigate the interference of granularity effects in quantitative spectral analysis. The methodology begins with the SAM model to precisely segment larger particles in microscopic images of coal samples, generating binarized mask images that effectively capture the sample's granularity characteristics. Subsequently, CNNs deep learning is employed to analyze the intrinsic relationship between these mask images and ash measurement errors, thereby establishing a robust granularity effect correction model. Experimental validation demonstrates that this novel method markedly improves the accuracy and consistency of ash content predictions, reducing the standard deviation (SD) and root mean square error (RMSE) by approximately 20%. By integrating cutting-edge image processing and deep learning technologies, this study not only offers an effective new solution for addressing the impact of granularity effects in traditional coal quality spectral analysis but also holds promise for application in the complex sample analysis of other resources and materials. This innovation is expected to enhance the automation and efficiency of online analysis systems, driving technological advancement in the industry.