1. Coal type identification with application result quantification based on deep-ensemble learning and image-encoded reflectance spectroscopy.
- Author
-
Yan, Zelin, Xiao, Dong, Sun, Hui, Zhang, Lizhi, and Yin, Lingyu
- Subjects
- *
ANTHRACITE coal , *BITUMINOUS coal , *REFLECTANCE spectroscopy , *SPECTRAL imaging , *POWER resources - Abstract
• Explored the feasibility of processing image-encoded coal reflectance spectroscopy using deep-ensemble learning method. • Proposed a novel diffusion model for generating a large quantity of spectral images, overcoming the challenge of data acquisition in mining environments. • For the task of identifying anthracite coal, bituminous coal and lignite, a deep-ensemble learning model IRDE-Net is proposed with the advantages of high accuracy and speed. • Based on the comparison of the identification results of the different methods, the costs and pollution caused by the different methods in the coal application process were quantified. Accurate coal type identification is essential for efficient coal utilization. This study proposes a coal type identification method based on image-encoded reflectance spectra and deep-ensemble learning. Firstly, we encode the reflectance spectra into images using methods based on Gramian Angular Summation Field (GASF), Gramian Angular Difference Field (GADF), Markov Transition Field (MTF), and Recurrence Plot (RP). These encoding methods facilitate the establishment of relationships between spectral bands, reflectance properties, and coal types. Subsequently, a novel diffusion model is introduced for learning from the spectral images and generating a substantial number of spectral images to augment the dataset. Finally, this study proposes a deep-ensemble learning model, IRDE-Net, designed to effectively learn both global features and local key features of spectral images. The experimental results demonstrate that IRDE-Net achieves optimal identification performance on spectral images encoded using the MTF method. The model attained an Accuracy of 94.29%, Precision of 94.36%, Recall of 94.29%, and F1 score of 94.31%. This study conducts a comparison with various advanced methods, quantifying the costs and pollution resulting from misidentification in the coal application process by different methods. It clearly demonstrates the significant contribution of the approach proposed in this study to the practical application of energy resources. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF