Although coal remains essential for global energy production, the inevitable self-heating within coal piles poses significant economic and safety risks. Existing thermal anomaly detection methods based on low-resolution (LR) infrared (IR) imaging often fail to identify subtle thermal signatures, especially across large storage areas. To address these limitations, we systematically implemented guided depth super-resolution (GDSR), specifically tailored for coal pile monitoring. Due to the scarcity of realistic field data, we developed a novel testbed using gravel piles embedded with controlled heaters to simulate real-world conditions. Multi-modal images were captured using synchronized RGB and IR sensors. We established a framework evaluated on both synthetic testbed datasets and real-world coal-field data to ensure practical applicability. In this study, we propose a novel Edge-Guided Thermal Super-Resolution (EGTSR) network, benchmarked against representative methods such as the efficient Discrete Cosine Transform Network (DCTNet) and the accuracy-oriented Structure-Guided Network (SGNet). Crucially, EGTSR integrates a segmentation-guided alignment module to explicitly correct spatial discrepancies, which enables the network to maximize the utilization of gradient-based structural guidance from RGB images. Experimental results demonstrate that the proposed EGTSR outperforms existing methods, achieving up to a 31% reduction in root mean squared error compared to the baseline. This AI-based framework significantly improves thermal imaging resolution and detection accuracy, offering safer and more efficient coal storage management.
Jung et al. (Tue,) studied this question.