Answers are grounded in this paper and cite their sources.
Evaluates high-precision anomaly detection in rock fractures, indicating enhanced monitoring techniques.
Infrared radiation anomaly zones generated during rock fracture serve as vital indicators for geotechnical engineering safety monitoring. However, their accurate detection and segmentation under limited-sample scenarios remain challenging. Current end-to-end instance segmentation models heavily depend on costly pixel-level annotations and demonstrate limited generalization in such settings, particularly in low-contrast infrared images containing fine cracks, where inherent feature competition between detection and segmentation tasks leads to high missed detection rates and low segmentation precision. Traditional manual measurement methods are prone to subjective errors and cannot meet the accuracy and efficiency demands of engineering applications. To address these issues, this study constructs a dedicated limited-sample benchmark dataset for infrared radiation anomaly detection in rock fracture, and systematically evaluates eleven object detection models and eight segmentation algorithms under limited-sample constraints, establishing a reference for model selection and performance comparison in this field. Furthermore, we propose a customized cascaded detection–filtering–segmentation framework tailored to the aforementioned limited-sample scenarios, which significantly reduces reliance on pixel-level segmentation annotation data while achieving high-precision detection and segmentation of infrared radiation anomaly zones. Based on the segmentation outputs, we integrate synchronously acquired mechanical data and acoustic emission monitoring data to construct a cross-modal correlation analysis framework linking visual characteristics with mechanical–acoustic multi-source responses. Not only does this study provide an effective technical solution for infrared radiation anomaly recognition under limited-sample conditions, but the proposed cascaded framework also offers a valuable reference paradigm for visual detection and segmentation tasks in related engineering contexts. The developed model is ultimately validated in real-world downstream mining scenarios.
No takes yet. Share an insight, caveat, or question.
Shang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: