Infrared thermal image detection method of stressed sandstone fracture based on deep learning
Hai Sun*, Xinyi Hou, Liqiang Ma, Wenshuang Gao, Kun Wang
Deep Resources Engineering. 2026, 3(1): 100207.
Abstract: Under stress, multiple cracks in the localized damaged zones in mine rock masses will expand and connect, which can lead to the rock mass instability and failure. Rapid, accurate identification of cracks within damaged zones forms a critical foundation for predicting and preventing rock mass fracture instability. To address the need for non-contact precise measurement of mine rock mass fractures, this study proposes an automated method for detecting and identifying fractures in infrared thermal images using deep convolutional neural networks. First, infrared thermal images containing temperature anomalies during sandstone fracturing are captured via infrared thermal imaging. Next, deep learning is used to develop sandstone fracture detection models for infrared thermal images based on the SSD and YOLOv5 algorithms. Compared to the YOLOv5 algorithm, the SSD algorithm detects sandstone fracture regions with higher accuracy and confidence. Building on this, a convolutional attention mechanism is integrated to optimize the SSD algorithm. The optimized algorithm achieves 90.2% detection accuracy for successive difference infrared thermal images, with improved precision in sandstone fracture detection. The research results can provide the development of computer vision-based fracture detection technologies for stressed rock masses.
Highlights:
• A novel methodology integrating infrared thermal imaging and deep learning is proposed.
• The SSD model demonstrates superior fracture localization precision and boundary delineation accuracy in thermal imagery compared to YOLOv5, validated by confidence score analysis.
• The study confirms that successive difference infrared thermal images outperform original infrared thermal images in sandstone fracture detection.
Keywords: Rock mass; Crack; Infrared thermal imaging; Convolutional attention mechanism; Deep learning
Cite: Sun, H.; Hou, X.Y.; Ma, L.Q.; Gao, W.S.; Wang, K., Infrared thermal image detection method of stressed sandstone fracture based on deep learning, Deep Resources Engineering 2026, 3 (1): 100207. https://doi.org/10.1016/j.deepre.2025.100207
Citations: ……
