Deep learning-based image segmentation for highwall stability monitoring in open pit mines

Current highwall monitoring methods are effective in tracking mass slope movements but often fail to adequately assess the dynamic impact of mining operations on highwall stability, particularly concerning the impact of rock fragmentation techniques, which can lead to structural instability and hazardous rockfall events. This study proposes a novel approach that leverages high-resolution imagery and deep learning-based U-Net segmentation model for automatic detection of cracks and fractures on highwalls in open pit mines. The developed methodology involves five key phases: obtaining high-quality imagery, pre-processing the acquired data, utilizing the U-Net image segmentation model, generating segmentation masks, and identifying cracks and fractures. Robustness testing was then conducted, comparing three U-Net model training configurations and the canny edge detector for crack segmentation. The results revealed that the model trained on a combination of original and augmented images achieved superior performance, boasting a 97 % accuracy, an intersection over union (IoU) of 0.77, and identifying cracks and fractures closely resembling the ground truth. This innovative approach not only enhances the efficiency of highwall monitoring but also minimizes the risk of hazardous incidents, thereby significantly improving safety standards and the overall impact on operational effectiveness in open-pit mining operations.