Characterizing segregation in blast rock piles a deep-learning approach leveraging aerial image analysis

Chengeng Liu, Sihong Liu, Chaomin Shen, Yupeng Gao, Yuxuan Liu

Abstract

Blasted rock material serves a critical role in various engineering applications, yet the phenomenon of segregation—where particle sizes vary significantly along the gradient of a quarry pile—presents challenges for optimizing quarry material storage and handling. This study introduces an advanced image analysis methodology to characterize such segregation of rock fragments. The accurate delineation of detailed rock fragment size distributions was achieved through the analysis of drone-captured imagery, coupled with the application of an enhanced Unet semantic segmentation model integrated with an expansion-based post-processing technique. The quarry slope was stratified into four vertical sections, with the size distribution of each section quantified via ellipsoid shape approximations. Our results disclose pronounced vertical segregation patterns, with finer particles concentrated in the upper slope regions and coarser particles in the lower. Utilizing relative characteristic diameters, we offered insight into the degree of segregation, thereby illustrating the spatial heterogeneity in fragment size more clearly. The techniques outlined in this study deliver a scalable and accurate method for assessing fragment size distribution, with the potential to better inform resource management and operational decisions in quarry management.

Introduction

In civil engineering, blasted rock is a fundamental component for road subgrades, dikes, and concrete production. However, these rock piles are rarely uniform. A natural phenomenon known as segregation occurs during dumping: larger, heavier blocks gain more momentum and settle at the foot of the slope, while finer particles cluster at the crest.

Ignoring this spatial heterogeneity can lead to significant engineering risks, such as using material that is too coarse (creating structural gaps) or too fine (compromising drainage and stability). Traditional "sieve analysis"—physically sorting tons of rock through mesh screens—is labor-intensive, expensive, and often impractical for large-scale quarry slopes. This research presents an automated, non-contact alternative using drone-based computer vision and deep learning.

Data Acquisition and Study Area

The study focused on a chamber blast quarry in Huangshan City, Anhui Province, China. Researchers utilized a DJI Mavic 2 Zoom drone to capture high-resolution (4096 × 3072 pixels) aerial imagery of a 48-meter slope with a 38-degree incline. To analyze vertical segregation, the slope was stratified into four distinct sections (S1 to S4) from top to bottom.

Study site location in Huangshan City and the drone-based image acquisition strategy over the blasted rock dump.

The resulting dataset comprised 660 image patches (512 × 512 pixels) used for training and validating the segmentation models.

Methodology

The core of the proposed workflow is an "Enhanced U-Net" architecture designed to handle the complex, overlapping nature of rock fragments. The methodology incorporates several key innovations:

  1. CARAFE Upsampling: Content-Aware ReAssembly of FEatures (CARAFE) replaces standard bilinear interpolation to better preserve the sharp edges of small rock fragments during the upsampling process.
  2. Ghost-ECA Modules: The integration of GhostConv and Efficient Channel Attention (ECA) reduces the model's parameter count by 16.8% while enhancing its ability to focus on critical visual features.
  3. Dual-Labeling and Post-Processing: The model predicts both the "body" and "boundary" of fragments. An expansion-based post-processing algorithm then grows the body seeds until they meet a boundary, effectively separating touching rocks.
  4. 3D Quantification: 2D segmented areas are converted into 3D volumes using an equivalent ellipsoid model, enabling the generation of volume-based Particle Size Distribution (PSD) curves.
The comprehensive automated workflow for image preparation, deep learning segmentation, and fragment size quantification.

Key Results and Insights

The enhanced model achieved a Mean Intersection over Union (mIoU) of 72.18%, significantly outperforming the baseline U-Net. This precision was vital for accurate fragment delineation, particularly at boundaries.

The analysis confirmed a stark vertical segregation gradient. The average particle size increased from 5.39 cm at the top section (S1) to 13.33 cm at the bottom (S4). Interestingly, the study found that $d_{10}$ (the diameter where 10% of the material is finer) is a more sensitive indicator of segregation than the median diameter ($d_{50}$). The relative characteristic diameter for $d_{10}$ increased by 379% from top to bottom, compared to a 159% increase for $d_{90}$, highlighting that smaller particles are more susceptible to the segregation process.

Conclusion

This research demonstrates that a deep-learning-based aerial analysis workflow can accurately characterize rock segregation in quarries. While the system currently faces limitations with extremely fine particles (smaller than 1 cm) and surface occlusion, it offers a scalable and efficient alternative to manual sieving. Future work may incorporate depth data from LiDAR or multi-view stereo to further refine 3D volume estimations and account for complex surface topographies.

Summary updated

Abstract

Blasted rock material serves a critical role in various engineering applications, yet the phenomenon of segregation-where particle sizes vary significantly along the gradient of a quarry pile-presents challenges for optimizing quarry material storage and handling. This study introduces an advanced image analysis methodology to characterize such segregation of rock fragments. The accurate delineation of detailed rock fragment size distributions was achieved through the analysis of drone-captured imagery, coupled with the application of an enhanced Unet semantic segmentation model integrated with an expansion-based post-processing technique. The quarry slope was stratified into four vertical sections, with the size distribution of each section quantified via ellipsoid shape approximations. Our results disclose pronounced vertical segregation patterns, with finer particles concentrated in the upper slope regions and coarser particles in the lower. Utilizing relative characteristic diameters, we offered insight into the degree of segregation, thereby illustrating the spatial heterogeneity in fragment size more clearly. The techniques outlined in this study deliver a scalable and accurate method for assessing fragment size distribution, with the potential to better inform resource management and operational decisions in quarry management.
BibTeX
@misc{liu2024characterizing,
  author = {Chengeng Liu and Sihong Liu and Chaomin Shen and Yupeng Gao and Yuxuan Liu},
  title = {{Characterizing segregation in blast rock piles a deep-learning approach leveraging aerial image analysis}},
  year = {2024},
  eprint = {2406.04149},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2406.04149},
}
Figure