How Drones and AI Measure Coal Stockpiles?

Imagine needing to know how much coal is in a massive pile without getting close. In the past, people used walking wheels, total stations or laser scanners, which were slow and risky. Now drones (UAS) are changing the game. In Ohio, two drone types flew over a coal depot: a quadcopter with LiDAR, which "sees" the ground with lasers, and a fixed‑wing with multispectral cameras. Researchers tested three ways to calculate volume. The first uses LiDAR data, considered the reference. The second exploits Structure‑from‑Motion (SfM) photogrammetry to create a digital surface model (DSM) and a digital terrain model (DTM). The third combines the SfM DSM with an interpolated DTM via kriging, a statistical technique. To outline the pile they built an algorithm that merges slope thresholds, near‑infrared filters and the Canny edge detector. The results? The two SfM versions produced errors of only 147 m³ (RMSE) and a mean percentage error of 2 %, almost identical to LiDAR. They also found that a resolution between 0.06 m and 5 m keeps accuracy, while at 10 m the error jumps to 12 %. In practice, drones with photogrammetry and automated analysis offer a fast, cheap, and precise way to measure coal piles, as well as mines, agricultural storage or forests.

The SfM‑intDTM approach leverages 3D reconstruction from overlapping images, generating a dense point cloud. Kriging, a geostatistical interpolation method, estimates the DTM based on spatial correlation of points, reducing errors from irregular surfaces. The NIR filter enhances discrimination between coal and vegetation, as coal strongly absorbs near‑infrared while vegetation reflects it. The Canny algorithm, originally designed for edge detection in 2D images, has been adapted to identify stockpile contours in the DSM by combining elevation gradients and intensity thresholds. Sensitivity to resolution stems from the trade‑off between point density (more points = finer detail) and noise: at resolutions finer than 0.06 m, noise can obscure details, while at 10 m the loss of topographic detail significantly impacts calculated volume. Moreover, the difference between RTK (real‑time corrections) and PPK (post‑processing) affects absolute positional accuracy, with RTK typically better for critical missions, but PPK offers greater operational flexibility. Integrating these methods allows balancing cost, processing time, and accuracy, making the solution scalable from single piles to entire industrial basins.