benchmark for rock blasting

Description

The dataset is unique in its comprehensiveness: it covers the blast pile, loading, truck/wagon haulage, and conveyor stages of the mining process. This allows for the training and testing of algorithms across a wide range of operational environments — from the mine pit to the processing plant "Open pits": Images were captured via a tripod-mounted motorized turntable system to ensure controlled rotation. The optical setup consisted of a 5MP Dalsa Genie Nano M2590NIR grayscale camera (1-inch sensor) paired with an LMZ25300M3P-IR 12x zoom lens. Conditions: High environmental variability was achieved by conducting 10 field experiments across all four seasons and diverse weather patterns (sunny, cloudy, and snowy). The polygon annotations are provided in COCO JSON. Annotation method: Manual. "Rocks asbest": Field samples were collected by a geologist at the Bazhenovskoye deposit (Asbest, Sverdlovsk Oblast, Russia). For oversized geological sections, a multi-frame stitching approach was used, where segments were captured separately to maintain focus and detail. Conditions: Autumn and winter conditions, including sunny, cloudy, rainy, and snowy weather. Annotation method: Automated segmentation via SAM "Rocks UAV": High-resolution aerial photography was performed using DJI drone equipment at a consistent operational altitude of 15–20 meters to provide a nadir view of the fragmentation. Conditions: Winter deployment, characterized by the presence of snow covering the blasted rock. Annotation method: Automated segmentation via SAM. "Bucket excavation": Video stream extraction from an onboard camera located inside the excavator cabin, capturing the loading process into railcars. The video was sampled at a fixed interval of 500 ms to capture temporal progression. Conditions: Dynamic loading environment with potential occlusions from cabin components. Annotation method: Hybrid ( pre-trained Mask R-CNN, followed by rigorous manual verification to remove false positives like cab debris). "40 70mm":Fixed overhead camera mounted at the Seversky Granite Quarry weigh station. The setup uses a non-standard oblique angle, introducing significant geometric distortion. Conditions: Monitoring of crushed stone (40–70 mm) in truck beds. Annotation method: SAM2.1-assisted segmentation, followed by manual cleaning to exclude non-target objects outside the truck bed. "Unloading a wagon": Proprietary. Acquisition setup: Video frames extracted from handheld mobile device recordings. The footage captures the discharge of limestone from railcars into a receiving hopper. Annotation method: Auto SAM. Avg. instances: 190. "Transport": A Dalsa Genie Nano M2590NIR machine vision camera with a 1-inch sensor, optimized for high sensitivity in the near-infrared (NIR) spectrum. The camera was positioned 900 mm from a 400 mm wide conveyor belt to ensure full-width coverage. Conditions: Stationary conveyor belt at an asbestos processing plant. Annotation method: Manual.

Institutions

• Ural Federal University

  Sverdlovsk Oblast, Yekaterinburg

Categories

Licence