UAV image-based identification and block size analysis of tunnel muck using deep learning

Drill-and-blast tunneling generates large volumes of rock spoil, and efficient and accurate characterization of block size distribution is critical for enhancing resource utilization. However, traditional manual measurement methods are time-consuming and limited in spatial coverage. To address these limitations, a UAV-based image acquisition approach combined with deep learning was employed to automatically extract rock block contours. The block size distribution and resource utilization potential of tunnel spoil were investigated using a case study from Dangshun Tunnel spoil yard in Qinghai Province, China. The results show that the median block diameter is approximately 300 mm, with maximum sizes exceeding 1 500 mm. Both the size and morphology of rock blocks exhibit systematic spatial variation along the slope, with particle size gradually increasing from fine to coarse from the slope crest to the slope toe, while block shape transitioning progressively from subrounded to angular. In recently dumped areas, rock blocks largely preserve their original blast-induced fragmentation features, with a relatively high proportion of large fragments exceeding 1 000 mm in size. This suggests that optimization of charging structure and blasthole spacing is necessary. Based on gravity-induced sorting characteristics, a zoned resource utilization strategy is proposed in which medium- and fine-grained materials at the crest and slope face zones can be directly used as fill or road construction materials, whereas coarse blocks accumulated at the slope toe should be crushed for recycling. The findings provide a technical basis for efficient resource utilization of tunnel spoil and optimization of blasting design in tunnel engineering.