In the intelligentization process of open-pit mines, small target detection for unmanned mining trucks is confronted with technical challenges caused by complex surface conditions and extreme multi-scale target distributions. An enhanced lightweight small target detection algorithm, PCSF-RTDETR, was proposed based on the RT-DETR frame work, with specific optimization for open-pit mining environments. Firstly, the lightweight PConv-Block module was integrated into the backbone feature extraction network, which enables effective capture of small target characteristics while significantly reducing computational complexity. Secondly, a cascaded group attention mechanism was introduced into the intra-scale feature interaction module, through which the processes of feature focusing and noise suppression were collaboratively optimized. Furthermore, through the integration of a dynamic upsampler, multi-scale feature fusion, and the Slimneck architecture, the Slimneck-DSF cross-scale feature fusion framework was proposed. The adaptability was enhanced to multi-scale targets while reducing the computational load of the network. Finally, the Focaler-GIoU loss function was adopted, which incorporates a dynamic focusing mechanism. This mechanism differentially adjusts the weight allocation for samples with varying difficulty levels, thereby significantly improving the accuracy of small target detection. Experimental results demonstrate that, compared with the baseline RT-DETR model, PCSF-RTDETR achieves a 31.5% reduction in parameter count and a 23.4% reduction in FLOPs, along with a 4.7% improvement in detection accuracy and an 8.2% improvement in detection speed. Notably, the mAP50 metric reaches 92.5%, which conclusively validates the effectiveness of the proposed PCSF-RTDETR-based lightweight small target detection method for open-pit mining scenarios.
WANG Peng-hao, XIE Li-rong, BIAN Yi-fan, AN Yi, YANG Zhi-yong, HUANG De-qi.
PCSF-RTDETR: A Lightweight Small Target Detection Algorithm for Open-pit Mines[J].
Science Technology and Engineering, 2026, 26(13): 5604-5615 DOI:10.12404/j.issn.1671-1815.2503631
式中:IoUfocaler为重构后的Focaler-IoU值;IoU为原始交并比,参数d和u的取值范围介于0~1,通过调节这两个参数,可动态控制IoUfocaler对不同质量样本的关注强度;B与Bgt分别为预测框(predicted box)和真实标注框(ground truth box)。其损失函数L定义如下。
WangS. Effectiveness of traditional augmentation methods for rebar counting using UAV imagery with Faster R-CNN and YOLOv10-based transformer architectures[J]. Scientific Reports, 2025, 15(1): 33702.
[2]
FigueiredoA, AmaralJ, MendesM, et al. Experimental dataset of video and radar detection for cooperative perception in urban environment[J]. Data in Brief, 2025, 63: 112091.
[3]
WangY Z, NiuP H, GuoX Y, et al. Single shot multibox detector with deconvolutional region magnification procedure[J]. IEEE Access, 2021, 9: 47767.
DingLina, YuanLiying. Improved pedestrian detection method for small targets in dense scenes based on YOLOv8[J]. Science Technology and Engineering, 2025, 25(28): 12088-12097.
[6]
ChenJ, LiuN, SunH, et al. Freq-DETR: frequency-aware transformer for real-time small object detection in unmanned aerial vehicle imagery[J]. Expert Systems With Applications, 2026, 298: 129710.
[7]
ZhaoY, LüW Y, XuS L, et al. Detrs beat YOLOs on real-time object detection[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2024: 16965-16974.
LiuGuangwei, LeiJian, GuoZhiqing, et al. Cross-modal obstacle detection method for open-pit mine[J]. Coal Science and Technology, 2025, 53(11): 327-340.
[10]
WangY, YangZ, ZhaoY, et al. FA-YOLO: multi-scale feature fusion for spectral image object detection in complex scenes[J]. Multimedia Systems, 2025, 31(6): 423.
[11]
GuoG, ZhangL, JiaB, et al. Tool-YOLO: a target detection network based on feature extraction and feature fusion[J]. Multimedia Systems, 2025, 31(6): 424.
[12]
PanJ, SunD, ZhangJ, et al. Dual convolutional neural networks for low-level vision[J]. International Journal of Computer Vision, 2022, 130(6): 1440-1458.
[13]
WangW, DaiJ, ChenZ, et al. Internimage: exploring large-scale vision foundation models with deformable convolutions[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. Vancouver: IEEE, 2023: 14408-14419.
ChenShuang, WangRunhao, LuoLinqi, et al. Salient object detection algorithm with multi-scale feature information fusion[J]. Science Technology and Engineering, 2025, 25(23): 9919-9926.
[16]
ZhangM, BaiH, ZhangJ, et al. Rkformer: Runge-kutta transformer with random-connection attention for infrared small target detection[C]// Proceedings of the 30th ACM International Conference on Multimedia. Lisbon: ACM, 2022: 1730-1738.
[17]
ChenY, LiuZ, ZhangL, et al. MFFNet: a lightweight multi-feature fusion network for UAV infrared object detection[J]. The Egyptian Journal of Remote Sensing and Space Sciences, 2024, 27(2): 268-276.
[18]
ChenD, MiaoD. Control distance IoU and control distance IoU loss for better bounding box regression[J]. arXiv: 2103.11696, 2023.
JinZiyue, LiHaitao, YinHaichen, et al. YOLO11n-seg-RF: an improved lightweight rock fracture detection and segmentation algorithm based on YOLO11n-seg[J/OL]. Acta Scientiarum Naturalium Universitatis Pekinensis[2026-02-02]. DOI: 10.13209/j.0479-8023.2025.063.
[21]
ChenJ, KaoS, HeH, et al. Run, don’t walk: chasing higher FLOPS for faster neural networks[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 12021-12031.
ChengQing, YanHengzhi, LiYandong. Lightweight detection method of small UAVs on improved YOLOv8n[J]. Science Technology and Engineering, 2025, 25(28): 12098-12107.
[24]
LiuX, PengH, ZhengN, et al. EfficientVIT: memory efficient vision transformer with cascaded group attention[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver: IEEE, 2023: 14420-14430.
[25]
ChenX, YinH. DCC-DETR: a real-time lightweight gesture recognition network for home human-robot interaction[J]. Journal of Real-Time Image Processing, 2025, 22(5): 184.
[26]
LiuW, LuH, FuH, et al. Learning to upsample by learning to sample[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Vancouver: IEEE, 2023: 6027-6037.
[27]
LiH, LiJ, WeiH, et al. Slim-neck by GSConv: a lightweight-design for real-time detector architectures[J]. Journal of Real-Time Image Processing, 2024, 21(3): 62.
[28]
KangM, TingC, TingF, et al. ASF-YOLO: a novel YOLO model with attentional scale sequence fusion for cell instance segmentation[J]. Image and Vision Computing, 2024, 147: 105057.
[29]
MaS, LiR, HuH. Train track fastener defect detection algorithm based on MGSF-YOLO[J]. The Journal of Supercomputing, 2025, 81(3): 494.
[30]
LiY, LiZ, TengS, et al. AutoMine: an unmanned mine dataset[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans: IEEE, 2022: 21308-21317.
ChengShuoqi, YaermaimaitiYilihamu, XieLirong, et al. Research on lightweight obstacle detection algorithm for unmanned mining trucks[J]. Coal Science and Technology, 2025, 53(7): 263-274.