基于改进YOLOv11的露天矿复杂背景下小目标检测

朱永军; 蔡光琪; 韩进; 缪燕子; 马小平; 焦文华

doi:10.13272/j.issn.1671-251x.2025020018

基于改进YOLOv11的露天矿复杂背景下小目标检测

朱永军^1,,
蔡光琪²,
韩进²,
缪燕子^{1, 3},
马小平¹,
焦文华^{1, 3, ,}

1.
中国矿业大学信息与控制工程学院，江苏徐州　221116
2.
中煤平朔集团有限公司，山西朔州　036006
3.
中国矿业大学人工智能研究院，江苏徐州　221116

基金项目:

国家自然科学基金项目（62473370）。

详细信息

作者简介:
朱永军（1998—），男，安徽六安人，硕士研究生，研究方向为计算机视觉，E-mail：yj.zhu@cumt.edu.cn

通讯作者:
焦文华（1975—），男，北京人，研究员，博士，研究方向为机器视觉与感知及时间序列分析预测，E-mail:wjiao@cumt.edu.cn。

中图分类号: TD67/804
计量
- 文章访问数: 104
- HTML全文浏览量: 18
- PDF下载量: 35
出版历程
- 收稿日期: 2025-02-12
- 修回日期: 2025-04-26
- 网络出版日期: 2025-05-07
- 刊出日期: 2025-04-14

Small object detection in complex open-pit mine backgrounds based on improved YOLOv11

ZHU Yongjun^1,,
CAI Guangqi²,
HAN Jin²,
MIAO Yanzi^{1, 3},
MA Xiaoping¹,
JIAO Wenhua^{1, 3, ,}

1.
School of Information and Control Engineering, China University of Mining and Technology, Xuzhou 221116, China
2.
China Coal Pingshuo Group Co., Ltd., Shuozhou 036006, China
3.
Artificial Intelligence Research Institute, China University of Mining and Technology, Xuzhou 221116, China

摘要

摘要:
露天矿小目标检测任务面临视角广、检测距离远导致目标成像小的挑战，现有目标检测模型存在图像逐层下采样操作引发的特征衰减问题。针对该问题，提出了一种改进YOLOv11模型，并将其用于露天矿复杂背景下小目标检测。改进YOLOv11模型通过引入鲁棒特征下采样（RFD）模块替换跨步卷积下采样模块，有效保留了小目标的特征信息；设计了小目标特征增强颈部（STFEN）网络替代原有特征金字塔结构的颈部网络，在模型颈部引入跨阶段部分融合模块，整合来自不同层级的特征图；将原有的CIoU损失函数替换为Powerful−IoU（PIoU）损失函数，解决了训练过程中锚框膨胀问题，使模型快速精准聚焦小目标。在露天矿区小目标数据集上的实验结果表明：① RFD模块使模型参数量减少的同时mAP提升了1.5%；STFEN网络虽使模型参数量有所增加，但mAP提升了2.2%；PIoU损失函数在未改变模型参数量及每秒浮点运算次数的前提下使mAP提升了1.7%；三者联合应用最终使模型mAP提升了3.9%。② 改进YOLO11模型在保持较高推理速度的同时实现了精度提升，其mAP较YOLOv5m，YOLOv8m，YOLOv11m和RtDetr−L分别提高了2.6%，1.5%，0.9%和2.2%，且模型参数量更小，易于边缘部署。
- 露天矿 /
- 小目标检测 /
- YOLOv11 /
- 鲁棒特征下采样 /
- 小目标特征增强颈部 /
- PIoU损失函数
Abstract:
Small object detection in open-pit mines faces challenges such as wide viewing angles and long detection distances, which result in small target imaging. Existing object detection models suffer from feature attenuation caused by progressive image downsampling operations. To address this issue, an improved YOLOv11 model was proposed and applied to small object detection under complex backgrounds in open-pit mines. The improved YOLOv11 model introduced a Robust Feature Downsampling (RFD) module to replace the stride convolution downsampling module, effectively preserving the feature information of small objects. A Small Target Feature Enhancement Neck (STFEN) network was designed to replace the original feature pyramid structure in the neck, incorporating a cross-stage partial fusion module to integrate feature maps from different levels. The original CIoU loss function was replaced with the Powerful-IoU (PIoU) loss function to solve the anchor box expansion issue during training, enabling the model to rapidly and accurately focus on small targets. Experimental results on a small object dataset from open-pit mining areas showed that: ① the RFD module reduced model parameters while increasing mAP by 1.5%. Although the STFEN network increased the number of parameters, it improved mAP by 2.2%. The PIoU loss function improved mAP by 1.7% without changing the number of parameters or FLOPs. The combination of all three led to a total mAP improvement of 3.9%. ② The improved YOLOv11 model achieved higher accuracy while maintaining a high inference speed, with mAP improvements of 2.6%, 1.5%, 0.9%, and 2.2% over YOLOv5m, YOLOv8m, YOLOv11m, and RtDetr-L, respectively, and with fewer parameters, making it more suitable for edge deployment.
- open-pit mine /
- small object detection /
- YOLOv11 /
- robust feature downsampling /
- small target feature enhancement neck /
- PIoU loss function

HTML全文

图 1 改进YOLOv11模型结构

Figure 1. Structure of improved YOLOv11 model

下载: 全尺寸图片幻灯片

图 2 RFD模块结构

Figure 2. Structure of RFD module

下载: 全尺寸图片幻灯片

图 3 SPDConv模块结构

Figure 3. Structure of SPDConv module

下载: 全尺寸图片幻灯片

图 4 CSPF模块结构

Figure 4. Structure of CSPF module

下载: 全尺寸图片幻灯片

图 5 OKM结构

Figure 5. Structure of OKM

下载: 全尺寸图片幻灯片

图 6 DCAM和FSAM结构

Figure 6. Structure of DCAM and FSAM

下载: 全尺寸图片幻灯片

图 7 露天矿作业现场常见小目标

Figure 7. Common small targets in open-pit mine operation sites

下载: 全尺寸图片幻灯片

图 8 不同模型检测结果

Figure 8. Detection results of different models

下载: 全尺寸图片幻灯片

表 1 重载云台相机及其配套设备硬件型号

Table 1 Heavy-duty gimbal camera and its supporting equipment hardware models

名称	型号
云台相机	TIC7862−IRL
云台相机电源	PWR−DC4806
硬盘录像机	NVS−B200−18
工业级光收发器	MTX100−A3K1020
5口工业交换机	MTX100−A5K0050
工业级室外无线AP	ZoneFree5886
防雷器	SMTRJ45/E1000−220 V

下载: 导出CSV

表 2 消融实验结果

Table 2 Ablation experiment results

模型	mAP/%	每秒浮点运算次数/$ {10}^{9} $	处理速度/ （$ \mathrm{帧} \cdot {\mathrm{s}}^{-1} $）	参数量/$ {10}^{6} $个
YOLOv11	75.8	21.3	372.5	9.41
E1	77.3	23.8	233.2	9.26
E2	78.0	43.0	198.4	11.16
E3	77.5	21.3	352.9	9.41
E4	79.0	45.4	208.1	11.01
E5	78.9	43.0	198.0	11.16
E6	78.6	23.8	233.6	9.26
E7	79.7	45.4	174.0	11.01

下载: 导出CSV

表 3 对比实验结果

Table 3 Comparative experiment results

模型	mAP/%	每秒浮点运算次数/$ {10}^{9} $	处理速度/ （$ \mathrm{帧} \cdot {\mathrm{s}}^{-1} $）	参数量/ $ {10}^{6} $个	模型大小/MiB
YOLOv5m	77.1	47.9	232.6	20.86	40.29
YOLOv8m	78.2	78.7	205.3	25.84	49.63
YOLOv11m	78.8	67.7	218.5	20.03	38.66
RtDetr−L	77.5	100.6	54.8	28.45	56.33
改进YOLOv11	79.7	45.4	174.0	11.01	21.43

下载: 导出CSV

参考文献(21)

[1]	翟海燕. 露天煤矿开采安全生产技术研究[J]. 内蒙古煤炭经济,2023(2):97-99. DOI: 10.3969/j.issn.1008-0155.2023.02.034 ZHAI Haiyan. Study on safety production technology of open-pit coal mining[J]. Inner Mongolia Coal Economy,2023(2):97-99. DOI: 10.3969/j.issn.1008-0155.2023.02.034
[2]	TONG Kang,WU Yiquan. Deep learning-based detection from the perspective of small or tiny objects:a survey[J]. Image and Vision Computing,2022,123. DOI: 10.1016/j.imavis.2022.104471.
[3]	秦学斌,薛宇强,景宁波,等. 露天煤矿自动驾驶矿卡前障碍物检测算法研究[J]. 金属矿山,2025(2):145-151. QIN Xuebin,XUE Yuqiang,JING Ningbo,et al. Study on obstacle detection algorithm in front of autonomous mining truck in open-pit coal mine[J]. Metal Mine,2025(2):145-151.
[4]	LIU Yang,SUN Peng,WERGELES N,et al. A survey and performance evaluation of deep learning methods for small object detection[J]. Expert Systems with Applications,2021,172. DOI: 10.1016/j.eswa.2021.114602.
[5]	江松,孔若男,李鹏程,等. 融合Swin Transformer与CNN的露天矿车前障碍物智能检测算法[J]. 金属矿山,2023(5):228-236. JIANG Song,KONG Ruonan,LI Pengcheng,et al. Intelligent detection algorithm of obstacles in front of open-pit mine cars based on Swin Transformer and CNN[J]. Metal Mine,2023(5):228-236.
[6]	薛小勇,何新宇,姚超修,等. 基于改进YOLOv8n的采掘工作面小目标检测方法[J]. 工矿自动化,2024,50(8):105-111. XUE Xiaoyong,HE Xinyu,YAO Chaoxiu,et al. Small object detection method for mining face based on improved YOLOv8n[J]. Journal of Mine Automation,2024,50(8):105-111.
[7]	阮顺领,鄢盛钰,顾清华,等. 基于多特征融合的露天矿区道路负障碍检测[J]. 煤炭学报,2024,49(5):2561-2572. RUAN Shunling,YAN Shengyu,GU Qinghua,et al. Negative obstacle detection on open pit roads based on multi-feature fusion[J]. Journal of China Coal Society,2024,49(5):2561-2572.
[8]	TAN Mingxing,PANG Ruoming,LE Q V. EfficientDet:scalable and efficient object detection[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition,Seattle,2020:10778-10787.
[9]	XIAO Jinsheng,GUO Haowen,ZHOU Jian,et al. Tiny object detection with context enhancement and feature purification[J]. Expert Systems with Applications,2023,211. DOI: 10.1016/j.eswa.2022.118665.
[10]	RABBI J,RAY N,SCHUBERT M,et al. Small-object detection in remote sensing images with end-to-end edge-enhanced gan and object detector network[J]. Remote Sensing,2020,12(9). DOI: 10.20944/preprints202003.0313.v2.
[11]	曹帅,董立红,邓凡,等. 基于YOLOv7−SE的煤矿井下场景小目标检测方法[J]. 工矿自动化,2024,50(3):35-41. CAO Shuai,DONG Lihong,DENG Fan,et al. A small object detection method for coal mine underground scene based on YOLOv7-SE[J]. Journal of Mine Automation,2024,50(3):35-41.
[12]	LIU Can,WANG Kaige,LI Qing,et al. Powerful-IoU:more straightforward and faster bounding box regression loss with a nonmonotonic focusing mechanism[J]. Neural Networks,2024,170:276-284. DOI: 10.1016/j.neunet.2023.11.041
[13]	LU Wei,CHEN Sibao,TANG Jin,et al. A robust feature downsampling module for remote-sensing visual tasks[J]. IEEE Transactions on Geoscience and Remote Sensing,2023,61. DOI: 10.1109/TGRS.2023.3282048.
[14]	吴一全,童康. 基于深度学习的无人机航拍图像小目标检测研究进展[J]. 航空学报,2025,46(3):181-207. WU Yiquan,TONG Kang. Research advances on deep learning-based small object detection in UAV aerial images[J]. Acta Aeronautica et Astronautica Sinica,2025,46(3):181-207.
[15]	WANG C Y,MARK LIAO H Y,WU Y H,et al. CSPNet:a new backbone that can enhance learning capability of CNN[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops,Seattle,2020:1571-1580.
[16]	CUI Yuning,REN Wenqi,KNOLL A. Omni-kernel network for image restoration[C]. The 38th AAAI Conference on Artificial Intelligence,Vancouver,2024:1426-1434.
[17]	ZHAO Qi,LIU Binghao,LYU Shuchang,et al. TPH-YOLOv5++:boosting object detection on drone-captured scenarios with cross-layer asymmetric transformer[J]. Remote Sensing,2023,15(6). DOI: 10.3390/rs15061687.
[18]	GUI Zhiyong,CHEN Jianneng,LI Yang,et al. A lightweight tea bud detection model based on Yolov5[J]. Computers and Electronics in Agriculture,2023,205. DOI: 10.1016/j.compag.2023.107636.
[19]	WANG Xueqiu,GAO Huanbing,JIA Zemeng,et al. BL-YOLOv8:an improved road defect detection model based on YOLOv8[J]. Sensors,2023,23(20). DOI: 10.3390/s23208361.
[20]	CAO Yukang,PANG Dandan,ZHAO Qianchuan,et al. Improved YOLOv8-GD deep learning model for defect detection in electroluminescence images of solar photovoltaic modules[J]. Engineering Applications of Artificial Intelligence,2024,131. DOI: 10.1016/j.engappai.2024.107866.
[21]	ZHAO Yian,LV Wenyu,XU Shangliang,et al. DETRs beat YOLOs on real-time object detection[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition,Seattle,2024:16965-16974.

施引文献(6)

期刊类型引用(4)

1.	蔡春林，江成玉，苟仁涛，夏勋鹏，王沉，刘文昌. 基于负压衰减效应的抽采长钻孔残余瓦斯分布规律研究. 煤矿安全. 2024(11): 61-68 . 百度学术
2.	张大琦，孙志海，谢宏，朱瑞浩. 煤层瓦斯含量快速测定中煤样现场制备关键技术的实验验证. 华北科技学院学报. 2024(06): 32-36+72 . 百度学术
3.	焦继红，王玉杰，李辉. 煤层瓦斯含量测定负压排渣定点取样应用研究. 煤炭与化工. 2023(08): 95-100 . 百度学术
4.	李鹏，马金魁，许江涛. 千米钻孔水锁现象效果影响及处理技术研究. 机械管理开发. 2022(03): 72-73+76 . 百度学术