Integrated and improved YOLOv8 and triangulated network method for extracting indicators in open-pit mine mining areas
-
摘要:
基于深度学习的露天矿山遥感影像研究为露天矿山采场的快速识别与提取提供了方向，但在露天矿山的实际应用仍局限于识别阶段，存在露天矿山边界提取不准确、模型训练时样本分布不平衡等问题。针对上述问题，提出了一种整合改进YOLOv8与三角网的露天矿山采场指标提取方法。在YOLOv8的基础上进行以下改进，得到Mine−YOLO：添加高效多尺度注意力（EMA）模块，以提高模型对矿山采场边界细节的识别与分割精度；添加全局注意力机制（GAM）模块，从全局尺度保留露天矿山采场特征数据，提高采场目标识别精度；采用Focaler−IoU损失函数优化，增强模型对正样本的区分能力。根据无人机获取的露天矿山数字高程模型（DEM）数据，结合Mine−YOLO模型进行识别与分割处理，获取露天矿山采场区域DEM影像，并自动建立不规则三角网，实现对露天矿山采场面积、体积和采深的精确定量监测。实验结果表明，Mine−YOLO模型在采场识别与分割方面的平均精度均值分别达0.942和0.865，具有较高的识别精度和较好的分割效果。实际应用结果表明，基于Mine−YOLO模型提取的采场数据与传统测量值相差不大，平均面积误差为5.8%，平均体积误差为4.9%，最小采深误差仅为0.2%。
Abstract:The research on remote sensing imagery of open-pit mines based on deep learning has provided a direction for the rapid identification and extraction of open-pit mining areas. However, its practical application in open-pit mining is still limited to the recognition stage, with issues such as inaccurate boundary extraction and unbalanced sample distribution during model training. To address these issues, an improved method for extracting mining field indicators by integrating YOLOv8 with a triangulated network was proposed. Based on YOLOv8, the following improvements were made to obtain Mine-YOLO: the addition of an Efficient Multi-Scale Attention (EMA) module to enhance the model's recognition and segmentation accuracy of mining field boundaries; the inclusion of a Global Attention Mechanism (GAM) module to retain open-pit mining field feature data at a global scale, improving target recognition accuracy; and the optimization of the Focaler-IoU loss function to enhance the model's ability to distinguish positive samples. By utilizing digital elevation model (DEM) data of the open-pit mine obtained by UAVs and combining it with the Mine-YOLO model for recognition and segmentation, DEM images of the mining area were obtained, and a triangulated irregular network was automatically generated. This enabled precise quantitative monitoring of the mining field's area, volume, and depth. Experimental results showed that the Mine-YOLO model achieved average accuracies of 0.942 for recognition and 0.865 for segmentation, demonstrating high recognition accuracy and good segmentation results. Practical application results showed that the mining field data extracted using the Mine-YOLO model were similar to traditional measurement values, with an average area error of 5.8%, an average volume error of 4.9%, and a minimum depth error of only 0.2%.
-
-
![]()
图 3 整合改进YOLO模型与三角网的露天矿山采场指标提取流程
Figure 3. Process of extracting indicators for open-pit mining sites by combining improved YOLO model with a triangular network
![]()
图 9 露天矿山采场区域识别结果对比
Figure 9. Comparison of identification results of open-pit mining area
表 1 实验软硬件配置
Table 1 Experimental software and hardware configuration
参数 配置 CPU Intel（R） Xeon（R） Gold 6348 2.59 GHz RAM 256 GiB，3 200 MHz GPU NVIDIA GeForce RTX 3090，24 GiB cuDNN 8.9.7 CUDA 12.4 Deep learning framework Pytorch−2.2.1+python−3.9 
下载: 导出CSV
表 2 实验参数设置
Table 2 Experimental parameters setting
参数 描述 数值 Epochs 训练的周期数 250 batch 每批次的图像数量 10 imgsz 输入图像的大小 640 workers 数据加载的工作线程数 8 lr0 初始学习率 0.01 momentum 学习动量 0.937 box 盒损失增益 7.5 dfl 类别损失增益 0.5 cls dfl损失增益 1.2
下载: 导出CSV
表 3 改进损失函数性能验证
Table 3 Performance validation of improved loss function
模型 参数量/106个 PD RD $ {{\mathrm{mAP}}}_{{\mathrm{D}}}{@}_{0.5} $ YOLOv7n 36.5 0.734 0.768 0.713 YOLOv8n 3.1 0.725 0.754 0.786 YOLOv7n+
Focaler−IoU40.5 0.722 0.783 0.751 YOLOv8n+
Focaler−IoU4.7 0.737 0.791 0.829
下载: 导出CSV
表 4 消融实验结果
Table 4 Results of ablation experiment
Focaler−IoU EMA GAM $ {P}_{{\mathrm{D}}} $ $ {R}_{{\mathrm{D}}} $ $ {{\mathrm{mAP}}}_{{\mathrm{D}}}{@}_{0.5} $ $ {P}_{{\mathrm{S}}} $ $ {R}_{{\mathrm{S}}} $ $ {{\mathrm{mAP}}}_{{\mathrm{S}}}{@}_{0.5} $ × × × 0.762 0.685 0.792 0.725 0.754 0.786 √ × × 0.777 0.698 0.833 0.753 0.783 0.818 × √ × 0.789 0.817 0.858 0.764 0.797 0.834 × × √ 0.783 0.843 0.824 0.783 0.743 0.783 √ √ × 0.838 0.858 0.873 0.778 0.793 0.815 √ × √ 0.816 0.843 0.838 0.753 0.828 0.788 × √ √ 0.844 0.858 0.872 0.821 0.797 0.852 √ √ √ 0.905 0.863 0.942 0.842 0.803 0.865
下载: 导出CSV
表 5 对比实验结果
Table 5 Comparative experimental results
模型 参数量/
106个$ {P}_{{\mathrm{D}}} $ $ {R}_{{\mathrm{D}}} $ $ {{\mathrm{mAP}}}_{{\mathrm{D}}}{@}_{0.5} $ $ {P}_{{\mathrm{S}}} $ $ {R}_{{\mathrm{S}}} $ $ {{\mathrm{mAP}}}_{{\mathrm{S}}}{@}_{0.5} $ YOLOv8n 3.1M 0.762 0.685 0.792 0.725 0.754 0.786 YOLOv8x 68.3M 0.845 0.732 0.811 0.823 0.785 0.804 YOLOv7n 36.5M 0.803 0.828 0.835 0.734 0.768 0.713 YOLOv5n 1.8M 0.688 0.754 0.653 0.618 0.647 0.638 YOLOv5x 86.8M 0.823 0.772 0.817 0.769 0.676 0.759 Mask R−CNN 42.5M 0.856 0.893 0.843 0.837 0.762 0.854 Mine−YOLO 46.5M 0.905 0.863 0.942 0.842 0.803 0.865
下载: 导出CSV
表 6 露天矿山采场信息提取结果
Table 6 Extraction results of open-pit mining information
矿山 日期（年−月） 采场面积/m2 采场体积/m3 采深/m sE sT vE vT hE hT a 2022−08 479 454 466 253 14 990 096 14 506 500 212.5 212.2 2023−08 489 230 467 800 12 022 578 12 453 500 203.6 203.4 2023−12 485 322 469 287 11 890 552 11 003 600 197.2 197.2 b 2022−08 168 341 165 445 8 503 314 8 649 732 396.4 396.5 2023−04 169 212 166 521 7 544 540 7 523 549 384.2 384.2 2023−08 166 801 167 480 6 688 480 6 623 547 376.5 376.2 c 2022−08 253 208 211 018 11 566 816 9 978 245 109.5 110.3 2023−04 228 319 215 514 9 983 984 9 768 521 101.5 100.4 2023−08 248 341 221 842 9 864 360 9 088 532 93.9 93.9
下载: 导出CSV
-
[1] 张兆长,李岩,南贵军. 河北省非金属露天矿山水平分层式开采探索与实践[J]. 中国矿业,2024,33(6):203-209. DOI: 10.12075/j.issn.1004-4051.20240222 ZHANG Zhaochang,LI Yan,NAN Guijun. Exploration and practice of horizontal slicing mining in non-metal open-pit mines in Hebei Province[J]. China Mining Magazine,2024,33(6):203-209. DOI: 10.12075/j.issn.1004-4051.20240222
[2] TOMBE R,VIRIRI S. Remote sensing image scene classification:advances and open challenges[J]. Geomatics,2023,3(1):137-155. DOI: 10.3390/geomatics3010007
[3] JI Haowei,LUO Xianqi. Implementation of ensemble deep learning coupled with remote sensing for the quantitative analysis of changes in arable land use in a mining area[J]. Journal of the Indian Society of Remote Sensing,2021,49(11):2875-2890. DOI: 10.1007/s12524-021-01430-6
[4] XIANG Jie,CHEN Jianping,SOFIA G,et al. Open-pit mine geomorphic changes analysis using multi-temporal UAV survey[J]. Environmental Earth Sciences,2018,77(6). DOI: 10.1007/s12665-018-7383-9.
[5] LIN C H,WANG Tingyou. A novel convolutional neural network architecture of multispectral remote sensing images for automatic material classification[J]. Signal Processing:Image Communication,2021,97. DOI: 10.1016/j.image.2021.116329.
[6] ZHANG Wei,TANG Ping,ZHAO Lijun. Remote sensing image scene classification using CNN-CapsNet[J]. Remote Sensing,2019,11(5). DOI: 10.3390/rs11050494.
[7] LIU Yanfei,ZHONG Yanfei,FEI Feng,et al. Scene semantic classification based on random-scale stretched convolutional neural network for high-spatial resolution remote sensing imagery[C]. IEEE International Geoscience and Remote Sensing Symposium,Beijing,2016:763-766.
[8] BALANIUK R,ISUPOVA O,REECE S. Mining and tailings dam detection in satellite imagery using deep learning[J]. Sensors,2020,20(23). DOI:10.3390/ s20236936.
[9] XIE Hongbin,PAN Yongzhuo,LUAN Jinhua,et al. Semantic segmentation of open pit mining area based on remote sensing shallow features and deep learning[C]. Big Data Analytics for Cyber-Physical System in Smart City,Shanghai,2020:52-59.
[10] CHEN Tao,ZHENG Xiaoxiong,NIU Ruiqing,et al. Open-pit mine area mapping with Gaofen-2 satellite images using U-net+[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2022,15:3589-3599. DOI: 10.1109/JSTARS.2022.3171290
[11] XIE Hongbin,PAN Yongzhuo,LUAN Jinhua,et al. Open-pit mining area segmentation of remote sensing images based on DUSegNet[J]. Journal of the Indian Society of Remote Sensing,2021,49(6):1257-1270. DOI: 10.1007/s12524-021-01312-x
[12] LIU Yong,LI Cheng,HUANG Jiade,et al. MineSDS:a unified framework for small object detection and drivable area segmentation for open-pit mining scenario[J]. Sensors,2023,23(13). DOI: 10.3390/s23135977.
[13] MENG Xiaoliang,ZHANG Ding,DONG Sijun,et al. Open-pit granite mining area extraction using UAV aerial images and the novel GIPNet[J]. Remote Sensing,2024,16(5). DOI: 10.3390/rs16050789.
[14] 郭栋梁,张延军. 基于轻量化PAM−M−YOLO 模型的煤矸石图像检测[J]. 矿业研究与开发,2024,44(5):220-227. GUO Dongliang,ZHANG Yanjun. Coal gangue image detection based on light-weighted PAM-M-YOLO model[J]. Mining Research and Development,2024,44(5):220-227.
[15] 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.
[16] LIU Yichao,SHAO Zongru,HOFFMANN N. Global attention mechanism:retain information to enhance channel-spatial interactions[EB/OL]. [2024-10-15]. https://arxiv.org/abs/2112.05561v1.
[17] HAO Wangli,REN Chao,HAN Meng,et al. Cattle body detection based on YOLOv5-EMA for precision livestock farming[J]. Animals,2023,13(22). DOI: 10.3390/ani13223535.
[18] REN Zili,WANG Liguan,HE Zhengxiang. Open-pit mining area extraction from high-resolution remote sensing images based on EMANet and FC-CRF[J]. Remote Sensing,2023,15(15). DOI: 10.3390/rs15153829.
[19] ZHANG Hao,ZHANG Shuaijie. Focaler-IoU:more focused intersection over union loss[EB/OL]. [2024-10-15]. https://arxiv.org/abs/2401.10525v1.
[20] 胡佳乐,周敏,申飞. 面向无人机小目标的RTDETR改进检测算法[J]. 计算机工程与应用,2024,60(20):198-206. DOI: 10.3778/j.issn.1002-8331.2404-0114 HU Jiale,ZHOU Min,SHEN Fei. Improved detection algorithm of RTDETR for UAV small target[J]. Computer Engineering and Applications,2024,60(20):198-206. DOI: 10.3778/j.issn.1002-8331.2404-0114
[21] OUYANG Daliang,HE Su,ZHANG Guozhong,et al. Efficient multi-scale attention module with cross-spatial learning[C].The 48th IEEE International Conference on Acoustics,Speech and Signal Processing,Rhodes,2023:1-5.
[22] 易磊,黄哲玮,易雅雯. 改进YOLOv8的输电线路异物检测方法[J]. 电子测量技术,2024,47(15):125-134. YI Lei,HUANG Zhewei,YI Yawen. Improved YOLOv8 foreign object detection method for transmission lines[J]. Electronic Measurement Technology,2024,47(15):125-134.
[23] 李德永,王国法,郭永存,等. 基于CFS−YOLO算法的复杂工况环境下煤矸图像识别方法[J]. 煤炭科学技术,2024,52(6):226-237. DOI: 10.12438/cst.2023-1967 LI Deyong,WANG Guofa,GUO Yongcun,et al. Image recognition method of coal gangue in complex working conditions based on CES-YOLO algorithm[J]. Coal Science and Technology,2024,52(6):226-237. DOI: 10.12438/cst.2023-1967
[24] 谭笑,朱博,王晋扬,等. 一种面向离散高程点的高效TIN生长改进算法[J]. 海军工程大学学报,2023,35(2):25-30. DOI: 10.7495/j.issn.1009-3486.2023.02.004 TAN Xiao,ZHU Bo,WANG Jinyang,et al. An efficient improved TIN growth algorithm for discrete elevation points[J]. Journal of Naval University of Engineering,2023,35(2):25-30. DOI: 10.7495/j.issn.1009-3486.2023.02.004
计量
- 文章访问数: 46
- HTML全文浏览量: 6
- PDF下载量: 8
