基于级联查询-位置关系的输电线路多金具检测方法

doi:10.11996/JG.j.2095-302X.2025020288

图学学报 ›› 2025, Vol. 46 ›› Issue (2): 288-299.DOI: 10.11996/JG.j.2095-302X.2025020288

• 图像处理与计算机视觉 • 上一篇下一篇

基于级联查询-位置关系的输电线路多金具检测方法

翟永杰(), 王璐瑶, 赵晓瑜, 胡哲东, 王乾铭(), 王亚茹

华北电力大学自动化系，河北保定 071003

收稿日期:2024-07-16 接受日期:2024-10-24 出版日期:2025-04-30 发布日期:2025-04-24
通讯作者:王乾铭(1995-)，男，讲师，博士。主要研究方向为电力视觉、知识推理。E-mail：qianmingwang@ncepu.edu.cn
第一作者:翟永杰(1972-)，男，教授，博士。主要研究方向为电力视觉。E-mail：zhaiyongjie@ncepu.edu.cn
基金资助:
国家自然科学基金(U21A20486);河北省自然科学基金(F2021502008);中央高校基本科研业务费专项资金(2023JC006);中央高校基本科研业务费专项资金(2024MS136)

Multi-fitting detection for transmission lines based on a cascade query-position relationship method

ZHAI Yongjie(), WANG Luyao, ZHAO Xiaoyu, HU Zhedong, WANG Qianming(), WANG Yaru

Department of Automation, Huabei Electric Power University, Baoding Hebei 071003, China

Received:2024-07-16 Accepted:2024-10-24 Published:2025-04-30 Online:2025-04-24
First author：ZHAI Yongjie (1972-), professor, Ph.D. His main research interest covers power vision. E-mail：zhaiyongjie@ncepu.edu.cn
Supported by:
National Natural Science Foundation of China(U21A20486);National Natural Science Foundation of Hebei(F2021502008);the Fundamental Research Funds for the Central Universities(2023JC006);the Fundamental Research Funds for the Central Universities(2024MS136)

摘要/Abstract

摘要：

针对输电线路航拍图像的金具目标尺寸小与密集遮挡问题，提出了基于级联查询-位置关系的输电线路多金具检测方法(CQPR)。首先提出了级联稀疏查询模块，通过小尺度特征图上小目标的粗略位置来查询大尺度特征图中的小目标精确位置，提高小目标金具检测的准确性。接着，提出了位置特征关系模块(PRM)，通过利用图像中不同金具之间的位置关系建立PRM，提取金具位置关系，丰富遮挡区域的特征，进而优化了密集遮挡下的金具检测效果。多个基线模型上的实验结果表明，将CQPR应用到基线检测框架时，Faster R-CNN，Cascade R-CNN，Libra R-CNN和Dynamic R-CNN的准确率分别达到82.9%，82.4%，83.7%和77.3%，优于其他先进目标检测模型，对其中小目标金具和存在遮挡情况的金具检测准确率的提高较为明显，推理速度也有一定的提高，同时兼顾定位精度与检测实时性。

关键词: 输电线路, 金具, 深度学习, 目标检测, 小目标, 密集遮挡

Abstract:

To address the challenges posed by the small-size and dense occlusion of fittings in aerial images of transmission lines, a multi-fitting detection method for transmission lines based on a cascade query-position relationship (CQPR) was proposed. Firstly, a cascade sparse query module was proposed to query the precise location of small-size objects in large-scale feature maps using the rough location of small-size objects on small-scale feature maps, thereby improving the accuracy of small-size fitting detection. Then, a positional feature relationship module (PRM) was proposed to optimize the detection results at the occlusion using the positional relationships between different fittings in the image to establish the positional PRM. This enriched the features of occluded areas and optimized fitting detection performance under conditions of dense occlusion. Experimental results on multiple baselines demonstrated that when the proposed CQPR for mult-fitting detection in transmission lines was applied to the baseline, achieving the accuracy of Faster R-CNN, Cascade R-CNN, Libra R-CNN, and Dynamic R-CNN at 82.9%, 82.4%, 83.7%, and 77.3%, respectively. These results surpassed those of other state-of-the-art object detection models, particularly in the accuracy of detection for small-size fittings and fittings with occlusion. The inference speed was also improved, achieving a balance between localization accuracy and real-time detection.

Key words: transmission lines, fitting, deep learning, object detection, small-size object, dense occlusion

中图分类号:

TP391
TM75

翟永杰, 王璐瑶, 赵晓瑜, 胡哲东, 王乾铭, 王亚茹. 基于级联查询-位置关系的输电线路多金具检测方法[J]. 图学学报, 2025, 46(2): 288-299.

ZHAI Yongjie, WANG Luyao, ZHAO Xiaoyu, HU Zhedong, WANG Qianming, WANG Yaru. Multi-fitting detection for transmission lines based on a cascade query-position relationship method[J]. Journal of Graphics, 2025, 46(2): 288-299.

图/表 14

参考文献 30

[1]	中能传媒研究院. 中国能源大数据报告(2023): 第一章能源发展综述[EB/OL]. (2023-06-20) [2024-04-06]. https://news.bjx.com.cn/html/20230620/1313822.shtml
	China Energy Media Academy. China energy big data report (2023)[EB/OL]. (2023-06-20) [2024-04-06]. https://news.bjx.com.cn/html/20230620/1313822.shtml (in Chinese).
[2]	李德海, 李丹丹. 基于视觉的输电线路自动监测技术综述[J]. 黑龙江电力, 2019, 41(6): 559-564.
	LI D H, LI D D. Overview of vision-based transmission line automatic monitoring technology[J]. Heilongjiang Electric Power, 2019, 41(6): 559-564 (in Chinese).
[3]	ZHAO Z B, LIU N. The recognition and localization of insulators adopting SURF and IFS based on correlation coefficient[J]. Optik, 2014, 125(20): 6049-6052.
[4]	陈文师. 高压输电线路的视觉巡检方法研究[J]. 现代制造, 2017(18): 152-153.
	CHEN W S. Research on visual inspection method of high voltage transmission line[J]. Mechanical and Electrical Information, 2017(18): 152-153 (in Chinese).
[5]	高文婷, 刘越. 面向移动增强现实的实时深度学习目标检测方法综述[J]. 图学学报, 2021, 42(4): 525-534.
	GAO W T, LIU Y. Review of real-time deep learning-based object detection for mobile augmented reality[J]. Journal of Graphics, 2021, 42(4): 525-534 (in Chinese).
[6]	孙宗康, 饶睦敏, 曹裕灵, 等. 基于小样本不均衡数据的供水管道泄漏智能检测算法[J]. 图学学报, 2022, 43(5): 825-831.
	SUN Z K, RAO M M, CAO Y L, et al. Water supply pipeline leakage intelligent detection algorithm based on small and unbalanced data[J]. Journal of Graphics, 2022, 43(5): 825-831 (in Chinese).
[7]	赵振兵, 吕雪纯, 王帆帆, 等. 基于改进YOLOx-S的输电线路上金具检测方法[J]. 无线电工程, 2023, 53(11): 2664-2672.
	ZHAO Z B, LYU X C, WANG F F, et al. Fitting detection on transmission line based on improved YOLOx-S[J]. Radio Engineering, 2023, 53(11): 2664-2672 (in Chinese).
[8]	赵振兵, 吕雪纯, 苗思雨, 等. 基于共有特征评分的解耦知识蒸馏输电线路金具检测方法[EB/OL]. (2023-11-13) [2024- 04-03]. https://doi.org/10.13336/j.1003-6520.hve.20231541.
	ZHAO Z B, LÓ X C, MIAO S Y, et al. Decoupled knowledge distillation transmission line fittings inspection method based on shared feature scoring[EB/OL]. (2023-11-13)[2024-04-03]. https://doi.org/10.13336/j.1003-6520.hve.20231541 (in Chinese).
[9]	张涛, 高绅, 陈万培, 等. 基于深度学习的间隔棒故障识别方法[J]. 无线电工程, 2021, 51(6): 502-506.
	ZHANG T, GAO S, CHEN W P, et al. Spacer bar fault detection based on deep learning[J]. Radio Engineering, 2021, 51(6): 502-506 (in Chinese).
[10]	HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 4700-4708.
[11]	VIEIRA-E-SILVA A L B, DE CASTRO FELIX H, DE MENEZES CHAVES T, et al. STN PLAD: a dataset for multi-size power line assets detection in high-resolution UAV images[C]// The 34th SIBGRAPI Conference on Graphics, Patterns and Images. New York: IEEE Press, 2021: 215-222.
[12]	戚银城, 赵席彬, 耿劭锋, 等. 基于遮挡关系推理的输电线路图像金具检测[J]. 智能系统学报, 2022, 17(6): 1154-1162.
	QI Y C, ZHAO X B, GENG S F, et al. Fittings detection in transmission line images with occlusion relation inference[J]. CAAI Transactions on Intelligent Systems, 2022, 17(6): 1154-1162 (in Chinese).
[13]	YANG C H Y, HUANG Z H, WANG N Y. QueryDet: cascaded sparse query for accelerating high-resolution small object detection[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 13658-13667.
[14]	GRAHAM B, VAN DER MAATEN L. Submanifold sparse convolutional networks[EB/OL]. [2024-05-15]. https://arxiv.org/abs/1706.01307.
[15]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]// 2017 IEEE International Conference on Computer Vision. New York: IEEE Press, 2017: 2980-2988.
[16]	HU H, GU J Y, ZHANG Z, et al. Relation networks for object detection[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 3588-3597.
[17]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// The 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6000-6010.
[18]	PASZKE A, GROSS S, MASSA F, et al. PyTorch: an imperative style, high-performance deep learning library[EB/OL]. [2024-05-15]. https://dblp.uni-trier.de/db/conf/nips/nips2019.html#PaszkeGMLBCKLGA19.
[19]	CHEN K, WANG J Q, PANG J M, et al. MMDetection: open MMLab detection toolbox and benchmark[EB/OL]. [2024-05-15]. https://arxiv.org/abs/1906.07155.
[20]	REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-Time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. DOI PMID
[21]	CAI Z W, VASCONCELOS N. Cascade R-CNN: delving into high quality object detection[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 6154-6162.
[22]	PANG J M, CHEN K, SHI J P, et al. Libra R-CNN: towards balanced learning for object detection[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 821-830.
[23]	ZHANG H K, CHANG H, MA B P, et al. Dynamic R-CNN: towards high quality object detection via dynamic training[C]// The 16th European Conference on Computer Vision. Cham: Springer, 2020: 260-275.
[24]	LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: common objects in context[C]// The 13th European Conference on Computer Vision. Cham: Springer, 2014: 740-755.
[25]	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 7464-7475.
[26]	JOCHER G, CHAURASIA A, QIU J. Ultralytics YOLO (Version 8.0.0)[EB/OL]. [2024-07-06]. https://github.com/ultralytics/ultralytics.
[27]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]// The 14th European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
[28]	KONG T, SUN F C, LIU H P, et al. FoveaBox: beyound anchor-based object detection[J]. IEEE Transactions on Image Processing, 2020, 29: 7389-7398.
[29]	FENG C, ZHONG Y, GAO Y, et al. Tood: Task-aligned one-stage object detection[C]// 2021 IEEE/CVF International Conference on Computer Vision. IEEE Computer Society, 2021: 3490-3499.
[30]	CHEN Z H, YANG C H Y, LI Q F, et al. Disentangle your dense object detector[C]// The 29th ACM International Conference on Multimedia. New York: ACM, 2021: 4939-4948.

模型	方法	AP_50-95	AP₅₀	AP₇₅	AP_S	AP_M	AP_L	AR₁₀₀	FPS
Faster R-CNN^[20]	Baseline	39.4	78.9	35.7	14.7	26.9	41.5	48.1	16.4
Faster R-CNN^[20]	CQPR Faster	42.9	82.9	39.9	18.8	29.4	44.8	50.6	25.0
Cascade R-CNN^[21]	Baseline	41.5	79.4	39.3	18.0	29.5	43.7	51.2	14.7
Cascade R-CNN^[21]	CQPR Cascade	43.9	82.4	43.1	20.2	34.7	45.8	52.3	10.8
Libra R-CNN^[22]	Baseline	40.5	80.3	36.4	18.3	28.4	42.4	49.6	10.7
Libra R-CNN^[22]	CQPR Libra	43.0	83.7	40.5	18.5	35.7	44.8	51.0	24.0
Dynamic R-CNN^[23]	Baseline	38.7	75.1	36.2	12.8	29.1	40.7	46.6	25.9
Dynamic R-CNN^[23]	CQPR Dynamic	41.7	77.3	40.0	19.4	35.4	43.3	49.4	24.7

模型	方法	AP_50-95	AP₅₀	AP₇₅	AP_S	AP_M	AP_L	AR₁₀₀	FPS
Faster R-CNN^[20]	Baseline	39.4	78.9	35.7	14.7	26.9	41.5	48.1	16.4
Faster R-CNN^[20]	CQPR Faster	42.9	82.9	39.9	18.8	29.4	44.8	50.6	25.0
Cascade R-CNN^[21]	Baseline	41.5	79.4	39.3	18.0	29.5	43.7	51.2	14.7
Cascade R-CNN^[21]	CQPR Cascade	43.9	82.4	43.1	20.2	34.7	45.8	52.3	10.8
Libra R-CNN^[22]	Baseline	40.5	80.3	36.4	18.3	28.4	42.4	49.6	10.7
Libra R-CNN^[22]	CQPR Libra	43.0	83.7	40.5	18.5	35.7	44.8	51.0	24.0
Dynamic R-CNN^[23]	Baseline	38.7	75.1	36.2	12.8	29.1	40.7	46.6	25.9
Dynamic R-CNN^[23]	CQPR Dynamic	41.7	77.3	40.0	19.4	35.4	43.3	49.4	24.7

模型	方法	PT	UT	SP	WT	YP	HB	CT	BT	AB	PG
Faster R-CNN^[20]	Baseline	84.5	76.7	85.3	76.2	77.7	76.2	56.5	94.0	83.9	78.2
Faster R-CNN^[20]	CQPR Faster	93.9	77.5	89.4	82.5	81.0	79.9	62.5	94.2	85.8	81.9
Cascade R-CNN^[21]	Baseline	91.4	76.6	88.1	70.5	76.3	77.2	59.5	94.7	84.7	75.4
Cascade R-CNN^[21]	CQPR Cascade	91.6	78.5	88.0	79.1	78.9	79.1	65.3	93.2	85.2	85.0
Libra R-CNN^[22]	Baseline	89.1	78.9	87.5	77.1	77.5	78.4	62,1	95.5	84.6	72.0
Libra R-CNN^[22]	CQPR Libra	93.7	80.5	89.4	82.5	79.8	80.1	64.5	94.7	85.2	86.3
Dynamic R-CNN^[23]	Baseline	91.3	74.1	79.9	58.8	67.4	75.7	52.6	93.9	83.8	72.8
Dynamic R-CNN^[23]	CQPR Dynamic	85.3	76.4	82.3	59.5	73.9	77.2	56.0	93.3	84.1	85.4

模型	方法	PT	UT	SP	WT	YP	HB	CT	BT	AB	PG
Faster R-CNN^[20]	Baseline	84.5	76.7	85.3	76.2	77.7	76.2	56.5	94.0	83.9	78.2
Faster R-CNN^[20]	CQPR Faster	93.9	77.5	89.4	82.5	81.0	79.9	62.5	94.2	85.8	81.9
Cascade R-CNN^[21]	Baseline	91.4	76.6	88.1	70.5	76.3	77.2	59.5	94.7	84.7	75.4
Cascade R-CNN^[21]	CQPR Cascade	91.6	78.5	88.0	79.1	78.9	79.1	65.3	93.2	85.2	85.0
Libra R-CNN^[22]	Baseline	89.1	78.9	87.5	77.1	77.5	78.4	62,1	95.5	84.6	72.0
Libra R-CNN^[22]	CQPR Libra	93.7	80.5	89.4	82.5	79.8	80.1	64.5	94.7	85.2	86.3
Dynamic R-CNN^[23]	Baseline	91.3	74.1	79.9	58.8	67.4	75.7	52.6	93.9	83.8	72.8
Dynamic R-CNN^[23]	CQPR Dynamic	85.3	76.4	82.3	59.5	73.9	77.2	56.0	93.3	84.1	85.4

模型	PRM	CSQN	AP	AP₅₀	AP₇₅	AP_S	AP_M	AP_L	FPS
Faster R-CNN			39.4	78.9	35.7	14.7	26.9	41.5	16.4
	√		42.3	82.0	39.4	16.2	32.6	44.2	15.1
		√	42.2	82.0	39.2	19.5	32.2	44.1	27.4
	√	√	42.9	82.9	39.9	18.8	29.4	44.8	25.0
Cascade R-CNN			41.5	79.4	39.3	18.0	29.5	43.7	14.7
	√		43.0	80.5	42.2	18.7	31.7	45.0	7.1
		√	44.1	81.9	42.5	21.0	30.5	46.1	23.0
	√	√	43.9	82.4	43.1	20.2	34.7	45.8	10.8
Libra R-CNN			40.5	80.3	36.4	18.3	28.4	42.4	10.7
	√		43.4	81.7	40.7	18.8	37.5	45.2	14.8
		√	43.2	81.9	38.5	20.8	32.5	45.2	25.9
	√	√	43.0	83.7	40.5	18.5	35.7	44.8	24.0
Dynamic R-CNN			38.7	75.1	36.2	12.8	29.1	40.7	25.9
	√		40.8	75.6	39.6	18.4	36.3	42.5	15.8
		√	40.5	75.8	39.3	20.1	31.2	42.5	26.9
	√	√	41.7	77.3	40.0	19.4	35.4	43.3	24.7

基于级联查询-位置关系的输电线路多金具检测方法

Multi-fitting detection for transmission lines based on a cascade query-position relationship method

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 30

相关文章 15

编辑推荐

Metrics

本文评价

模型	PRM	CSQN	AP	AP₅₀	AP₇₅	AP_S	AP_M	AP_L
Faster R-CNN			42.9	80.6	42.7	13.9	31.1	45.2
	√		46.3	82.6	47.5	15.3	35.2	48.3
	√	√	44.8	82.7	45.7	19.7	35.2	46.8
Cascade R-CNN			44.6	81.2	45.8	17.5	32.4	46.9
	√		46.5	81.9	49.2	14.4	32.8	48.9
	√	√	47.3	83.6	51.1	22.1	35.9	49.4
Libra R-CNN			43.8	81.0	41.8	18.6	32.6	45.8
	√		46.2	82.8	48.0	17.3	35.8	48.3
	√	√	45.9	84.3	48.7	19.8	35.0	47.9
Dynamic R-CNN			39.8	69.1	43.5	8.8	27.8	42.1
	√		44.6	76.9	45.8	15.3	34.3	46.7
	√	√	45.3	79.3	46.2	24.5	34.1	47.3

模型	AP_50-95	AP₅₀	AP₇₅	AP_S	AP_M	AP_L	AR₁₀₀	AR_S	FPS
YOLOv7	36.4	73.4	32.1	14.9	25.9	38.3	48.3	22.9	33.1
YOLOv8	41.1	77.7	40.6	13.6	27.4	43.7	51.0	25.0	50.4
SSD	18.6	46.5	12.1	0.8	6.3	20.6	31.2	5.7	22.9
Foveabox	32.2	65.3	28.0	8.8	23.5	34.3	46.7	20.0	14.8
TOOD	41.3	79.4	37.3	9.4	36.3	43.4	53.2	23.9	20.4
DDOD	41.9	80.1	37.6	12.6	30.7	44.1	53.2	23.9	25.8
Faster R-CNN	39.4	78.9	35.7	14.7	26.9	41.5	48.1	21.9	16.4
Cascade R-CNN	41.5	79.4	39.3	18.0	29.5	43.7	51.2	26.5	14.7
Libra R-CNN	40.5	80.3	36.4	18.3	28.4	42.4	49.6	26.8	10.7
Dynamic R-CNN	38.7	75.1	36.2	12.8	29.1	40.7	46.6	16.6	25.9
Our-Fas	42.9	82.9	39.9	18.8	29.4	44.8	50.6	27.8	25.0
Our-Cas	43.9	82.4	43.1	20.2	34.7	45.8	52.3	27.3	10.8
Our-Lir	43.0	83.7	40.5	18.5	35.7	44.8	51.0	27.6	24.0
Our- Dyn	41.7	77.3	40.0	19.4	35.4	43.3	49.4	24.9	24.7

[1]	张立立, 杨康, 张珂, 魏薇, 李晶, 谭洪鑫, 张翔宇. 面向柴油车辆排放黑烟的改进型YOLOv8检测算法研究[J]. 图学学报, 2025, 46(2): 249-258.
[2]	潘树焱, 刘立群. MSFAFuse：基于多尺度特征信息与注意力机制的SAR和可见光图像融合模型[J]. 图学学报, 2025, 46(2): 300-311.
[3]	张天圣, 朱闽峰, 任怡雯, 王琛涵, 张立冬, 张玮, 陈为. BPA-SAM：面向工笔画数据的SAM边界框提示增强方法[J]. 图学学报, 2025, 46(2): 322-331.
[4]	孙禾衣, 李艺潇, 田希, 张松海. 结合程序内容生成与扩散模型的图像到三维瓷瓶生成技术[J]. 图学学报, 2025, 46(2): 332-344.
[5]	陈瑞启, 刘晓飞, 万峰, 侯鹏, 沈金屹. 数字孪生驱动的卫星太阳翼展开测试仿真与预测方法[J]. 图学学报, 2025, 46(2): 449-458.
[6]	程旭东, 史彩娟, 高炜翔, 王森, 段昌钰, 闫晓东. 面向域自适应目标检测的一致无偏教师模型[J]. 图学学报, 2025, 46(1): 114-125.
[7]	崔克彬, 耿佳昌. 基于EE-YOLOv8s的多场景火灾迹象检测算法[J]. 图学学报, 2025, 46(1): 13-27.
[8]	汪颜, 张牧雨, 刘秀珍. 基于深度学习的电影海报视觉互动意义评价方法[J]. 图学学报, 2025, 46(1): 221-232.
[9]	苑朝, 赵明雪, 张丰羿, 冯晓勇, 李冰, 陈瑞. 基于点云特征增强的复杂室内场景3D目标检测[J]. 图学学报, 2025, 46(1): 59-69.
[10]	王杨, 马唱, 胡明, 孙涛, 饶元, 袁振羽. 基于多尺度特征融合的轻量型野外蝙蝠检测[J]. 图学学报, 2025, 46(1): 70-80.
[11]	孙前来, 林绍杭, 刘东峰, 宋晓阳, 刘佳耀, 刘瑞珍. 基于元学习的小样本指针式仪表检测方法[J]. 图学学报, 2025, 46(1): 81-93.
[12]	李琼, 考月英, 张莹, 徐沛. 面向无人机航拍图像的目标检测研究综述[J]. 图学学报, 2024, 45(6): 1145-1164.
[13]	刘灿锋, 孙浩, 东辉. 结合Transformer与Kolmogorov Arnold网络的分子扩增时序预测研究[J]. 图学学报, 2024, 45(6): 1256-1265.
[14]	宋思程, 陈辰, 李晨辉, 王长波. 基于密度图多目标追踪的时空数据可视化[J]. 图学学报, 2024, 45(6): 1289-1300.
[15]	李珍峰, 符世琛, 徐乐, 孟博, 张昕, 秦建军. 基于MBI-YOLOv8的煤矸石目标检测算法研究[J]. 图学学报, 2024, 45(6): 1301-1312.

模型	方法	正常	暗化	增亮	有雾	有雨
Faster R-CNN	Baseline	78.9	65.5	76.8	78.3	74.0
Faster R-CNN	Ours	82.9	67.8	79.4	79.8	76.0
Cascade R-CNN	Baseline	79.4	64.2	76.1	77.3	73.7
Cascade R-CNN	Ours	82.4	67.4	78.6	79.3	75.1
Libra R-CNN	Baseline	80.3	66.3	75.4	79.6	73.2
Libra R-CNN	Ours	83.7	69.2	79.2	81.6	76.0
Dynamic R-CNN	Baseline	75.1	61.0	70.2	71.5	69.4
Dynamic R-CNN	Ours	77.3	61.8	73.5	75.3	71.8

模型	方法	正常	暗化	增亮	有雾	有雨
Faster R-CNN	Baseline	78.9	65.5	76.8	78.3	74.0
Faster R-CNN	Ours	82.9	67.8	79.4	79.8	76.0
Cascade R-CNN	Baseline	79.4	64.2	76.1	77.3	73.7
Cascade R-CNN	Ours	82.4	67.4	78.6	79.3	75.1
Libra R-CNN	Baseline	80.3	66.3	75.4	79.6	73.2
Libra R-CNN	Ours	83.7	69.2	79.2	81.6	76.0
Dynamic R-CNN	Baseline	75.1	61.0	70.2	71.5	69.4
Dynamic R-CNN	Ours	77.3	61.8	73.5	75.3	71.8