面向积水干扰的变电设备渗漏油精准分割方法

doi:10.11996/JG.j.2095-302X.2026020296

图学学报 ›› 2026, Vol. 47 ›› Issue (2): 296-310.DOI: 10.11996/JG.j.2095-302X.2026020296

• 图像处理与计算机视觉 • 上一篇下一篇

面向积水干扰的变电设备渗漏油精准分割方法

赵振兵¹^,²^,³(), 张靖梁¹, 唐辰康¹, 毕雨轩¹, 李浩鹏¹^,²

¹ 华北电力大学电子与通信工程系，河北保定 071003
² 华北电力大学河北省电力物联网技术重点实验室，河北保定 071003
³ 华北电力大学复杂能源系统智能计算教育部工程研究中心，河北保定 071003

收稿日期:2025-07-31 接受日期:2025-10-10 出版日期:2026-04-30 发布日期:2026-05-20
通讯作者:赵振兵，E-mail：zhaozhenbing@ncepu.edu.cn
基金资助:
国家自然科学基金(62571189);国家自然科学基金(62373151);国家自然科学基金(62371188);国家自然科学基金(62303184);中央高校基本科研业务费专项资金(2023JC006)

Precise-oil leakage segmentation for substation equipment under water-accumulation interference

ZHAO Zhenbing¹^,²^,³(), ZHANG Jingliang¹, TANG Chenkang¹, BI Yuxuan¹, LI Haopeng¹^,²

¹ Department of Electronic and Communication Engineering, North China Electric Power University, Baoding Hebei 071003, China
² Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding Hebei 071003, China
³ Engineering Research Center of Intelligent Computing for Complex Energy Systems, Ministry of Education, North China Electric Power University, Baoding Hebei 071003, China

Received:2025-07-31 Accepted:2025-10-10 Published:2026-04-30 Online:2026-05-20
Contact: ZHAO Zhenbing，E-mail：zhaozhenbing@ncepu.edu.cn
Supported by:
National Natural Science Foundation of China(62571189);National Natural Science Foundation of China(62373151);National Natural Science Foundation of China(62371188);National Natural Science Foundation of China(62303184);Fundamental Research Funds for the Central Universities(2023JC006)

摘要/Abstract

摘要：

变电设备渗漏油的准确分割对于保障电力系统安全运行至关重要。然而，渗漏油与积水的高度视觉相似性、自身形态的不规则性以及此类干扰场景下训练数据的匮乏，对现有分割方法构成了严峻挑战。针对上述问题，提出了一种从数据增强到网络模型优化的综合解决方案。首先，设计了一种新颖的基于扩散模型的时间步自适应调谐方法(EvoTune)，通过动态调整U-Net在图像生成过程中的特征贡献，有效扩充了数据集中积水干扰场景的样本数量与多样性。其次，在数据增强的基础上，提出了一种高性能的渗漏油分割网络HyDR-Net。该网络通过判别式边界抗扰模块(DBAIM)，增强渗漏油与积水等易混淆背景的特征判别能力，并有效抑制背景噪声；以及多尺度注意力校准模块(MAAM)，对渗漏油特征进行多尺度上下文感知和精细化的边界校准，以适应渗漏油不规则的形态。实验结果表明，EvoTune在SSIM，PSNR和NIQE指标上分别达到0.918 8，26.790 2 dB和5.713 3，显著提升了训练数据的质量和积水区域的生成真实感；而HyDR-Net在F1和PA指标上分别达到80.46%和92.15%，在各项关键评价指标上均大幅超越了现有主流分割方法，尤其在复杂积水干扰场景下展现出卓越的分割精度与鲁棒性。为解决特定视觉干扰下的数据稀缺问题提供了有效途径，并为变电设备渗漏油的智能、精准检测提供了强有力的技术支撑。

关键词: 渗漏油分割, 变电设备, 数据增强, 扩散模型, 积水干扰

Abstract:

Accurate segmentation of oil leakage from substation equipment is crucial for ensuring the safe operation of power systems. However, existing segmentation methods face significant challenges due to the high visual similarity between oil leakage and water accumulation, the irregular morphology of oil spills, and the scarcity of training data for such interference scenarios. To address these issues, a comprehensive solution from data augmentation to network-model optimization was proposed. First, a novel diffusion model-based timestep-adaptive tuning method (Evolutionary Tuning, EvoTune) was designed to dynamically adjust the feature contributions of U-Net during image generation, effectively expanding the quantity and diversity of water interference scenarios in the dataset. Second, based on the data augmentation, a high-performance oil leakage segmentation network HyDR-Net (Hydro Discriminative Refining Network) was proposed. The network incorporated a Discriminative Boundary Anti-Interference Module (DBAIM) to enhance feature discrimination between oil leakage and confusing backgrounds such as water accumulation while effectively suppressing background noise, and a Multi-Scale Attentive Alignment Module (MAAM) for multi-scale context-aware processing and fine-grained boundary calibration of oil leakage features to accommodate the irregular morphology of oil spills. Experimental results showed that EvoTune achieved SSIM, PSNR, and NIQE scores of 0.918 8, 26.790 2 dB, and 5.713 3, respectively, significantly improving training-data quality and the realism of generated water-accumulation regions, while HyDR-Net achieved F1 and PA scores of 80.46% and 92.15%, respectively, substantially outperforming existing mainstream segmentation methods across all key evaluation metrics, and particularly exhibiting superior segmentation accuracy and robustness under complex water-interference scenarios. The research provided an effective approach for addressing data scarcity under specific visual-interference conditions and offered robust technical support for intelligent and precise detection of oil leakage in substation equipment.

Key words: oil-leakage segmentation, substation equipment, data augmentation, diffusion models, water-accumulation interference

中图分类号:

赵振兵, 张靖梁, 唐辰康, 毕雨轩, 李浩鹏. 面向积水干扰的变电设备渗漏油精准分割方法[J]. 图学学报, 2026, 47(2): 296-310.

ZHAO Zhenbing, ZHANG Jingliang, TANG Chenkang, BI Yuxuan, LI Haopeng. Precise-oil leakage segmentation for substation equipment under water-accumulation interference[J]. Journal of Graphics, 2026, 47(2): 296-310.

图/表 12

图1 时间步自适应调谐方法整体网络结构图

Fig. 1 Overall network architecture diagram of timestep adaptive tuning method

图2 HyDR-Net网络结构图

Fig. 2 HyDR-Net network architecture diagram

图3 判别式边界抗扰模块网络结构图

Fig. 3 Network architecture diagram of discriminative boundary anti-interference model

图4 差异感知边界精炼单元(DABRU)网络结构图

Fig. 4 Network architecture diagram of difference-aware boundary refinement unit (DABRU)

图5 多尺度注意力校准模块网络结构图

Fig. 5 Network architecture diagram of multi-scale attentive alignment model

表1 图像生成质量对比结果

Table 1 The comparison results of image generation quality

方法	SSIM	PSNR/dB	NIQE
Dall-E 2	0.712 4	22.456 0	7.623 4
Imagen	0.738 9	23.127 0	7.289 1
SD	0.730 8	23.017 0	7.010 0
SD +LoRA	0.746 5	23.023 7	6.985 7
SD + LoRA +EvoTune	0.918 8	26.790 2	5.713 3

图6 不同方法的生成结果((a) 原始图像；(b) Dall-E 2；(c) Imagen；(d) SD；(e) SD+LoRA；(f) SD+LoRA+EvoTune)

Fig. 6 Generation results of different methods ((a) Original image; (b) Dall-E 2; (c) Imagen; (d) SD; (e) SD+LoRA;(f) SD+LoRA+EvoTune)

表2 在变电设备渗漏油分割数据集上的对比实验结果

Table 2 The comparison experimental results on substation equipment oil leakage segmentation dataset

方法	F1/%	IOU/%	PA/%	Params/M	FLOPs/G
ANN	73.61	58.24	88.71	43.93	190.0
DeeeplabV3+	75.42	60.54	90.64	41.29	182.0
UPerNet(Twins)	77.62	63.42	90.99	90.19	275.0
KNet + PSPNet	71.85	56.06	89.13	59.89	190.0
UPerNet(Swin)	73.80	58.47	90.58	122.88	306.0
DMNet	74.53	59.40	90.22	50.88	201.0
TransUnet	75.78	61.01	90.41	91.52	128.0
SegFormer	76.48	61.91	90.79	27.36	56.9
DSACP	77.95	63.87	91.38	42.48	101.0
SAM	68.99	41.28	85.26	89.70	677.0
Ours	80.46	67.31	92.15	32.83	239.0

图7 Nemenyi检验可视化结果

Fig. 7 Visualization results of Nemenyi test

图8 积水干扰场景下渗漏油不同网络的分割结果((a) 原始图像；(b) 标签图像；(c) ANN；(d) SegFormer；(e) UPerNet(Twins)；(f) DSACP；(g) HyDR-Net)

Fig. 8 Segmentation results of oil leakage by different networks under water accumulation interference scenarios ((a) Original image; (b) Ground truth; (c) ANN; (d) SegFormer; (e) UPerNet (Twins); (f) DSACP; (g) HyDR-Net)

表3 HyDR-Net消融实验结果

Table 3 The ablation experimental results of HyDR-Net

基线	DBAIM	MAAM	F1/%	IOU/%	PA/%
√			76.48	61.91	90.79
√	√		78.88	65.13	91.64
√		√	78.84	65.07	91.53
√	√	√	80.46	67.31	92.15

图9 不同网络深层特征xfuse的热力图映射结果((a) 原始图像；(b) 标签图像；(c) 基线网络；(d) 基线+DBAIM；(e) 基线+MAAM；(f) 基线+DBAIM+MAAM)

Fig. 9 Heatmap visualization results of deep feature xfuse from different networks ((a) Original image; (b) Ground truth; (c) Baseline network; (d) Baseline+DBAIM; (e) Baseline+MAAM; (f) Baseline+DBAIM+MAAM)

参考文献 29

[1]	韩笑, 郭剑波, 蒲天骄, 等. 电力人工智能技术理论基础与发展展望(一): 假设分析与应用范式[J]. 中国电机工程学报, 2023, 43(8): 2877-2890.
	HAN X, GUO J B, PU T J, et al. Theoretical foundation and directions of electric power artificial intelligence (Ⅰ): hypothesis analysis and application paradigm[J]. Proceedings of the CSEE, 2023, 43(8): 2877-2890 (in Chinese).
[2]	盛戈皞, 钱勇, 罗林根, 等. 面向新型电力系统的电力设备运行维护关键技术及其应用展望[J]. 高电压技术, 2021, 47(9): 3072-3084.
	SHENG G H, QIAN Y, LUO L G, et al. Key technologies and application prospects for operation and maintenance of power equipment in new type power system[J]. High Voltage Engineering, 2021, 47(9): 3072-3084 (in Chinese).
[3]	赵振兵, 欧阳文斌, 冯烁, 等. 基于类内稀疏先验与改进YOLOv8的绝缘子红外图像检测方法[EB/OL]. (2025-04-28) [2025-07-29]. https://link.cnki.net/urlid/10.1034.T.20250428.1048.002.
	ZHAO Z B, OUYANG W B, FENG S, et al. A thermal image detection method for insulators incorporating within-class sparse prior knowledge and improved YOLOv8[EB/OL]. (2025-04-28) [2025-07-29]. https://link.cnki.net/urlid/10.1034.T.20250428.1048.002 (in Chinese).
[4]	王鹏, 赵春晖, 周华良, 等. 基于多时相巡检图像的变电设备抗干扰缺陷检测[J]. 控制与决策, 2024, 39(3): 885-892.
	WANG P, ZHAO C H, ZHOU H L, et al. Anti-interference defect detection of substation equipment based on multi-temporal inspection images[J]. Control and Decision, 2024, 39(3): 885-892 (in Chinese).
[5]	张迎晨, 王波, 马富齐, 等. 基于域适应网络的电力设备不规则外表面缺陷高精度检测方法[J]. 高电压技术, 2022, 48(11): 4516-4526.
	ZHANG Y C, WANG B, MA F Q, et al. High-precision detection method of irregular outer surface defects of power equipment based on domain adaptation network[J]. High Voltage Engineering, 2022, 48(11): 4516-4526 (in Chinese).
[6]	赵文清, 刘亮, 胡嘉伟, 等. 基于深度可分离空洞卷积金字塔的变压器渗漏油检测[J]. 智能系统学报, 2023, 18(5): 966-974.
	ZHAO W Q, LIU L, HU J W, et al. Detection of transformer oil leakage based on deep separable atrous convolution pyramid[J]. CAAI Transactions on Intelligent Systems, 2023, 18(5): 966-974 (in Chinese).
[7]	CHEN H, XIANG Q, HU J X, et al. Comprehensive exploration of diffusion models in image generation: a survey[J]. Artificial Intelligence Review, 2025, 58(4): 99. DOI
[8]	LI C, QI Y L, ZENG Q T, et al. Comparison of image generation methods based on diffusion models[C]// The 4th International Conference on Computer Vision, Image and Deep Learning. New York: IEEE Press, 2023: 1-4.
[9]	ZHANG C, YADAV A, TANG L Y, et al. Algorithm optimization of image generation model based on diffusion model[C]// The 8th International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud). New York: IEEE Press, 2024: 1369-1374.
[10]	ROMBACH R, BLATTMANN A, LORENZ D, et al. High-resolution image synthesis with latent diffusion models[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 10674-10685.
[11]	涂晴昊, 李元琪, 刘一凡, 等. 基于扩散模型的文本生成材质贴图的泛化性优化方法[J]. 图学学报, 2025, 46(1): 139-149. DOI
	TU Q H, LI Y Q, LIU Y F, et al. Generalization optimization method for text to material texture maps based on diffusion model[J]. Journal of Graphics, 2025, 46(1): 139-149 (in Chinese). DOI
[12]	YEO Y, UM D. The butterfly effect: color-guided image generation from unconditional diffusion models[J]. IEEE Access, 2025, 13: 27794-27804. DOI URL
[13]	RUIZ N, LI Y Z, JAMPANI V, et al. DreamBooth: fine tuning text-to-image diffusion models for subject-driven generation[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 22500-22510.
[14]	KIDDER B L. Advanced image generation for cancer using diffusion models[J]. Biology Methods and Protocols, 2024, 9(1): bpae062. DOI URL
[15]	AN T, XUE B, HUO C L, et al. Efficient remote sensing image super-resolution via lightweight diffusion models[J]. IEEE Geoscience and Remote Sensing Letters, 2024, 21: 6000105.
[16]	RAMESH A, DHARIWAL P, NICHOL A, et al. Hierarchical text-conditional image generation with CLIP latents[EB/OL]. [2025-05-31]. https://arxiv.org/abs/2204.06125.
[17]	SAHARIA C, CHAN W, SAXENA S, et al. Photorealistic text-to-image diffusion models with deep language understanding[C]// The 36th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2022: 2643.
[18]	HU E J, SHEN Y, WALLIS P, et al. LoRA: low-rank adaptation of large language models[EB/OL]. [2025-05-31]. https://arxiv.org/pdf/2106.09685.pdf.
[19]	叶文龙, 陈斌. PanoLoRA: 基于Stable Diffusion的全景图像生成的高效微调方法[J]. 图学学报, 2025, 46(5): 980-989. DOI
	YE W L, CHEN B. PanoLoRA: an efficient finetuning method for panoramic image generation based on Stable Diffusion[J]. Journal of Graphics, 2025, 46(5): 980-989 (in Chinese). DOI
[20]	LI X X, LIU X J, XIAO Y, et al. An improved U-Net segmentation model that integrates a dual attention mechanism and a residual network for transformer oil leakage detection[J]. Energies, 2022, 15(12): 4238. DOI URL
[21]	GENG P, TAN Z Y, LUO J, et al. ACPA-Net: atrous channel pyramid attention network for segmentation of leakage in rail tunnel linings[J]. Electronics, 2023, 12(2): 255. DOI URL
[22]	梁翀, 李程启, 张彬彬. 基于深度学习和超分辨率重构技术的图像缺陷诊断算法[J]. 山东农业大学学报(自然科学版), 2020, 51(3): 510-513.
	LIANG C, LI C Q, ZHANG B B. A defect diagnosis algorithm for images based on the deep learning and super-resolution reconstruction technique[J]. Journal of Shandong Agricultural University (Natural Science Edition), 2020, 51(3): 510-513 (in Chinese).
[23]	CHU X X, TIAN Z, WANG Y Q, et al. Twins: revisiting the design of spatial attention in vision transformers[C]// The 35th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2021: 716.
[24]	ZHAO Z B, LIU B, ZHAI Y J, et al. Dual graph reasoning network for oil leakage segmentation in substation equipment[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 3502415.
[25]	CHENG L X, LI Y, ZHAO K J, et al. A two-stage oil spill detection method based on an improved superpixel module and DeepLab V3+ using SAR images[J]. IEEE Geoscience and Remote Sensing Letters, 2025, 22: 4000505.
[26]	CUI G Y, FAN J C, ZOU Y R. Enhanced unsupervised domain adaptation with iterative pseudo-label refinement for inter-event oil spill segmentation in SAR images[J]. International Journal of Applied Earth Observation and Geoinformation, 2025, 139: 104479. DOI URL
[27]	CHEN Y Q, SUN Y H, YU W, et al. A novel lightweight bilateral segmentation network for detecting oil spills on the sea surface[J]. Marine Pollution Bulletin, 2022, 175: 113343. DOI URL
[28]	FAN J C, ZHANG S, WANG X Z, et al. Multifeature semantic complementation network for marine oil spill localization and segmentation based on SAR images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2023, 16: 3771-3783. DOI URL
[29]	ZHU Q Q, ZHANG Y N, LI Z Q, et al. Oil spill contextual and boundary-supervised detection network based on marine SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5213910.

面向积水干扰的变电设备渗漏油精准分割方法

Precise-oil leakage segmentation for substation equipment under water-accumulation interference

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 29

相关文章 15

编辑推荐

Metrics

本文评价

[1]	向婷, 唐卓, 郑佳丽, 陈长建, 吕斐, 李肯立. 基于生成模型的图像数据增强方法综述[J]. 图学学报, 2026, 47(2): 235-250.
[2]	陈梦琪, 赵俊莉, 邓晓丹. 基于大模型的皮肤病图像掩膜生成与分割[J]. 图学学报, 2026, 47(2): 322-331.
[3]	周金, 周一, 徐鹏飞, 黄惠. PDF-Sketch：基于笔画段距离场与离散扩散的布局式草图生成方法[J]. 图学学报, 2026, 47(2): 380-389.
[4]	刘景豪, 游振国, 杜冬. 基于潜在扩散模型的CAD条件生成[J]. 图学学报, 2026, 47(2): 390-401.
[5]	叶文龙, 陈斌. PanoLoRA：基于Stable Diffusion的全景图像生成的高效微调方法[J]. 图学学报, 2025, 46(5): 980-989.
[6]	雷松林, 赵征鹏, 阳秋霞, 普园媛, 谷金晶, 徐丹. 基于可解耦扩散模型的零样本风格迁移[J]. 图学学报, 2025, 46(4): 727-738.
[7]	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.
[8]	李纪远, 管哲予, 宋海川, 谭鑫, 马利庄. 人在环路的细分行业logo生成方法[J]. 图学学报, 2025, 46(2): 382-392.
[9]	方程浩, 王康侃. 基于半监督学习的单视角点云三维人体姿态与形状估计[J]. 图学学报, 2025, 46(2): 393-401.
[10]	涂晴昊, 李元琪, 刘一凡, 过洁, 郭延文. 基于扩散模型的文本生成材质贴图的泛化性优化方法[J]. 图学学报, 2025, 46(1): 139-149.
[11]	张冀, 崔文帅, 张荣华, 王文彬, 李亚琦. 基于关键视图的文本驱动3D场景编辑方法[J]. 图学学报, 2024, 45(4): 834-844.
[12]	王吉, 王森, 蒋智文, 谢志峰, 李梦甜. 基于深度条件扩散模型的零样本文本驱动虚拟人生成方法[J]. 图学学报, 2023, 44(6): 1218-1226.
[13]	曹义亲, 周一纬, 徐露. 基于E-YOLOX的实时金属表面缺陷检测算法[J]. 图学学报, 2023, 44(4): 677-690.
[14]	王道累, 康博, 朱瑞. 基于深度学习的电力设备铭牌文本检测方法[J]. 图学学报, 2023, 44(4): 691-698.
[15]	曾武, 朱恒亮, 邢树礼, 林江宏, 毛国君. 显著性检测引导的图像数据增强方法[J]. 图学学报, 2023, 44(2): 260-270.