基于YOLOv8的轻量化变电设备红外图像故障区域分割方法

王媛彬; 潘志远; 吴冰超

doi:10.19783/j.cnki.pspc.260227

欢迎来到能源电力期刊网！

English Version 期刊登录

您当前的位置：

首页 >

文章列表页 >

基于YOLOv8的轻量化变电设备红外图像故障区域分割方法

应用研究 | 更新时间：2026-09-23

- 基于YOLOv8的轻量化变电设备红外图像故障区域分割方法
- A lightweight YOLOv8-based fault region segmentation method for infrared images of substation equipment
- 电力系统保护与控制 2026年54卷第18期
- 作者机构：
- 作者简介：
- 基金信息：
- DOI：10.19783/j.cnki.pspc.260227
  中图分类号：
- 纸质出版：2026
- 稿件说明：
移动端阅览
王媛彬, 潘志远, 吴冰超. 基于YOLOv8的轻量化变电设备红外图像故障区域分割方法[J]. 电力系统保护与控制, 2026,54(18).

WANG Yuanbin, PAN Zhiyuan, WU Bingchao. A lightweight YOLOv8-based fault region segmentation method for infrared images of substation equipment[J]. Power System Protection and Control, 2026, 54(18).
王媛彬, 潘志远, 吴冰超. 基于YOLOv8的轻量化变电设备红外图像故障区域分割方法[J]. 电力系统保护与控制, 2026,54(18). DOI： 10.19783/j.cnki.pspc.260227.

WANG Yuanbin, PAN Zhiyuan, WU Bingchao. A lightweight YOLOv8-based fault region segmentation method for infrared images of substation equipment[J]. Power System Protection and Control, 2026, 54(18). DOI： 10.19783/j.cnki.pspc.260227.

摘要

针对变电站设备红外图像中故障区域普遍存在形状不规则、目标尺度小及边缘特征模糊等问题，提出一种基于YOLOv8的轻量化红外图像故障区域分割模型。首先，采用MobileNetV3作为主干网络，将深度可分离卷积与注意力机制相结合，在保证特征提取能力的同时显著降低了参数量与计算复杂度。其次，为增强模型对多尺度、不规则目标的感知能力，设计语义细节注入模块，促进高层语义信息与低层细节特征的有效融合。接着，针对红外图像边缘模糊的特性，构建了上采样反向瓶颈特征融合结构，以强化模型对故障区域边界轮廓的精准捕捉。最后，通过引入焦点损失与骰子损失的复合损失函数，有效缓解了正负样本不均衡的问题。实验结果表明，与原始YOLOv8模型相比，所提方法在分割精度上提升了2.38%。模型参数量仅为9.17 M，较原始YOLOv8模型降低了26.46%。与4种主流分割模型的对比表明，所提模型在分割精度和计算效率上均展现出显著优势。

Abstract

To address irregular shapes

small target sizes

and blurred edge features commonly observed in fault regions in infrared images of substation equipment

this paper proposes a lightweight fault region segmentation model based on YOLOv8. First

MobileNetV3 is employed as the backbone

where depthwise separable convolutions are combined with an attention mechanism to significantly reduce the number of parameters and computational complexity while preserving feature extraction capability. Second

a semantic detail injection module is designed to enhance multi-scale and irregular target perception by promoting effective fusion of high-level semantic information and low-level detailed features. Subsequently

an upsampling reverse-bottleneck feature fusion architecture is developed to address the blurred edges characteristic of infrared images

thereby enhancing the model’s ability to accurately capture the boundary contours of fault regions. Finally

a compound loss function combining focal loss and dice loss is introduced to alleviate the positive-negative sample imbalance. Experimental results show that

compared with the original YOLOv8

the proposed method improves segmentation precision by 2.38%

while reducing the number of model parameters to only 9.17 M

a 26.46% reduction compared with the original YOLOv8 model. Comparison with four mainstream segmentation models further demonstrates that the proposed model achieves significant advantages in both segmentation accuracy and computational efficiency.

关键词

Keywords

references

ZHANG Z, LÜ L. Application status and prospects of digital twin technology in distribution grid[J]. Energy Rport, 2022, 8: 14170-14182.

PASHAEI M, KAUHANIEMI K, LAAKSONEN H, et al. Development of virtualized centralized protection and WAMPAC systems: a review[J]. Protection and Control of Modern Power Systems, 2025, 11(1): 1-25.

LIU X, ZHANG Z, HAO Y, et al. Optimized OTSU segmentation algorithm-based temperature feature extraction method for infrared images of electrical equipment[J]. Sensors, 2024, 24(4).

XING Z, HE Y. A two-step image segmentation based on clone selection multi-object emperor penguin optimizer for fault diagnosis of power transformer[J]. Expert Systems with Applications, 2024, 244.

CHEN Y, WANG Z, BAI X. Fuzzy sparse subspace clustering for infrared image segmentation[J]. IEEE Transactions on Image Processing, 2023, 32: 2132-2146.

LIN J, HAN M, LI G, et al. A novel defect segmentation for pulse infrared images based on improved fuzzy C-means algorithm with weighted distance[J]. IEEE Transactions on Magnetics, 2023, 59(2): 1-13.

ZHOU J, LIU Y, GU Y, et al. A box-supervised instance segmentation method for insulator infrared images based on shuffle polarized self-attention[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1-11.

SHU J, HE J, LI L. MSIS: multispectral instance segmentation method for power equipment[J]. Computational Intelligence and Neuroscience, 2022(1).

PENG Y, SONKA M, CHEN D. U-Net v2: rethinking the skip connections of U-Net for medical image segmentation[J]. arXiv preprint arXiv, 2023, 2311, 17791.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于TimeMixer时序预测的电力系统宽频振荡快速辨识

基于双层优化迭代模数并联双驱的电力系统负荷建模方法

基于模型辅助准比例谐振扩张状态观测器的单相构网型逆变器电压控制

基于 CVaR 约束的双层主从博弈优化模型及其在多政策场景下的应用研究

考虑转供率和容载比的主动配电网供电能力提升研究