基于跨层次局部特征融合的光伏热斑缺陷检测

项新建; 孙思琪; 罗秋霞; 唐小辉; 陈祯力; 郑永平

doi:doi:10.19912/j.0254-0096.tynxb.2024-2032

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

基于跨层次局部特征融合的光伏热斑缺陷检测

更新时间：2026-04-08

- 基于跨层次局部特征融合的光伏热斑缺陷检测
- Vol. 47, Issue 3, Pages: 732-738(2026)
- 作者机构：
- 作者简介：
- 基金信息：
- DOI：doi:10.19912/j.0254-0096.tynxb.2024-2032
  CLC： TP183
- Online First：07 April 2026，
  
  Published：2026
- 稿件说明：
移动端阅览
项新建, 孙思琪, 罗秋霞, et al. 基于跨层次局部特征融合的光伏热斑缺陷检测[J]. 2026, 47(3): 732-738.
DOI：

项新建, 孙思琪, 罗秋霞, et al. 基于跨层次局部特征融合的光伏热斑缺陷检测[J]. 2026, 47(3): 732-738. DOI： doi:10.19912/j.0254-0096.tynxb.2024-2032.

摘要

针对光伏热斑缺陷检测过程中

热斑的特征细节丢失而导致微小热斑漏检的问题

提出一种基于YOLOv8的跨层次局部特征融合改进方法。在YOLOv8中引入BiFormer思想改进主干网络

使模型能够更好地提取细节信息;使用Dysample动态上采样模块替代传统上采样方式

以更好地保留高分辨率特征图的细节信息;并设计跨层次局部特征融合的FCLAHead检测头

在不同层级的特征图间建立特征关联

实现信息互补

通过融合操作增强微小热斑的特征表达

提高模型在多场景下对微小热斑的检测能力。改进后的模型相较于基础模型YOLOv8n

平均精度mAP由86.3%提升至90%

精确率和召回率分别提升1.1个百分点和3.1个百分点

参数量仅为3.08×106

计算量略有增加

适用于精度要求高且硬件资源有限的应用场景。

Abstract

To address the issue of missed detections of small hotspots caused by the loss of detailed features in photovoltaic hotspot defect detection

this paper proposes an improved method for cross-level local feature fusion based on YOLOv8. The method introduces the BiFormer concept into the YOLOv8 backbone to enhance the model’s ability to extract detailed information. It replaces traditional upsampling methods with a Dysample dynamic upsampling module

better preserving the details of high-resolution feature maps. Additionally

an FCLAHead detection head is designed for cross-level local feature fusion

establishing feature associations across different feature map levels to achieve information complementarity. This fusion operation enhances the feature representation of small hotspots

improving the model’s detection capability for small hotspots across various scenarios. Compared to the base model YOLOv8n

the improved model increases mean average precision （mAP） from 86.3% to 90%

with precision and recall improving by 1.1 and 3.1 percentage points

respectively. The model’s parameter count is only 3.08×10

with a slight increase in computation

making it suitable for applications requiring high accuracy in hardware-limited environments.

关键词

Keywords

references

尤海侠. 光伏发电效率影响因素分析[J]. 能源技术与管理, 2022, 47(6): 147-149.

王乐, 陈雪, 张舒, 等. 光伏组件热斑效应研究[J]. 太阳能学报, 2023, 44(7): 155-161.

MUÑOZ J, LORENZO E, MARTÍNEZ-MORENO F, et al. An investigation into hot-spots in two large grid-connected PV plants[J]. Progress in photovoltaics: research and applications, 2008, 16(8): 693-701.

HOCINE L, SAMIRA K M, TAREK M, et al.Automatic detection of faults in a photovoltaic power plant based on the observation of degradation indicators[J]. Renewable energy, 2021, 164: 603-617.

王道累, 李明山, 姚勇, 等. 改进SSD的光伏组件热斑缺陷检测方法[J]. 太阳能学报, 2023, 44(4): 420-425.

郎庆凯, 高方玉, 吴琼, 等. 基于改进YOLOv7的光伏组件红外图像热斑目标检测方法[J]. 计算机应用, 2023, 43(增刊2): 191-195.

SUN Y J, HUANG H, YUN X, et al.Triplet attention multiple spacetime-semantic graph convolutional network for skeleton-based action recognition[J]. Applied intelligence, 2022, 52(1): 113-126.

ZHU L, WANG X J, KE Z H, et al.BiFormer: vision transformer with bi-level routing attention[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Vancouver, BC, Canada, 2023: 10323-10333.

LIU W Z, LU H, FU H T, et al.Learning to upsample by learning to sample[C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV). Paris, France, 2024: 6004-6014.

WANG C Y, MARK LIAO H Y, WU Y H, et al. CSPNet: a new backbone that can enhance learning capability of CNN[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops(CVPRW).Seattle, WA, USA, 2020: 1571-1580.

LIU S, QI L, QIN H F, et al.Path aggregation network for instance segmentation[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA, 2018: 8759-8768.

LAW H, DENG J.Cornernet: detecting objects as paired keypoints[C]//Proceedings of the European Conference on Computer Vision (CECCV). Munich, Germany, 2018: 734-750.

ZHENG Z H, WANG P, LIU W, et al.Distance-IoU loss: faster and better learning for bounding box regression[J]. Proceedings of the AAAI conference on artificial intelligence, 2020, 34(7): 12993-13000.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

YOLO-DSBF：一种新的太阳电池缺陷识别方法

基于HCF-YOLO的光伏组件红外图像热斑故障检测

OVERVIEW AND PROSPECT OF INTELLIGENT DETECTION OF HOT SPOT EFFECT OF PHOTOVOLTAIC MODULES

基于Transformer多尺度融合网络的暖通空调能耗预测模型

基于改进YOLOv8的光伏缺陷快速检测

Related Author

何羿颉

楚瀛

夏能弘

江正源

李曦

秦怡鸣

尹丽菊

高晓宁

Related Institution

College of Energy and Mechanical Engineering, Shanghai University of Electric Power

21ST Research Institute of China Electronic Technology Group Corporation

Gaussian Robotics

School of New Energy, Shenyang Institute of Engineering

Liaoning Solar Energy Research and Application Co., Ltd.

AI问答

⁰