Deep learning based defect detection algorithm for solar panels

Jiaqi Li; Hongxu Li; Yifan Wu; Hailiang Zhou; Luigi Manfredi; Peng Li; Hong Zhang

doi:10.1109/WRCSARA60131.2023.10261859

Deep learning based defect detection algorithm for solar panels

Jiaqi Li (Lead / Corresponding author)
, Hongxu Li
, Yifan Wu (Lead / Corresponding author)
, Hailiang Zhou
, Luigi Manfredi (Lead / Corresponding author)
, Peng Li
, Hong Zhang (Lead / Corresponding author)

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

1 Citation (Scopus)

301 Downloads (Pure)

Abstract

Defect detection of solar panels plays an essential role in guaranteeing product quality within automated production lines. However, traditional manual inspection of solar panel defects suffers from low efficiency. This paper proposes an enhanced YOLOv5 algorithm (EL-YOLOv5) fused with the CBAM hybrid attention module to ensure product quality. The algorithm focuses on detecting five common types of defects that frequently appear on photovoltaic production lines, namely hidden cracks, scratches, broken grids, black spots, and short circuits. This study utilizes publicly available solar panel datasets, as well as datasets collected from actual photovoltaic production lines. These datasets are annotated accordingly and used to train the proposed algorithm. The experimental results demonstrate that the proposed algorithm achieves good performance on both the public and actual solar panel defect datasets. Particularly in actual datasets, where defect features are often less apparent and defects are smaller in size, the proposed algorithm can still detect even minor black spots.

Original language	English
Title of host publication	2023 WRC Symposium on Advanced Robotics and Automation (WRC SARA)
Publisher	IEEE
Pages	438-443
Number of pages	6
ISBN (Electronic)	9798350307320
ISBN (Print)	9798350307337
DOIs	https://doi.org/10.1109/WRCSARA60131.2023.10261859
Publication status	Published - 27 Sept 2023
Event	5th WRC Symposium on Advanced Robotics and Automation (WRC SARA) - Chinese Institute of Electronics, Beijing, China Duration: 19 Aug 2023 → 19 Aug 2023

Publication series

Name
ISSN (Print)	2835-3366
ISSN (Electronic)	2835-3358

Conference

Conference	5th WRC Symposium on Advanced Robotics and Automation (WRC SARA)
Country/Territory	China
City	Beijing
Period	19/08/23 → 19/08/23

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 7 Affordable and Clean Energy

Keywords

Training
Production
Manuals
Inspection
Feature extraction
Product design
Quality assessment

ASJC Scopus subject areas

Artificial Intelligence
Control and Optimization
Human-Computer Interaction
Computer Vision and Pattern Recognition
Computer Science Applications

Access to Document

10.1109/WRCSARA60131.2023.10261859

Author Accepted ManuscriptAccepted author manuscript, 2.08 MB

Cite this

@inproceedings{5122102dd12343b5a5f9dc18fb11f55c,

title = "Deep learning based defect detection algorithm for solar panels",

abstract = "Defect detection of solar panels plays an essential role in guaranteeing product quality within automated production lines. However, traditional manual inspection of solar panel defects suffers from low efficiency. This paper proposes an enhanced YOLOv5 algorithm (EL-YOLOv5) fused with the CBAM hybrid attention module to ensure product quality. The algorithm focuses on detecting five common types of defects that frequently appear on photovoltaic production lines, namely hidden cracks, scratches, broken grids, black spots, and short circuits. This study utilizes publicly available solar panel datasets, as well as datasets collected from actual photovoltaic production lines. These datasets are annotated accordingly and used to train the proposed algorithm. The experimental results demonstrate that the proposed algorithm achieves good performance on both the public and actual solar panel defect datasets. Particularly in actual datasets, where defect features are often less apparent and defects are smaller in size, the proposed algorithm can still detect even minor black spots.",

keywords = "Training, Production, Manuals, Inspection, Feature extraction, Product design, Quality assessment",

author = "Jiaqi Li and Hongxu Li and Yifan Wu and Hailiang Zhou and Luigi Manfredi and Peng Li and Hong Zhang",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 5th WRC Symposium on Advanced Robotics and Automation (WRC SARA) ; Conference date: 19-08-2023 Through 19-08-2023",

year = "2023",

month = sep,

day = "27",

doi = "10.1109/WRCSARA60131.2023.10261859",

language = "English",

isbn = "9798350307337",

publisher = "IEEE",

pages = "438--443",

booktitle = "2023 WRC Symposium on Advanced Robotics and Automation (WRC SARA)",

}

TY - GEN

T1 - Deep learning based defect detection algorithm for solar panels

AU - Li, Jiaqi

AU - Li, Hongxu

AU - Wu, Yifan

AU - Zhou, Hailiang

AU - Manfredi, Luigi

AU - Li, Peng

AU - Zhang, Hong

PY - 2023/9/27

Y1 - 2023/9/27

N2 - Defect detection of solar panels plays an essential role in guaranteeing product quality within automated production lines. However, traditional manual inspection of solar panel defects suffers from low efficiency. This paper proposes an enhanced YOLOv5 algorithm (EL-YOLOv5) fused with the CBAM hybrid attention module to ensure product quality. The algorithm focuses on detecting five common types of defects that frequently appear on photovoltaic production lines, namely hidden cracks, scratches, broken grids, black spots, and short circuits. This study utilizes publicly available solar panel datasets, as well as datasets collected from actual photovoltaic production lines. These datasets are annotated accordingly and used to train the proposed algorithm. The experimental results demonstrate that the proposed algorithm achieves good performance on both the public and actual solar panel defect datasets. Particularly in actual datasets, where defect features are often less apparent and defects are smaller in size, the proposed algorithm can still detect even minor black spots.

AB - Defect detection of solar panels plays an essential role in guaranteeing product quality within automated production lines. However, traditional manual inspection of solar panel defects suffers from low efficiency. This paper proposes an enhanced YOLOv5 algorithm (EL-YOLOv5) fused with the CBAM hybrid attention module to ensure product quality. The algorithm focuses on detecting five common types of defects that frequently appear on photovoltaic production lines, namely hidden cracks, scratches, broken grids, black spots, and short circuits. This study utilizes publicly available solar panel datasets, as well as datasets collected from actual photovoltaic production lines. These datasets are annotated accordingly and used to train the proposed algorithm. The experimental results demonstrate that the proposed algorithm achieves good performance on both the public and actual solar panel defect datasets. Particularly in actual datasets, where defect features are often less apparent and defects are smaller in size, the proposed algorithm can still detect even minor black spots.

KW - Training

KW - Production

KW - Manuals

KW - Inspection

KW - Feature extraction

KW - Product design

KW - Quality assessment

UR - https://www.scopus.com/pages/publications/85174202017

U2 - 10.1109/WRCSARA60131.2023.10261859

DO - 10.1109/WRCSARA60131.2023.10261859

M3 - Conference contribution

SN - 9798350307337

SP - 438

EP - 443

BT - 2023 WRC Symposium on Advanced Robotics and Automation (WRC SARA)

PB - IEEE

T2 - 5th WRC Symposium on Advanced Robotics and Automation (WRC SARA)

Y2 - 19 August 2023 through 19 August 2023

ER -

Deep learning based defect detection algorithm for solar panels

Abstract

Publication series

Conference

UN SDGs

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this