PABW-RTDETR: An Improved Defect Detection Algorithm on Solar Panel Images
IEEE Sensors Journal · 01.05.2026 · 26 (9) · 13300-13313
DOI: 10.1109/jsen.2026.3672930 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 01.05.2026. Verfasst von Dong Zhu und 3 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Gefördert von: Jiangsu Provincial Science and Technology Support Project.
Abstract (Originalsprache)
During the production and transportation of solar panels, defects such as nonelectricity, black spot, scratch, and crossed concealment inevitably arise. However, due to the diverse features, miniature size, weak texture, and irregular distribution of small-target defects, traditional models like real-time DETR (RT-DETR) demonstrate suboptimal performance. To address these challenges, a lightweight PABW-RTDETR algorithm is proposed. First, the C2f-AP module is designed by fusing asymmetric padding with the C2f structure. The asymmetric receptive fields are constructed to enhance the directional extraction of multiscale defect features, which enables the precise capture of edges and textures in tiny defects. Second, adaptive sparse selfattention (ASSA) is employed to filter redundant background information through dynamic allocation of attention weight. Therefore, key defect features are focused on, and recognition accuracy for small-target defects in complex backgrounds is improved. Third, a novel bidirectional multibranch auxiliary fusion pyramid network (BiMAFPN) module is established to optimize interaction paths of cross-scale features. Feature representation for small-target defects is improved by enhancing feature complementarity among different scales. Finally, the WIoUv3 loss function is introduced to optimize bounding box regression, increasing detection accuracy for small targets and irregular defects. Experimental results on a self-built dataset demonstrate that mean average precision (mAP) 50 is up to 82.4%, which denotes an improvement of 3.1% over the original RT-DETR. As a result of the 27.1% reduction in parameter scale, the computational load decreases to 49.6 GFLOPs and the frames per second (FPS) reaches 177.1. The model significantly reduces miss detection and false detection rates while maintaining lightweight advantages.
Förderung
- Jiangsu Provincial Science and Technology Support Project
Zitieren
APA
Zhu, D., Zhu, Y., Zhang, J., & Feng, C. (2026). PABW-RTDETR: An Improved Defect Detection Algorithm on Solar Panel Images. IEEE Sensors Journal, 26(9), 13300-13313. https://doi.org/10.1109/jsen.2026.3672930
Vancouver
Zhu D, Zhu Y, Zhang J, Feng C. PABW-RTDETR: An Improved Defect Detection Algorithm on Solar Panel Images. IEEE Sensors Journal. 2026;26(9):13300-13313. doi:10.1109/jsen.2026.3672930
BibTeX
@article{Zhu2026,
author = {Zhu, Dong and Zhu, Yaning and Zhang, Ji and Feng, Chengtao},
title = {PABW-RTDETR: An Improved Defect Detection Algorithm on Solar Panel Images},
journal = {IEEE Sensors Journal},
year = {2026},
volume = {26},
number = {9},
pages = {13300--13313},
doi = {10.1109/jsen.2026.3672930}
}
EUVIMED