Marine Photovoltaic Module Salt Detection via Semantic-Driven Feature Optimization in Mask R-CNN

Abstract

Offshore floating photovoltaic systems are highly susceptible to salt crystallization on the surfaces of photovoltaic modules, highlighting the need for intelligent inspection and cleaning technologies to improve operational efficiency and overcome the limitations of conventional manual maintenance methods. However, the presence of surface gridlines on the photovoltaic modules introduces significant visual interference, which complicates the accurate identification of salt deposition regions. To address this challenge, a semantic information-guided detection framework is proposed to enable precise segmentation of salt-affected areas. The key innovation lies in the effective classification of gridlines as background features by extracting semantic priors through low-level thresholding, which are then fused with the original red-green-blue image to construct a four-channel input. This fusion enhances the model’s ability to extract and discriminate features related to salt crystallization. Experimental results demonstrate that the proposed method achieves a 4.6% improvement in segmentation accuracy and a 3.7% increase in recognition accuracy compared to conventional models, based on evaluation metrics such as mean average precision and F1-score. The proposed framework offers a robust technical foundation for developing intelligent maintenance systems tailored to offshore floating photovoltaic applications.

Keywords

Floating photovoltaic system / Salt deposition detection / Image segmentation / Semantic information / Mask R-CNN / Otsu threshold

Declaration of Generative AI and AI-Assisted Technologies in the Writing Process

During the preparation of this work the author(s) used ChatGPT in order to translate the manuscript. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.

Acknowledgements

This study is supported by the China Postdoctoral Science Foundation (No.2024T022TJ, No.2025M773253), National Key R&D Program of China (No. 2022YFB4200704), National Science Foundation of Tianjin Science and Technology Plan Project (Grant NO: 23ZYQYGX00140).

Author Contributions

X.G. and M.L.; Methodology, M.L.; Software, Y.L.; Validation, X.D., Q.M. and Y.L.; Formal Analysis, Q.M.; Investigation, X.D.; Resources, M.L.; Data Curation, X.X.; Writing—Original Draft Preparation, X.D.; Writing—Review & Editing, Q.M.; Visualization, X.D.; Supervision, X.G.; Project Administration, J.L.; Funding Acquisition, J.L.

Ethics Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data will be made available on request.

Funding

This research received no external funding.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References

[1] Mohammadi A, Mehrpooya M. A comprehensive review on coupling different types of electrolyzer to renewable energy sources. Energy 2018, 158, 632-655. doi:10.1016/j.energy.2018.06.073.

[2] Wall S, Hong X, Sha L, Xie J. High-efficiency PV inverter with SiC technology. IET Renew. Power Gener. 2018, 12, 149-156. doi:10.1049/iet-rpg.2017.0331.

[3] Venkateswari R, Rajasekar N. Power enhancement of PV system via physical array reconfiguration based Lo Shu technique. Energy Conv. Manag. 2020, 215, 22. doi:10.1016/j.enconman.2020.112885.

[4] Pietzcker RC, Stetter D, Manger S, Luderer G. Using the sun to decarbonize the power sector: The economic potential of photovoltaics and concentrating solar power. Appl. Energy 2014, 135, 704-720. doi:10.1016/j.apenergy.2014.08.011.

[5] Zhao Q, Su C, Qin M, Umar M. Is global renewable energy development a curse or blessing for economic growth? Evidence from China. Energy 2023, 285, 11. doi:10.1016/j.energy.2023.129403.

[6] Gorjian S, Sharon H, Ebadi H, Kant K, Scavo F.B, Tina G.M. Recent technical advancements, economics and environmental impacts of floating photovoltaic solar energy conversion systems. J. Clean. Prod. 2021, 278, 28. doi:10.1016/j.jclepro.2020.124285.

[7] Gao X, Li Y, Liu M, Lian J, Ma Q, Zhang J, et al. An exploratory framework for analyzing the impact of salt deposition on offshore photovoltaic system, Renew. Energy 2025, 242, 11. doi:10.1016/j.renene.2025.122439.

[8] Zhang Y, Yuan C. Effects of marine environment on electrical output characteristics of PV module. Renew. Sustain. Energy J. 2021, 13, 9. doi:10.1063/5.0060201.

[9] Nie J, Luo T, Li H. Automatic hotspots detection based on UAV infrared images for large-scale PV plant. Electron. Lett. 2020, 56, 193-194. doi:10.1049/el.2020.1542.

[10] Ullah A, Amin A, Haider T, Saleem M, Butt N.Z. Investigation of soiling effects, dust chemistry and optimum cleaning schedule for PV modules in Lahore, Pakistan. Renew. Energy 2020, 150, 456-468. doi:10.1016/j.renene.2019.12.090.

[11] Lu K, Li J, Zhou L, Hu X, An X, He H. Generalized Haar Filter-Based Object Detection for Car Sharing Services. IEEE Trans. Autom. Sci. Eng. 2018, 15, 1448-1458. doi:10.1109/TASE.2018.2830655.

[12] Song Z, Yang J, Zhang D, Wang S, Li Z. Semi-Supervised Dim and Small Infrared Ship Detection Network Based on Haar Wavelet. IEEE Access 2021, 9, 29686-29695. doi:10.1109/ACCESS.2021.3058526.

[13] Shirpour M, Khairdoost N, Bauer MA, Beauchemin SS. Traffic Object Detection and Recognition Based on the Attentional Visual Field of Drivers. IEEE T. Intell. Veh. 2023, 8, 594-604. doi:10.1109/TIV.2021.3133849.

[14] Cao T, Song K, Xu L, Feng H, Yan Y, Guo J. Balanced multi-scale target score network for ceramic tile surface defect detection. Measurement 2024, 224, 12. doi:10.1016/j.measurement.2023.113914.

[15] Xu S, Tang H, Li J, Wang L, Zhang X, Gao H. A YOLOW Algorithm of Water-Crossing Object Detection. Appl. Sci. 2023, 13, 15. doi:10.3390/app13158890.

[16] Guo H, Bai H, Qin W. ClouDet: A Dilated Separable CNN-Based Cloud Detection Framework for Remote Sensing Imagery. IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. 2021, 14, 9743-9755. doi:10.1109/JSTARS.2021.3114171.

[17] Cao Z, Li Z, Fang L, Li J, Yang H, Hui D. Research on coal and gangue segmentation based on MFCCM - Mask R - CNN. Energy Sci. Eng. 2024, 12, 2958-2973. doi:10.1002/ese3.1797.

[18] Nam K, Lee D, Kang S, Lee S. Performance evaluation of mask R-CNN for lung segmentation using computed tomographic images. J. Korean Phys. Soc. 2022, 81, 346-353. doi:10.1007/s40042-022-00532-9.

[19] Zhang C, Zhou J, Wang H, Lian J, Ma Q, Zhang J, et al. Multi-Species Individual Tree Segmentation and Identification Based on Improved Mask R-CNN and UAV Imagery in Mixed Forests. Remote Sens. 2022, 14, 22. doi:10.3390/rs14040874.

[20] Jia X, Ai B, Deng Y, Xu X, Peng H, Shen H. PC2D simulation and optimization of the selective emitter solar cells fabricated by screen printing phosphoric paste method. Jpn. J. Appl. Phys. 2015, 54, 9. doi:10.7567/JJAP.54.08KD23.

[21] Chanchangi YN, Ghosh A, Sundaram S, Mallick TK. An analytical indoor experimental study on the effect of soiling on PV, focusing on dust properties and PV surface material. Sol. Energy 2020, 203, 46-68. doi:10.1016/j.solener.2020.03.089.

[22] Khan Z, Yang J. Image segmentation via multi dimensional color transform and consensus based region merging. Multimed. Tools Appl. 2019, 78, 31347-31364. doi:10.1007/s11042-019-07906-5.

[23] Sagar P, Upadhyaya A, Mishra SK, Pandey RN, Sahu SS, Panda G. A Circular Adaptive Median Filter for Salt and Pepper Noise Suppression from MRI Images. J. Sci. Ind. Res. 2020, 79, 941-944.

[24] Moghaddasi H, Hendriks RC, van der Veen A, de Groot NMS, Hunyadi B. Classification of De novo post-operative and persistent atrial fibrillation using multi-channel ECG recordings. Comput. Biol. Med. 2022, 143, 14. doi:10.1016/j.compbiomed.2022.105270.

[25] Yang GJ, Huang TS. The effect of median filtering on edge location estimation. Comput. Graphics Image Proc. 1981, 15, 224-245.

[26] Arias-Castro E, Donoho DL. Does Median Filtering Truly Preserve Edges Better Than Linear Filtering? Ann. Stat. 2009, 37, 1172-1206, 10.1214/08-AOS604.

[27] Wang P, Kong M, Wang L, Ni L. The Effect of the Superhydrophobic Film on the Generation Efficiency of Photovoltaic Modules Affected by Salt-Containing Dust Deposition. IEEE J. Photovolt. 2019, 9, 1727-1732. doi:10.1109/JPHOTOV.2019.2930909.

[28] Anzaku ET, Wang H, Babalola A, Van Messem A, De Neve W. Re-assessing accuracy degradation: a framework for understanding DNN behavior on similar-but-non-identical test datasets. Mach. Learn. 2025, 114, 22. doi:10.1007/s10994-024-06693-x.

[29] Zhong Y. A Theory of Semantic Information. China Commun. 2017, 14, 1-17.

[30] Lin KC. On improvement of the computation speed of Otsu 's image thresholding. J. Electron. Imaging 2005, 14, 12. doi:10.1117/1.1902997.

[31] Gatos B, Pratikakis I, Perantonis SJ. Adaptive degraded document image binarization. Pattern Recognit. 2006, 39, 317-327. doi:10.1016/j.patcog.2005.09.010.

[32] Garcia G, Herman I. Advances in Computer Graphics: Images:Synthesis, Analysis, and Interaction; Springer: London, UK, 1986.

[33] Liu C, Xie F, Dong X, Gao H, Zhang H. Small Target Detection From Infrared Remote Sensing Images Using Local Adaptive Thresholding. IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. 2022, 15, 1941-1952. doi:10.1109/JSTARS.2022.3151928.

[34] Liu Z, Bai X, Sun C, Zhou F, Li Y. Infrared ship target segmentation through integration of multiple feature maps. Image Vis. Comput. 2016, 48-49, 14-25, 10.1016/j.imavis.2015.12.005.

[35] AlSaeed D, Bouridane A, El-Zaart A. A Novel Fast Otsu Digital Image Segmentation Method. Int. Arab J. Inf. Technol. 2016, 13, 427-434.

[36] Bouwmans T, Javed S, Zhang H, Lin Z, Otazo R. On the applications of robust pca in image and video processing. Proc. IEEE 2018, 106, 1427-1457.

[37] He K, Gkioxari G, Dollar P, Girshick R. Mask R-CNN. IEEE Trans. Pattern Anal. Mach. Intell. 2020, 42, 386-397. doi:10.1109/TPAMI.2018.2844175.

[38] Ren S, He K, Girshick R, Sun J. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Trans. Pattern Anal. Mach. Intell. 2017, 39, 1137-1149. doi:10.1109/TPAMI.2016.2577031.

[39] Khan M.A, Akram T, Zhang Y, Sharif M. Attributes based skin lesion detection and recognition : A mask RCNN and transfer learning - based deep learning framework. Pattern Recognit. Lett. 2021, 143, 58-66. doi:10.1016/j.patrec.2020.12.015.

[40] Perez-Borrero I, Marin-Santos D, Gegundez-Arias ME, Cortes-Ancos E. A fast and accurate deep learning method for strawberry instance segmentation. Comput. Electron. Agr. 2020, 178, 12. doi:10.1016/j.compag.2020.105736.

[41] Kumar NM. Subramaniam U, Mathew M, Ajitha A, Almakhles D.J. Exergy analysis of thin-film solar PV module in ground-mount, floating and submerged installation methods. Case Stud. Therm. Eng. 2020, 21, 11. doi:10.1016/j.csite.2020.100686.

[42] Lee MCH, Braet J, Springael J. Performance Metrics for Multilabel Emotion Classification : Comparing Micro, Macro, and Weighted F1-Scores. Appl. Sci. 2024, 14, 21. doi:10.3390/app14219863.

[43] Wang X, Wang S, Guo Y, Jia X, Hu K, Cheng G. Multi-scale coal and gangue detection in dense state based on improved Mask RCNN. Measurement 2023, 221, 15. doi:10.1016/j.measurement.2023.113467.