Abstract

Partial shading in photovoltaic (PV) systems creates multiple Local Maximum Power Points (LMPPs) on the power-voltage (P-V) curve, causing conventional Maximum Power Point Tracking (MPPT) algorithms to converge prematurely to suboptimal operating points and resulting in significant energy losses. This work proposes a novel vision-based framework that integrates computer vision with Global Maximum Power Point Tracking (GMPPT) to maximize energy harvesting in partially shaded PV systems. The proposed method captures real-time imagery of PV panels to identify cell-level shading patterns, which are subsequently mapped to module electrical characteristics to enable direct analytical estimation of the GMPP voltage. An instance segmentation model performs cell-level shade quantification, and the extracted shading information is used to update a physics-based PV model that accurately reconstructs the P-V curve under non-uniform irradiance conditions. To enhance robustness against residual inaccuracies in visual inference or PV modeling, the estimated GMPP voltage is used to initialize a hybrid adaptive Perturb and Observe (P&O) algorithm that provides closed-loop refinement around the predicted operating point. Experimental results and ablation studies validate the proposed framework, achieving 97.5% accuracy in cell-level shade detection. For a representative 5 kW installation under partial shading conditions, the proposed method reduces annual energy losses by approximately 1,661.7 kWh compared to periodic optimization-based GMPPT approaches, corresponding to annual financial savings of $897 and a total hardware cost of $155 with a payback period as few as 2.1 months depending on electricity price and scan interval. Compared to recently developed GMPPT methods, the proposed approach demonstrates superior tracking speed while maintaining comparable power extraction efficiency. These results confirm the framework's robustness and its ability to overcome the fundamental limitations of conventional periodic scanning-based GMPPT techniques, enabling consistent and efficient energy harvesting under real-world partial shading conditions.

Department(s)

Electrical and Computer Engineering

Publication Status

Full Text Access

Keywords and Phrases

Computer Vision; GMPP; Instance Segmentation; Partial Shading; Photovoltaic Systems

International Standard Serial Number (ISSN)

0196-8904

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2026 Elsevier, All rights reserved.

Publication Date

15 Jun 2026

Download

Full Text Link

Share

 
COinS