Near-infrared vision-guided adaptive electrostatic cleaning for energy-efficient power recovery of soiled photovoltaic panels

Dust deposition on photovoltaic (PV) panels causes power losses, and its spatial nonuniformity reduces the effectiveness of fixed-intensity cleaning. This study develops a near-infrared (NIR) vision-guided adaptive electrostatic cleaning method that converts local soiling information into zonal voltage control. NIR images were segmented using a U-Net model, transformed into a dust-coverage matrix, and mapped to dynamic voltages during electrode scanning. The U-Net model achieved an accuracy of 0.9936, a precision of 0.9629, a recall of 0.8842, an F1 score of 0.9219, and an intersection over union (IoU) of 0.8551. Under fixed-voltage cleaning, increasing dust density from 5 to 15 g m−2 reduced dust removal efficiency from 98.79% to 84.59%, while nonuniform soiling further increased local residue. Relative to the corresponding fixed-voltage results obtained under the same dust-loading and soiling-pattern conditions, the adaptive mode reduced residual soiling by 8.03%–32.76% and improved power recovery by 10.60%–34.17% under patchy soiling. Across striped, edge, and patchy soiling at the three dust-loading levels, it reduced the energy consumption per unit recovery gain by 37%–68%. The proposed method links visual soiling recognition with localized electrostatic actuation, offering a water-free and energy-efficient route for intelligent PV cleaning.