A Novel Framework for Solar Panel Segmentation From Remote Sensing Images: Utilizing Chebyshev Transformer and Hyperspectral Decomposition

Solar panel segmentation (SPS) is identifying and locating solar panels from remote sensing images, such as aerial or satellite imagery. SPS is critical for energy monitoring, urban planning, and environmental studies, as it can provide information on the distribution and deployment of solar energy systems and their impact on the climate and the economy. However, the existing methods face several challenges, such as low-quality remote sensing images, varying resolutions, and high computational costs. These factors make it challenging to distinguish solar panels from other objects or backgrounds and accurately and efficiently segment them. This article proposes a novel hyperspectral solar segmentation network (HSS-Net) method for SPS, combining Chebyshev transformation (CHT) and hyperspectral synthetic decomposition (HSD). Our method can enhance the image quality, select the optimal bands, and segment the solar panels. We validated the presented method on three publicly available SPS benchmark datasets, such as base de données apprentissage profond PV (BDAPPV), photovoltaic (PV), and DeepSolar. We compared the performance of HSS-Net with the state-of-the-art (SOTA) methods, including convolutional neural network (CNN)-based and transformer-based networks and existing hyperspectral segmentation techniques. We used the intersection over union (IoU), the
$F1$
-score, and the Kappa coefficient (KI) as the evaluation metrics. We demonstrated that HSS-Net significantly surpasses SOTA methods in terms of accuracy, efficiency, and scalability, can advance remote sensing applications, and may provide more precise results in various relevant fields.