Classified Identification and Estimation of behind-the-Meter Distributed Photovoltaic Panels Using High-Resolution Aerial Imagery

The continuously increasing penetration of behind-the-meter distributed photovoltaics (PV) proposes significant challenges to the operation of distributed network. Identifying the distributed PV using high-resolution aerial image is a promising and low-cost way to enhance the visibility of distributed PV. Existing studies typically establish a unified identification model, which cannot achieve satisfactory identification performance for different types of distributed PV in practice. To this end, this paper proposes a classified identification and estimation method to accurately acquire the location and size of the installed PV panels within a wide area. Firstly, K-means algorithm is adopted to cluster PV panel images with different size and installation scenarios. For each cluster, U-net is then used to build identification models where focal loss is used as the loss function for better identification of tiny-size panels. Secondly, density-based clustering algorithms and right-angle polygon fit algorithm are employed to optimize the identification results. Finally, the size of PV panels is estimated based on the segmentation masks. The effectiveness of the proposed method has been validated on a real-world aerial imagery dataset.