AI excels at pattern matching and identification, even for unusually shaped objects in low-resolution photographs. The proof is a new AI model, known as U-Net, which has been developed by Swedish scientists to identify small, decentralized roof-top mounted solar power systems, using only low-resolution aerial images. This model utilizes deep learning and image processing techniques, and is reported to yield exceptional results, and is currently being programmed to differentiate between photovoltaic and solar thermal systems.
The model was recently trained and tested on databases from Germany and Sweden in order to enhance its generalization capabilities. When compared to other convolutional neural network architectures (that is, those designed to processing images or spatial data), the U-Net model outperformed them, particularly in image segmentation tasks, which involves distinguishing meaningful objects within the image.
According to the research, the U-Net model can achieve high identification accuracy when trained on aerial images with a resolution of only 128 x 128 pixels. U-Net’s ability to use such a low resolution images reduces computer hardware requirements. The Swedish study demonstrates that the U-Net model can accurately assess rooftop mounted solar energy captured in aerial images. However, to ensure correct area estimation, the tilt and skew of of the rooftop solar panels must also be considered. The scientists’ next step is to combine LiDAR data with the U-Net methodology used in this study.
This AI system is considered a valuable tool for the photovoltaic and solar thermal industries because it can provide accurate type and quantity data to policymakers, authorities, and financial evaluators. The method used in this AI model employs deep learning and image processing techniques to collectively detect solar thermal and photovoltaic roof-top systems. The authors of the study believe that the AI model can be improved to effectively distinguish between these two technologies, despite their similar texture and color appearance.
The whytry.ai article you just read is a brief synopsis; the original article can be found here: Read the Full Article…
