A team of researchers from Stanford, came up with an AI solution, in an attempt to count the number of solar panels in the US to gauge the spread of the technology. The researchers developed a deep learning system, DeepSolar, that has the capability to map every visible solar panel in the US – which by the way stands at a whopping 1.47 million.
The researchers leveraged the neural network technology to transform satellite-based imagery of an area into tiles. Further, by classifying every pixel within those tiles and combining them, the researchers were successful in determining if solar panels have been in a given area or not, let it be large solar farms or individual rooftop installations.
The approach gives accurate and much faster results and requires only some basic oversight. The Stanford researchers were able to map the country in just a few weeks, which, in a conventional approach would take years to complete the assessment, and even then the results might be outdated. This deep learning system can help governments take important decisions on nation’s renewable energy strategies, track solar adoption rates of the general public, and even pinpoint economic differences based on the number of panels in a given neighborhood.
The researchers disclosed that California and the Southwest have the highest concentrations of solar panels, using the DeepSolar system. Based on data revealed by the system, the researchers were able to pinpoint locations which are ideal for panel deployment –basically, any place above a given sunlight level is virtually guaranteed to have panels. Income played a determining factor for installation of these panels, even in places with abundant sunlight, suggesting companies to re-strategize their approach if they expect their market presence.