Not surprisingly, satellite imagery and computer vision are a match made in heaven, spawning useful new insights and applications on a regular basis. Now engineers at Stanford University have developed a deep learning model that scans satellite imagery to detect the size and number of every solar panel in the United States. The model found 50 percent more solar panels than previous surveys, but the findings get really interesting when correlated with other data, such as amount of sunlight and neighborhood demographic information. For example, solar panels are more prevalent in areas with higher incomes—but only up to $150,000 per year, after which there is a significant drop off—suggesting that not many of the megamansions of the rich and famous are making use of renewable fuels, much less giving it back. Either way, the findings generated by this technology can be used by governments and utility companies to incentivize residents of sun-rich-but-solar-panel-light neighborhoods, to go greener, as well as increase or reduce traditional fossil-fuel power output on days when it’s cloudy or sunny, respectively.
sign up for our newsletter