There are all types of air high quality sensors in the marketplace that depend on all types of electro-physical results to detect gases or contaminants and report them again as a worth. [lucascreator] has as an alternative been investigating a way of figuring out air high quality that’s nearer to divination than measurement—utilizing pc imaginative and prescient and a skilled AI mannequin.
The system depends on an Unihiker K10—a microcontroller module primarily based across the ESP32-S3 at coronary heart. The chip is operating a light-weight convolutional neural community (CNN) skilled on 12,000 pictures of the sky. These pictures had been sourced from a public dataset; they had been taken in India and Nepal, and tagged with the related Air High quality Index on the time of seize. [lucascreator] used this information to coach their mannequin to take a look at a picture taken with a digicam connected to the ESP32 and estimate the air high quality index primarily based on what it has seen in that present dataset.
It would sound like a spurious idea, however it does have some worth. [lucascreator] cites research the place video information was used for low-cost air high quality estimation—not as a alternative for correct measurement, however as an extra information level that may very well be sourced from present surveillance infrastructure. Efficiency of such fashions has, in some instances, been remarkably correct.
[lucascreator] is pragmatic concerning the limitations of their implementation of this idea, noting that their very compact mannequin didn’t at all times carry out one of the best when it comes to figuring out precise air high quality. The idea might have some worth, however implementing it on an ESP32 isn’t really easy if you happen to’re searching for supreme accuracy. We’ve featured another nice air high quality tasks earlier than, although, if you happen to’re searching for different methods to seize this data. Video after the break.
