As you enjoy nature this summer (or winter, depending on your hemisphere), you’ll come across a variety of birds, plants, and insects that you, admittedly, have no idea what they’re called. While packing a nature book or two is one option, this build by hacker Dave Ripp instead uses a Raspberry Pi equipped with Google’s Coral Edge TPU accelerator to ID things for you.
The device packs a Zero W, an accelerator, and a Pi Camera module into an enclosure made out of a vintage Brownie Bakelite case. While you would normally have a hard time getting good closeups of nature with the Pi Camera, the build accommodates a Tiffen 17mm threaded teleconverter lens to blow your image up by a factor of two.
With this setup, you can take pictures, then use the custom Python GUI to pass them (or even previously stored images) to the accelerator. You just need to select whether to use the bird, insect, or plant library — stored locally, so there is no need for a WiFi connection. Once started, the process takes about three seconds, held back for the most part by the Raspberry Pi’s slow USB 2.0 transfer speeds, not so much either device’s processing capabilities.
A demo of the (updated) system is seen in the first video below, while the second video elaborates further on how it works in a previous iteration.
For more information on Google’s Coral machine learning accelerator, be sure to check out Alasdair Allan’s article on the subject here.