Home » Technology » Planet enlists machine learning experts to parse a treasure …


Planet, the satellite imaging company that operate the largest commercial Earth imaging constellation in existence, is hosting a new data science competition on the Kaggle platform, with the specific aim of developing machine learning techniques around forestry research. Planet will open up access to thousands of image ‘chips,’ or blocks covering around 1 sauce kilometre, and will give away a total of $60,000 to participants who place in the top three when coming up with new methods for analyzing the data available in these images.

Planet notes that each minute, we lose a portion of forest the size of approximately 48 football fields, which is a heck of a lot of forest. The hope is that by releasing this data and hosting this competition, Planet can encourage academics and researchers worldwide to apply advances in machine learning that have been put to great use in efforts like facial recognition and detect, to this pressing ecological problem.

“We’re putting together this competition as a way to get people excited about the kinds of data that Planet provides,” explained Planet machine learning engineer Kat Scott in an interview. “Particularly when you’re analyzing imaging and that sort of thing, everyone works off the same sort of jpgs, but our satellites have these sort of superpowers. We get multiple bands at very high resolution, and deep bit depth, so we put together this interesting data set of all the interesting things that are gong on right now that we’d like to monitor. So things like deforestation, new agriculture, what we call artisanal mining which is basically illegal mining, and all these other effects.”

The goal is to see if competitors can come up with new ways to monitor these situations with machine learning tools created to make sense of the data. It’s a bit like finding a needle in a haystack, according to Scott, which is why the need exists for this machine learning-driven approach, taken on from multiple teams…

Source link