Tech Talk: Add Wildlife Conservation To TensorFlow's Uses


Anybody who keeps up with tech news will tell you that AI technology, especially machine learning, is pretty awesome. The list of potential ways it could develop and become useful, or even self-sustaining, is nearly limitless. The right kind of people, like coders, happen to find watching an AI do its thing quite beautiful, as well. Coincidentally, almost all of those qualities also apply to nature, in a roundabout fashion; it’s beautiful, useful to humanity, and contains endless possibilities. Those possibilities manifest in the vast array of majestic animal and plant species we have on this planet, but we have seen more than a few make their exit in the time humanity has been around. When it comes to endangered manatees, also known as sea cows, Google’s AI tech is helping to ensure that they don’t join that list, and it is one of the most clever, and probably one of the best, uses for AI that have been dreamed up thus far.

Anybody with ties to the state of Florida has a special place in their heart for manatees, but they make frequent appearances outside of the Sunshine State and serve critical roles in balancing out their local ecosystems. The majestic creatures also sometimes find themselves playing critical roles in less vital activities, such as messing up boat propellers, getting tangled in trash and nets, and lining poachers’ pockets. Unfortunately, these not-so-fun-filled activities often end with the manatee population falling. With estimates putting their remaining population at less than 5,000, keeping a close eye on the concentration and movement of manatee populations is a step in the right direction for environmentalists, and an important one at that.

Normally, keeping track of manatee populations and movements involves tracking them manually while riding in a helicopter or biplane above the water. A clever researcher, Dr. Amanda Hodgson of Murdoch University, decided to make life a bit easier for everybody, and decidedly less hazardous, by using a drone to fly over the water and take pictures of the manatees going about their business. This is all well and good, except that a typical flyover produces somewhere in the neighborhood of 45,000 high-resolution photos, where individual manatees can take time to identify because they show up so small on the photos. Knowing that there had to be a better way than wasting time and manpower to sift through the photos manually, the good doctor teamed up with Dr. Frederic Maire of the Queensland University of Technology to develop a new solution.


Enter TensorFlow. Google’s AI backbone. The open-source AI protocol can be used and modified by just about anybody for just about any purpose, including agriculture. The team set out to use TensorFlow to help them solve their photograph conundrum, and wound up coming up with an AI that’s being slowly trained to spot manatees in the aerial photos that the drones send in. As with any other AI technology based on machine learning, the technology isn’t quite perfect yet, and will take some time to improve with experience. In its current, early state, the AI is able to find about 80% of the sea cows in the photos successfully. When performance improvements and tweaks, paired with machine learning from multiple experiences, eventually gets the machine’s accuracy up to 100%, the team will be able to shave countless hours off of their research time, leaving more time to actually plan and implement the kind of conservation efforts that will cater specifically to the environmental needs of manatees. This goal is still a long way off, but when the team can devote all of their manpower to figuring out how best to serve manatee populations instead of figuring where they’re going and what they’re doing, they will be able to do a lot more good, and it’s all thanks to the magic of machine learning.