Machine learning is more prone to overfit and human learning is more prone to underfit, comment plz

Hello Christian, Juan,

We started over 3 years ago in June 2019. We developed (or “parametrized”) our initial Business Cases first and then programmed our smart RC drones using AI with Machine Learning to monitor the extensive algae growth on the lake. AI with Machine Learning code is shared between all the drones, 1 to 12.

Even though we received commercial and city permits to fly our drones over the lake, some residents still took “practice shots” on our drones. Using embedded AI with Machine Learning allowed our DJI RC drones to fly about 100 feet off the lake. The drones can “recognize” gunshots and avoid them by flying faster (or higher) in certain areas of the lake. The shooters were also photographed and reported to authorities.

We allowed our RC drones to have the ability to “self-preserve” and help each other (eg. returned to base solo or helped by another drone if anything is broken, or low on battery) and tag the next available drone to finish the monitoring task. We also allow our drones to provide better solutions WHEN monitoring the lake (eg. wind, rain, or heavy fog).

The hyperspectral camera on our drones can easily indicate whether the blue-green algae is at “Attention”, “Warning”, or “Outbreak”. It uses the light spectrum of blue-green algae to check the current status digitally.

We are using AI modules from DJI (see AI Module – Specifications – DJI).

Regards,
Santiago

1 Like

Outstanding. Thanks for sharing.

So the AI Module, which I assume contains the trained model, is connected to each drone. What drone model are you using for this task?

Can you share any more information? may be there’s literature about this interesting case? I really would like to learn more about it.

Thanks,

Juan

1 Like

I think it is a great use case. Drones can efficiently petrol a large area, though it will expose to shooters, your drones has preloaded YOLO that can recognize human to stay away from potential danger. The hyperspectral camera is added by your team to the drone I assumed? If so, it is wonderful that the drone supports and is programmable to use new hardware. I don’t know the biology of algae growth, but unless it grows faster than the time for a round trip of a drone, we might be allowed to delay any decision about it to after the drone is back to the charging station. Then experts can make decisions based on the info gathered by the drones and perhaps plan some special monitoring missions?

I have read some articles on this matter, and this is an important work that can protect drinking water sources and prevent algal bloom outbreaks. Seems drones are the most effective solutions nowadays. I wonder what are some critical pain points in using drones for algae monitoring?

1 Like

Hello Juan,

Sorry but I don’t think DJI makes our simple, and less than 250grams “programmable” RC drones anymore.

We haven’t purchased any brand new RC drones for the past 2 years, except for replacement parts, propellers, batteries, etc.

As you might have guessed, AI-RC drones have become more expensive (over $3-$5,000+) and more “professional-grade” to give them more capabilities out of the box.

If you google for AI RC drones, you may find “hobby” technologies, like Blackhawk2 from EXO for about $900. I am NOT endorsing EXO or DJI. Or if you work or consult for Google (or Boeing) they may already have several AI drones with ML capabilities for our military.

Our military is great with technology. Unfortunately it’s not for normal, non-military people like me to use.

Good luck,
Santiago

1 Like