Hacker News Reader

Comment by brk | Hacker News Reader

Short summary is that this isn't going to work. I was involved with a company doing some development on whale detection AI. The vast majority of the time you just don't see enough of the animal to make reliable detections. And that is in a best case scenario, add in movement of the boat/camera, waves, sunlight reflections, and having to ignore a vast majority of potential things on/in the water that can look kind of like whales, but are not whales. Dolphins in particular.

This isn't a problem for AI/machine vision, IMO.

View on HN

Acoustic detection methods using hydrophone arrays can supplement visual AI by identifying whale vocalizations at distances of several kilometers even in poor visibility conditions.

Metal whale vocalizations especially. One might say it's the primary application.

Ok, but whales move, sometimes very quickly. If we assume that the whales are moving, that they are not sleeping, the ship might want to initially aim directly at them. By the time the ship gets there, they will have moved.

That’s your opinion not a summary of the article though

Correct, I didn’t mean to imply it was a summary of the article but my comment could be confusing I guess. It’s my opinion based on 18 years in various AI machine vision startups, and having worked on this specific problem.

Thanks for clearing that up. I trust what you know what you are talking about.

A comment section is logically for people to share their opinions. We can all read the article.

They use "heat-sensing cameras", so what does movement of the waves, reflections, and not seeing the animal have to do with that? Did you use the same type of sensors as them?

"heat sensing cameras", meaning thermal cameras. They solve some problems, but create other ones. You lose any color channel data, resolution is low, and in many cases you get reduced thermal contrast if the object you are trying to detect is close to the temperature of the surrounding environment. This can definitely be a problem for whales or marine animals.

Along these lines wouldn’t it be better to require ships to add some forward facing underwater radar to watch out for whale strikes? Or am I missing something here?

Radar is ineffective underwater because the radio waves are absorbed by the water. That is why there's no radar in use on submarines.

Sonar is not a valid solution here because it annoys the whales. Or at least high-power sonar, as used by the military, does. It would defeat the point.

Sonar is the solution; passive sonar using numerous hydrophones to chart the bearing of the whale. How that bearing changes over time allows you to deduce if you're on a collision course. This is tech developed by the military.

Even if they could watch out there is very little most merchant ships or cruise liners could do if a whale crossed their path, it takes too much time to alter course or stop.

Would an upgraded passive sonar supported by AI make sense for cargo vessels? Is it too expensive to maintain and operate?

Reply on news.ycombinator.com