[dengxiaopeng]

I used to work professionally in the US doing this sort of task with computer vision. The challenging part isn’t so much the labeling/classification of a fish within an image; instead, it’s a connection to cloud environments to do processing. Most of the projects I worked on were at hydroelectric dams, where-as an ironic punchline goes—it’s punishingly hard to get access to reliable power or water.

If anyone is interested/curious happy to answer any questions on here or via DM.

[m3047]

Have seen this firsthand in (literal) field demos. Radios don't work. Field equipment is noisy, sound and electromagnetically. Just things. The "cloud" isn't effectively down the street. The nearest fiber is over a mile away. And so on.

The current brick wall to bang heads against is putting all of that under "connectivity" and working on that, and trying to sell solutions which are not as reliable as they should be; instead of doing compute more locally, or applying traditional system analysis to figure out what really needs to be realtime, and what "realtime" really means.

Would love to get a job in the field working on the compute / local telemetry solution somewhere in the Pacific Northwest USA. Definitionally I suppose that means somewhere you have to drive because there are no commercial flights; there's a quite lot of space like that, surprising / not surprising depends on your bias I suppose.

[fragmede]

Well, why is it so hard to get reliable access to power or water? I can think up reasons, but you've got the background to tell me why.

Are there innovations in material science and technology that make things possible today that weren't 20 years ago? I have a shower attachment to tell me the temperature of the water, and it's powered by water through the device so there are no batteries to swap. I imagine the industrial versions must be so much more advanced than a gadget I got off Amazon.

[dengxiaopeng]

Great question! One would intuit that the procurement function of a large organization (like one which operates a hydroelectric dam) would be able to see the value of some kind of industrial sensor that minimizes integration costs and provides them with better understanding of operations data, but practically it’s hard enough to coordinate folks to install conduit to route cables! When it requires political capital to get something as simple as cat-5 installed, clear efficiency gains easily fall off the table. It’s sort of like the infrastructure is “set in stone” and built to service the operations of the dam first, and everything else is secondary, as other commenters have suggested.

[harrisi]

It's probably a much simpler answer. Why set up a bunch of plumbing and networking inside a dam if you don't need it? Many (most?) were built before the internet, as well. And I don't think they generally are staffed around the clock. Any more complexity than you need is a waste of money.

[bombcar]

I bet it's exactly this. You've a massive concrete structure where adding water lines and power is going to require quite a bit of paperwork before you can even begin to consider how you run it. And they can be quite long, too, and your sensors likely need to be somewhere that power (and clean water) currently isn't.

And then you will discover just how well wifi travels through concrete whose thickness is measured in meters.

[ska]

> where-as an ironic punchline goes

I knew someone who grew up a reasonable walking distance from a major operational dam, but had no grid access. They did have a generator I think, but would have had to pay the full cost of running lines to their land, and it was too expensive.

[sd9]

Have you tried Starlink? I work in motorsports which has a similar problem, tracks being remote and internet access being minimal, but Starlink all but solved that for us (except unfortunately in countries without service).

[dengxiaopeng]

Late reply, but yes we had a pilot project with Starlink for a couple of locations which showed great results. Unfortunately we couldn’t use it in all cases because of stringent information security requirements; otherwise, nothing but good things to say.