AI insect identification

General discussion on entomology
Post Reply
Bub
Junior Member
Junior Member
Reactions:
Posts: 6
Joined: Tue May 24, 2022 6:56 pm
United States of America

AI insect identification

Post by Bub »

Here's a link.
https://hackaday.com/2022/07/16/neural-network-identifies-insects-outperforming-humans/

What do you think?
User avatar
nomihoudai
Junior Member
Junior Member
Reactions:
Posts: 39
Joined: Mon May 23, 2022 10:22 am
Netherlands

Re: AI insect identification

Post by nomihoudai »

Hi Bub, thanks for sharing the article.

I think that AI can and will do a lot of the insect classification in the future. It already does right now. Personally, I first use Google Lens for things that I come across when on a walk. If that doesn't give me a suitable result I ask among peers and lastly I turn to forums like this one.

In the article that you posted they touch the topic of hard to distinguish species or cryptic species. The article shows a set of Oxythyrea species. While the ones shown at the top are not hard to distinguish, they have a pair later in the article that their model wasn't able to distinguish, O. funesta and O. pantherina which look very similar. I do think that we can train AI models that can also distinguish these if we have more input images and use a multi-step approach.

Personally, I want to create a model to distinguish Colias hyale and Colias alfacarensis, and one for Pieris (rapae, napi, mannii). I do think that there must be subtle features that can tell the imagos apart. Even tough we humans have a hard time I think that the AI models should give a statistically sound answer. The only thing holding me back on this endeavour is lack of time and a properly labelled set of images of Colias hyale/alfacarensis and of the other species.

p.S. I don't understand why they bother so much about image size in your article, as CNNs work quite independently of image size. I didn't read it in full detail.
Lepidoptera distribution maps: lepimap.click
Chuck
Premium Member - 2024
Premium Member - 2024
Reactions:
Posts: 886
Joined: Mon May 23, 2022 2:30 pm
Solomon Islands

Re: AI insect identification

Post by Chuck »

The article does defer to experts as currently able to ID the more difficult determinations.

It is amazing where the technology is. For decades researchers hoped to reach this point; for decades, the archetypical example was a tiger in the grass, which humans could pick out, but "AI" (software) could not. Now it can.

It's still not perfect. One of the constraints is the image library. For example, Seek & iNaturalist cannot ID which ssp of Ornithoptera victoriae because there aren't enough source images. And in some cases (particularly common and widespread insects with multiple similar sp and ssp) the source data is garbage; garbage in, garbage out.

I often use Seek, which is aligned with (uses data from) iNaturalist. It is spectacular, and lightning fast, for identifying flowering plants in USA. "AI" isn't so much an algorithm as it is the speed of image comparisons. That said, I used Seek to identify Cat, which is black and fuzzy, and Seek identified Cat as a fox. Still some work to do.

By way of example, I recently posted an image of Fritillaries/ Speyeria cybele? Seek identified all except one as cybele; the one Eurytides called into question was not fully identified by Seek, which would only label it "Greater Fritillaries" which suggests Seek has the same challenges humans do when it comes to detail.

AI could improve insect ID by NOT using crowd sourced, and using images from institutions, but this is unlikely to happen due to cost. So to some extent, AI identification will be limited, as is almost everything, by human fault.
Post Reply

Create an account or sign in to join the discussion

You need to be a member in order to post a reply

Create an account

Not a member? register to join our community
Members can start their own topics & subscribe to topics
It’s free and only takes a minute

Register

Sign in