Plant recognition apps not as reliable as we might hope......

[Toddy]

The New Scientist has a new article on plant apps, and the problems thereof.
Worth a read through.

Apps that identify plants can be as little as 4 per cent accurate

There are many smartphone apps that aim to identify plants from photographs, but tests have found that most are not very accurate

www.newscientist.com

I think it's worth the practice, and a good heads up on potential, but I still think learning what grows where, accurately and recognisably is more worthwhile.

M

 

[Robson Valley]

Start with the trees, that's often reduced in diversity number. Lesson: Not all trees grow everywhere. There will be particular combinations. And, each of those combinations will have a characteristic understory group of plant species.

Be wary of gardens and landscapings. Many times the species are decorative ornamental specimens which come from all over the world.

 

[Souledman]

Really interesting article. It didn’t look like the app I have used was part of the review but interesting to see the one that that was most accurate one is based on a combination of algorithm and expert review.

 

[Wander]

Very interesting.
I do use one of the apps mentioned - Plant Net (and I'm pleased to see it was the most accurate).
Mind you, this was recommended to me by a conservationist friend who works for a Wildlife Trust and whose specialism is plants and grasses!
He said he found it useful and worth having.
Go figure.
Mind you, Plant Net (and I don't know about the others) works by comparing your picture with its library of 'approved' images. And these 'approved' images are ones approved by the user. So, for example, I will take a picture and up load it for confirmation. The app then comes back with most likely species (in descending order). In fairness, Plant Net never says 'that is...such-and-such', it always gives a percentage likelihood - e.g. 88%, etc, so it does entertain doubt. Once I find the right species I can then (and it is only an option) 'agree' with it and my picture is added to the library.
Which I've always thought is fundamentally flawed since if I am using the app for confirmation then, by definition, my knowledge is not of a required level for me 'approve' anything.
If you see what I mean.
For that reason I never, officially, 'agree' with any of the suggestions since I may pollute the collective knowledge.
I've always been dubious about that aspect.
That said, I've always come away happy with the results and when you find the species you think your sample may be there is lots of other images to look at for variation. Oh, and the app also uses your location as a filter for its results. Though, as highlighted in the report, that may not help when it comes to introduced or invasive species.
A very interesting report indeed.
I'm still sufficiently impressed by Plant Net that I'll keep using it (I've learnt a lot from it as well), but it does underline my own caution.
I tend to use these apps as a field guide only. When I get home I will double-check the image in a couple of different guides to ensure I got the right result.

 

[Kadushu]

Like everything: it's another tool for the toolkit which can compliment books, courses, etc. Don't treat it as the be all and end all and you'll be fine. As with most things, learning plants gets easier with practice.

One thing I used to joke with my late friend about (we both studied horticulture at university) was how people who know a handful of plant names then assign those names to everything that looks vaguely similar. For example every large leafed herbaceous perennial gets called a hosta and every broadleafed evergreen gets called a laurel. Actually that latter point arises a lot in common names. The only true laurel that you'll commonly find in the UK, Laurus nobilis, is called 'bay' whereas Japanese laurel, cherry laurel and laurestine are Aucuba, Prunus and Viburnum respectively.

 

[Broch]

I have had some success when trialling 'Seek' but I certainly wouldn't trust it. I agree with Wander, any app that allows the user to affect the certainty by putting their own confirmation in is as flawed as Amazon's review system

I have found Seek will get to the right family, often the right genus, and usually the right species. So, it will very often get you to the right section or even page in a good field guide (Harrap, Rose, or Blamey etc.)

 

[Brizzlebush]

An interesting article, thank you Toddy.

I like Wander's point about not always agreeing with an identification, which could pollute the collective knowledge.
As Broch says, it's a little like an am#@*n review, and prone to subjectivity, and user error.

I use Plantnet and Flora Incognita. Both are flawed, but pretty good. So I've found if I use both on the same plant and cross-reference the results, I can get pretty close to an accurate i.d. Then I'll go home and hit the books to make sure.

I agree, that they're handy in the field, but shouldn't be 100% relied upon.
Quite fun though and the pictures can be a nice reference point.

There's something in my Internet Paranoia (with sharing the location of these plants) that makes me wonder if the location of, say, a rare plant, is a good thing?

 

[Toddy]

That's a really good point. Ought any plant on the 'red list' really be so easily found ? or in finding more and noting where do we maybe encourage some kind of project to assist regeneration ?

I don't know the answer. I do know I think I'd be happier if there were something like the RHS, or some other academic/ farming organisation, overseeing an app like the ones we've spoken about.

Good amateurs are a brilliant and under-rated resource though, but it's a bit like Wikipedia, it needs not only to be editable, but from multiple viewpoints.

I don't know the answer, I don't wander far these days, but I still find it of interest to spot different things growing

 

[Suffolkrafter]

Plant net has been a great help in my plant I.D journey. It gives me pointers and ideas, which I can then fully confirm in books or elsewhere on the web. It provides something of a shortcut, and being always short on time, I need this. Over time I've got a feel for when I can trust it and when not. One area I've found it less reliable is with the apeaciae family, or umbelifers. Otherwise I think it's a great tool so long as you're aware of its limitations.

 

[TinkyPete]

I have used some apps in the past, but I have found reference books a lot more reliable, and there are a lot of good ones which are quite light as well and do not rely on batteries. Always get one suitable for your region.

 

[Suffolkrafter]

Came across my favourite patch of poison hemlock on my lunchtime walk today and thought I'd test the plant net app. Based on a photo of the leaves, it came up with 25 percent probability of it being poison hemlock (conium maculatum). But I think it's a tough ask. It illustrates that you need to know their limitations.

 

[SaraR]

my favourite patch of poison hemlock

Sounds....worrying

 

[Toddy]

I think so too.
The thing is, we, as in this kind of community, know that the white umbellifleurs are both tasty and potentially deadly. We are rightfully wary and very careful of our identification of them.....how about someone not so aware, with an 'app' out for a wander and it says wild carrot ?

Okay, I get that that's maybe a little unlikely, but thinking on it, is it ?
Wild carrot, hemlock ?...

I think having such a resource in a Star Trek like all in one communicator is a brilliant thing, but the reality is that sometimes it still needs thought, it still needs care, it still needs an application of common sense.....which unfortunately seems to be becoming more and more uncommon

I think for the apps to truly be reliable needs a lot more input from people who do know what they're looking at, what they're actually showing, but I wonder how many will bother to input and update?
I know I haven't even contributed to the one for my own area. It's not laziness, just I never have my phone on if I'm out and about. It's only for phoning home to say where I am or if I'm going to be late. It's not plugged into my mind kind of thing.

 

[SaraR]

The thing is, we, as in this kind of community, know that the white umbellifleurs are both tasty and potentially deadly. We are rightfully wary and very careful of our identification of them.....how about someone not so aware, with an 'app' out for a wander and it says wild carrot ?

Okay, I get that that's maybe a little unlikely, but thinking on it, is it ?
Wild carrot, hemlock ?...

I avoid this group for that very reason. Unless I am 100% sure about the identification of something AND I know what the potential dangerous look-alikes are in that region and how to tell them apart I just avoid them altogether. It's just not worth the risk. For example, there are plenty of species that I would be confident in identifying, but since I don't always know which other similar species are likely in the UK, I remain a bit cautious.

If I do want to be sure about something, I will take the time to key them out, but even that is not always so easy depending on time of year and group of plants.

 

[Suffolkrafter]

There's only so much you can mitigate for lack of common sense. I've even seen respected YouTubers confuse gorse with broom and white dead nettle for stinging nettle.

 

[Broch]

As I've said before, a well known foraging book has miss-labelled conium maculatum in one edition

Here's a summary spread sheet for anyone interested - no liability accepted

 

[demented dale]

The New Scientist has a new article on plant apps, and the problems thereof.
Worth a read through.

Apps that identify plants can be as little as 4 per cent accurate

There are many smartphone apps that aim to identify plants from photographs, but tests have found that most are not very accurate

www.newscientist.com

I think it's worth the practice, and a good heads up on potential, but I still think learning what grows where, accurately and recognisably is more worthwhile.

I asked a friend to ID a plant using a phone app. It said the plant was celery. It WAS NOT celery. It was Hemlock water dropwort. I think we need to check and double check and triple check until we are certain. Particularly with the umbellifer family. Better to be safe than sorry. These plant apps can be helpful but they are not 100% reliable and need to be used with other resources like books and online information and asking experts. .

 

[demented dale]

I asked a friend to ID a plant using a phone app. It said the plant was celery. It WAS NOT celery. It was Hemlock water dropwort. I think we need to check and double check and triple check until we are certain. Particularly with the umbellifer family. Better to be safe than sorry. These plant apps can be helpful but they are not 100% reliable and need to be used with other resources like books and online information and asking experts. .

 

[Nice65]

Any app for identifying a plant is only being asked to have a go at just that. They're not for identifying potential foods or plants to avoid eating.

 

[Pattree]

At College we were taught to use (deep breath): Tooting Clapham and Warburg for plant identification. I promise you that was far worse at identifying plants - you had to get every botanical element right or it sent you off on the wrong track. To test it I tried to identify the common Daisy (Bellis perenis ) with a sample in front of me. I ended up with a tree paeony.

I read the NS article. Was the best app something like 90% accurate. Others far worse.
Hang on - one of the things I’d lose with this phone is NS!

Got it:
The best was PI@ntNET @ 88.2% accurate.
I think I’d be careful if there were a 10% chance lunch would kill me.