[NatureNS] AI and species ID?

DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=dalu.onmicrosoft.com;
From: Stephen Shaw <srshaw@Dal.Ca>
To: "naturens@chebucto.ns.ca" <naturens@chebucto.ns.ca>
Thread-Topic: [NatureNS] AI and species ID?
Thread-Index: AQHUxuVItcdBv2bghkOWa2EHVuDXp6XkQreAgAAOGXmAABUlgIAAUEcA
Date: Mon, 18 Feb 2019 00:36:40 +0000
References: <DM6PR18MB250600B7D86A51E6B12F177DB7620@DM6PR18MB2506.namprd18.prod.outlook.com>
Accept-Language: en-US
authentication-results: spf=none (sender IP is ) smtp.mailfrom=srshaw@Dal.Ca;
received-spf: None (protection.outlook.com: Dal.Ca does not designate
Precedence: bulk
Return-Path: <naturens-mml-owner@chebucto.ns.ca>
Original-Recipient: rfc822;"| (cd /csuite/info/Environment/FNSN/MList; /csuite/lib/arch2html)"

next message in archive
next message in thread
previous message in archive
previous message in thread
Index of Subjects

&lt;div&gt;&lt;br&gt;
--_000_6730EB7AFB49442FB60CCFF336445A52dalca_
Content-Type: text/plain; charset="Windows-1252"
Content-Transfer-Encoding: quoted-printable

Re Dave=92s comment, whatever your ultimate interest, the thing you need at=
 some point is a species ID.

The current criterion used in the interesting link Ian provides is for at l=
east 20 good images of a species to be available that have been vetted by e=
xperts, for training the AI network so that it can identify a new image to =
a fairly high level of reliability, presumably from a variety of viewing an=
gles.  Given the popularity of digital cameras this could be useful for N A=
merican birds (most quite large, not too many species, hoards of amateur ob=
servers some of whom are expert photographers).  As for insects, it should =
work also for NS butterflies (few species, all IDs known except perhaps for=
 occasional incoming non-native strays, a few expert photographers, main ID=
 by wing pattern & colour (=3D easy), existing web site).

Re flies mentioned by Fritz, far less likely to happen for such much larger=
 Orders =97 far too many species, many too small to photo without special l=
enses or a microscope, fewer interested observers.  A few fly groups have l=
arger, photogenic species (families like syrphids =97 hoverflies) and BugGu=
ide.net<http://BugGuide.net> has hundreds of images of these (~236 come up =
on the page for the large, common alien hoverfly Eristalis tenax, though ~5=
% of images there are actually other species that have leaked into the E. t=
enax page).   Many flies, like tachinids, a huge group, have species that a=
re identifiable, if at all, by bristle patterns on particular cuticular pla=
tes that would need posed photographs by an expert down a microscope to see=
.

A larger problem is that whole blocks of the larger insect groups are not e=
ven fully or satisfactoriiy classified (or even identified) beyond the fami=
ly or genus level yet, to species.  This is likely to alter only very slowl=
y in the foreseeable future =97 few professional taxonomists, given more se=
emingly attractive career alternatives, just too many unsatisfactory insect=
 groupings which professionals assert need revision.  If you want a fly=92s=
 ID, your best initial bet is BugGuide and ask for help.

Steve (Hfx)
-------------------------------------------------------------

On Feb 17, 2019, at 3:49 PM, Ian Manning <ianmanning4@gmail.com<mailto:ianm=
anning4@gmail.com>> wrote:

Hi Fritz,

In the last six months or so, the inaturalist website added machine learnin=
g functionality to image observations, so now when an image is uploaded it =
will give you suggestions based on their machine learning image recognition=
 model, in the video I sent before around the 3-minute mark, Alex Shepard s=
ays the model can ID to common ancestor (higher level taxon) with an accura=
cy of 93%. Each time an observation is uploaded and vetted as "research gra=
de", that image is added to the image analysis training library (used to tr=
ain the model and make it more accurate down the line).

There's lots of other examples of Computer Science folks using Machine Lear=
ning for this type of thing for citizen science (see Merlin like Rick menti=
oned), and also a bunch of more specific/academic purposes, though as far a=
s I know, none are as accessible or ambitious as iNat. Here's some more inf=
o on the computer vision side of things. https://www.inaturalist.org/pages/=
computer_vision_demo


Cheers,
IM



On Sun, 17 Feb 2019 at 15:07, Fritz McEvoy <fritzmcevoy@hotmail.com<mailto:=
fritzmcevoy@hotmail.com>> wrote:
HI Ian,
     I've been on iNaturalist for a couple of years now and it is a great w=
ebpage. There are lots of sites like iNaturalist online now and they are re=
ally helpful for having others help you identify a species.
     The thing I'm asking about is a program that would allow someone to up=
load a photo and/or maybe some data (size, colour, location and whatever ot=
her ID info you have) of a fly, a leaf or a mouse etc. and have a AI comput=
er program take that photo and/or data and give you its guess as to Genus/s=
pecies identification based on it's algorithms - much like facial recogniti=
on and retina scans work today. All the best.
                        Fritz McEvoy
________________________________
From: naturens-owner@chebucto.ns.ca<mailto:naturens-owner@chebucto.ns.ca> <=
naturens-owner@chebucto.ns.ca<mailto:naturens-owner@chebucto.ns.ca>> on beh=
alf of Ian Manning <ianmanning4@gmail.com<mailto:ianmanning4@gmail.com>>
Sent: February 17, 2019 1:43 PM
To: naturens@chebucto.ns.ca<mailto:naturens@chebucto.ns.ca>
Subject: Re: [NatureNS] AI and species ID?

Try uploading a photo to iNaturalist and seeing what it comes up with. I th=
ink you'll be impressed.

Cheers,
IM


--
We're mapping research in Southwest Nova Scotia.
Link to your research project/publications here<https://goo.gl/forms/1JkoBR=
WIP3Kbvxl53>.



--_000_6730EB7AFB49442FB60CCFF336445A52dalca_
Content-Type: text/html; charset="Windows-1252"
Content-ID: <67CF34AC83B94E44802D79CB8AB5354E@CANPRD01.PROD.OUTLOOK.COM>
Content-Transfer-Encoding: quoted-printable

<html>
<head>
<meta http-equiv=3D"Content-Type" content=3D"text/html; charset=3DWindows-1=
252">
</head>
<body style=3D"word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-lin=
e-break: after-white-space;">
Re Dave=92s comment, whatever your ultimate interest, the thing you need at=
 some point is a species ID.
<div>&nbsp;&nbsp;
<div>The current criterion used in the interesting link Ian provides is for=
 at least 20 good images of a species to be available that have been vetted=
 by experts, for training the AI network so that it can identify a new imag=
e to a fairly high level of reliability,
 presumably from a variety of viewing angles. &nbsp;Given the popularity of=
 digital cameras this could be useful for N American birds (most quite larg=
e, not too many species, hoards of amateur observers some of whom are exper=
t photographers). &nbsp;As for insects, it
 should work also for NS butterflies (few species, all IDs known except per=
haps for occasional incoming non-native strays, a few expert photographers,=
 main ID by wing pattern &amp; colour (=3D easy), existing web site). &nbsp=
;
<div><br>
</div>
<div>Re flies mentioned by Fritz, far less likely to happen for such much l=
arger Or