AntMatcher dataset

This project has been taken off-line as outdated.

From the diary of an ant modeler

This project has been taken off-line as outdated.

AntMatcher page updated

This project has been taken off-line as outdated.

More results or better results?

This project has been taken off-line as outdated.

The less efficient AntMatcher blogger introduces himself

This project has been taken off-line as outdated.

Origins of AntMatcher

AntMatcher - automated ant identification system - was born from my PhD struggle to identify Malagasy ants and from Eli Parviainen's genius to make computers do things she wants them to do.

For my PhD, I need to identify tones of ants collected in Madagascar. I was getting rather stressed that I will never be able to recognize them. The identification was just too time consuming. To identify one ant you need to check a number of different characters on various parts of the body using ant identification keys or the amazing AntWeb. Even though, if you are not professional ant taxonomist to identify one ant can take you a day(s). Seeing that this process is extremely slow, I started to think would there be an easier way to read all those characters. I thought: "could a computer read all these characters?" I noticed that many characters to recognize an ant species are used from head characters (of course not all, but quite many). Naturally, I started to think if only ant head could be used to identify an ant. I imagined how great it could be if you could snap a head image of an ant, upload it online, match your picture with AntWeb images and a computer would tell you what that ant is! Everyone is happy; I am happy I have a name, ant is happy to continue its errands without being marinated in ethanol. I knew there must be an easier way to perform routine ant identification, but how to do that was still a mystery. All these ideas, however, remained my dreams for some time. I continued to spend countless hours identifying ants one by one. With no background in programming or any kind of software knowledge, my ideas would always have remained plans. I knew I needed someone who can make computers “see” and “talk”.

I was fortunate enough and the event called BCH "Connecting scientists" vol2 (2012) reached my ears. The event was aiming in connecting biologists with more technology-oriented people. After my  three minute talk, I discussed my idea with different people, but not much interest around. I was not a surprise though. People there were talking brains and things; me, I was running and trying to sell ants. However, I still had one person behind my sleeve I really wanted to talk to. I read Eli Parviainen’s profile before coming to the meeting and I thought she would be an interesting person to talk with. Finally, I had a chance to talk with here and I my judgment was absolutely right. Her enthusiasm and interest in the idea, and her zeal for a color splotches brought us where we are with the project now. We are preparing a manuscript, we have published an online system were you can actually try the identification.

Meeting Eli was great blessing. No, it didn’t solve my PhD problem, I still have piles of identified ants. You also can’t yet use AntMatcher to have 100% accuracy ant identification yet. However,  in its infancy phase, the program can help you to make AntWeb database more manageable, especially if you just starting identification for the first time in your life. AntMatcher can narrow down your search to for example 15 closest genus/species matches. This is something manageable compared with for example a million possible matches directly in AntWeb. Our system in future could also be successfully employed in for example insect collection inventory or pest management programs. The program will improve identification accuracy with more and more ant images added to AntWeb. The more training images available, the better results you will get from the AntMatcher.

AntMatcher currently uses ant head images to identify ants, but we understand the importance of adding other parts of ant body to increase identification accuracy. Eli is also developing and improving the classifier, which in future will allow to improve identification and to add new features. Most importantly, AntMatcher is an open source program, so that means that each of you can help to improve it. There are a million ways to do so. If you need ideas how you could help, please write us an email or check Help us page. You can also contribute to the improvement of AntMatcher by simply using it and sending us feedback. Together, let’s save insects and let’s make science move forward with technology. 

Silvija Budaviciute