Makersmiths Gets Down in the Weeds!

By: Shereef Sayed

Because Makersmiths is a non-profit, community-based organization, we do work with youth groups in a variety of ways. One of our partnerships is with Loudoun County Public Schools. This academic year, two students from Woodgrove High School taking an independent science research course contacted us for mentorship help.

For two Woodgrove High School students (Go Badgers!) and their Independent Research class, they needed to submit a research proposal that would address a problem. While visiting family in Italy, one of the students noticed that her Mom could eat foods containing gluten, unlike eating similar foods back home. “Why is that?”, she wondered. With help from her research partner, they came across a research paper that showed a strong correlation between Celiac disease and certain herbicides that are applied to wheat fields to hold weeds in check. They also learned that countries in Europe have banned the use of herbicides, but we are still using herbicides in the USA. “Hmm, how can we fix this?” they wondered. They determined that they should build a robot that can only spray herbicides on weeds, but not crops.

It turns out that their idea is one of a growing industry (pun intended) called Precision Agriculture. From academia to start-ups to corporations, the agriculture industry is developing innovative solutions to make better use of farm resources, including herbicides. 

My background is as a systems engineer and I volunteered to mentor the students.  When they described their initial research plan, it seemed like an awful lot of ground to cover in one academic year (pun intended). In order for a robot to spray only weeds, it seemed that we should first focus on solving the problem of distinguishing crops from non- crops. All you need is a neural network, right? They’re everywhere, right? 

Siri? Alexa? How do you build a neural network? 

To build a neural network you need to first train it with pictures of the objects you want to find. All we had to do was search the Internet for a bunch of pictures of weeds in a field, but we needed to find a lot of pictures...like, hundreds of pictures. Despite our best efforts to search the Internet, the only comprehensive image database of weeds that we found currently comes from Australia. Just to be sure that our Internet search was complete, we contacted a Professor at Virginia Tech in their Weed Sciences department, and asked what database his students use for their research. Turns out they use a database from Denmark! Well, I explained to the students, if the Internet is a sum of all human knowledge, and your search leads you back where you started, then you’ve just found a hole in the Internet; a place of incomplete knowledge. So, the research project was beginning to become more focused. The students decided that their research for this year's research class is to fill that gap and create the first North American Weed image database for training neural networks to find weeds in a field.

Right now, the team is focused on finding a camera that can take pictures in both the visible light and infrared spectrums. They are also learning how to annotate images to help train a neural network. 

With the help from a conservation specialist with Loudoun Soil and Water Conservation District, the students now have two farms that they can visit that are growing winter wheat.  Once the weather warms up, they will be busy taking lots of pictures, changing the world, one weed at a time.