The Cornfield Debugger

Working Environment
Working camera with correct configurations
View of the Debugger as it sees an insect
Final product- The Cornfield Debugger

Inspiration

One of our team members took AP Environment Science in high school and was enlightened about the growing issue of pest control in commercial crops, and thus inspired us to design a solution to fight against them. We decided that given the area we were in and the fact that we were partnered with John Deere, a powerhouse agricultural engineering company, it would be an excellent idea to try and implement a solution against these unwanted pests. Additionally, we were inspired by John Deere's See & Spray product, which detects unwanted weeds in crops. Given our past knowledge of insect pests and our general dislike of most bugs, we were very motivated to create this project.

What it does

This device moves forward until it detects one of the banners we put up, each with a distinct image of an insect on it. It then uses a trained model to identify the type of insect it is - in our model we used grasshoppers, dragonflies, butterflies, mosquitoes, and ladybugs - and either "squashes" them or avoids them based on whether or not they are a pest or not. We assigned grasshoppers and mosquitoes as pests, and dragonflies, butterflies, and ladybugs as harmless. The device, when identifying a pest, will flash a yellow light and then run it over. When identifying a harmless pest, it will flash a green light and move around the insect.

How we built it

We utilized a model using TensorFlow that could identify 5 species of insects/bugs (model was created by Rishi Rajak, https://rishirajak.medium.com/). We trained this model and loaded it into our Raspberry Pi. Then, we used the array data given by the camera to load into the model so it could determine the type of bug.

Challenges we ran into

We had trouble installing TensorFlow onto the Raspberry Pi. There were also some issues with the hardware such as accidentally breaking the pi hat and wiring it incorrectly. In the software, there were lots of debugging issues.

Accomplishments that we're proud of

We are proud that we were able to get the model to work and have a decent amount of accuracy, although it definitely would have been better if we had trained it for longer (used more epochs).

What we learned

We learned that building a robot and coding it with machine learning takes a lot more time than we expected, and has a lot of debugging and troubleshooting involved.

What's next for The Cornfield Debugger

We hope to continue building this and increase its accuracy, and possibly expand it to a wider array of insects (ones that are more pertinent to real life crop pests).