What is our project?

• Our project is a drone/model airplane that looks for areas which trees are doing well in/areas that are lacking trees and drops seeds. This is faster than planting by hand.
• We chose an aerial vehicle over a wheeled vehicle because it is easier to fly over difficult terrain than to drive through it.
• Our project also finds data about tree canopies and documents it as well as references it or future use. This would be very tricky for a person to do, because it is hard for a human to get to the top of a tree.

Who is it for?

• Anyone interested in planting plants in a large scale, or looking for a tool to help with wide range aerial view data.
• could also be for schools that want to do research, or ecological research sites.

Precedents

• One precedent is Airseed, which plant trees using a large drone and a pre-determined flight path. This would be different from our idea because ours drone/model airplane would use AI to find good spots to plant trees instead of someone manually mapping a path.
• Airseed puts their seeds in pods which are made out of a bunch of nutrients to help the trees grow, which is something we are seriously considering.

Prototypes

• The first seed dropper prototype was a cardboard cylinder, with a slider that pushed seeds out. The slider always jammed and it broke itself, and it was too tall for the plane.
• the next version was plywood, which made it stiffer and stopped itself from breaking.

Plane

Plane

• Model planes, or fixed wing drones, are much more efficient than a standard quadcopter.`
• We want to make the plane a lot lighter to be able to have a higher payload.

Code

• Here is our code for the identification of trees. This code takes in images from a camera and compares them using the model. If it identifies the right object it will send a signal to an Arduino where it runs code to control the dropping mechanism of the seeds.

yoloV8

• yolo v8 is an ai model. It allows you to collect your own data set of images and train yolo v8 with it. You give it images of what you're identifying and what you are not. Once you give your trained model an image it will compare its data sets and identify if there is your object in the image.
• yolov8 is specifically for images and video while there are other ai models that have been developed for specific text tasks. for example GitHub co pilot has been made to help programmers program.

Final Prototype

• This version needs a few more improvements, but it works for the most part. it will use rubber bands for pushing the seeds forward, and the original slider mechanism for pushing the seeds out.

Privacy issues?

• Assuming people are following law part 107 we can assume that Greens 4 Growth would not be disturbing anyones privacy.
• Accidentally recording someone's face

What new problems could we be creating?

• One issue the drone could cause is scaring away wildlife, because it is loud and has fast spinning propellers.
• There could be miss-usage or human error when setting up the drone or preparing for a planting session.

Issues we need to solve

• The plane needs a 4 cell battery instead of a 3 cell. Meaning more power.
• Integrating the seed dropper into the plane will be tricky, because it takes up a lot of space.
• Improve the identification different trees from the canopy view.
• The AI only recognizes tree or not tree, we would want to make more models to adapt the quantity of use cases.
• We need a better camera to go on the plane, due to poor image quality effecting the results.

What are trail cameras?

• Small motion-capture cameras that snap a photo or start recording a video if motion is detected
• Created originally for wildlife research - tech from camera is used on home/property cameras
• When photo or video is captured, the camera places the capture into an sd card that can be transferred to your phone/computer
• Most cameras are wrapped with a camouflage type of design to blend into the wilderness so they don't disrupt wildife activity

Why

We chose to tackle this project because after connecting with Harvard Forest, we found out that they did not have any sort of animal tracking data nor did they have any trail cameras around the campus. We proposed the idea of having an AI system connected with trail cameras that will automatically track animal data for them so they do not have to start something new from the ground up. All they have to do it set it up and let it track the animals and then after a while, they can check up on the data that the AI has created. It would also eliminate the constant labor of having a human needing to identify what each animal photo is. That would be one of the biggest issues that a forest organization would have in starting to track their animal data manually.

How it works - story board

Progress

Progress

Prompts + Next steps

• Specify AI to the area - Harvard forest
• Set up trail cameras at Harvard forest campus
• Build archive of data - so researchers can access
• Expanding data base - making AI more accurate
• Adding more animals to it
• Connecting software to a trail camera - testing with AI integration
• Building UI for data mapping

Thank You