This is a video series that builds, wires, tests and operates version 2 of an AI powered solar weeder. All files can be found at:
github.com/NathanBuildsDIY/we...
This video shows how to retrain an image classification model for use on a Raspberry PI. There are 3 steps:
1. Obtain photos - these must be as similar to what you'll classify as possible. We take photos with the robot and divide them into 224x224 size
2. Classify Photos - this is done by hand, but we automate it as much as possible so we can get a few hundred examples of each image class in a few hours
3. Retrain an existing model using Google Media Pipe Model Maker and then quantize it so we get a model_int8.tflite model that will run quickly on Raspberry Pi.
------
My goal is cleaner, cheaper food for everyone. Small AI powered robots like this one can remove weeds without use of chemicals and save farmers the time associated with weeding. My goal with this series is to show how to build and operate this robot so that anyone can pick it up and use it or improve on it, eventually reaching the goal of an improved robot that delivers cleaner, cheaper food for all.
This robot is controlled via a web interface. I'll show how to add to the image classification model to address weeds in your yard and I hope you'll share your training sets so that the model can continually improve and address more and more common weeds.