A simple trash/waste classifier developed using MaixPy (a MicroPython framework) to run on K210 MCU on Sipeed's Maix dev board.
I use this project for the live demo and talk at Pycon Thailand 2019, which I discussed about AIoT: Intelligence on Low Power Microcontroller, with MicroPython
The MicroPython script (trash_classifier.py) will grab image from camera. Using prepared ML model (trash.kmodel), it can classify captured image as Plastic, Glass, or Other trash. Then display the captured image, inference result, and inference's frame/second on LCD.
The demo of how it works can be viewed on this Youtube video, although that video also discuss other things.
- One of Sipeed Maix development board. I use Maix Go. May work for other boards.
- MaixPy. Please refer to MaixPy docs for setting up development environment.
- Model. Explained below.
Copy trash.kmodel model file to Maix board, either via microSD card or burn it to board's memory using k-flash GUI tool.
If you take k-flash route, make sure to burn trash.kfpkg file instead. You can use k-flash command version, by typing this:
sudo kflash -p /dev/cu.usbserial-00005014B -b 2000000 -B goE trash.kfpkg
/dev/cu.usbserial-00005014B is which port the board is connected to, and should be changed according to your environment.
Then make sure to adjust the code for loading the model, using microSD card or from memory.
- Model is trained using dataset taken from trashnet repo, and transfer learning from MobileNet architecture