1. On your PC, open WinSCP and connect to your Pi. Create a Lobe folder in your Pi's home directory and create a model folder in that directory.

2. Drag the resulting Lobe TF folder contents onto the Pi. Make note of the file path: /home/pi/Lobe/model

3. On the Pi, open a terminal and first download the TensorFlow Lite runtime library. Then install the lobe-python library for Python3 by running the following bash commands:

pip3 install https://dl.google.com/coral/python/tflite_runtime-2.1.0.post1-cp37-cp37m-linux_armv7l.whl

pip3 install lobe

4. Download the Trash Classifier code (rpi_trash_classifier.py) from this repo onto the Pi (click the "Code" button as shown in Photo 1).

  • Prefer to copy/paste? Snag the code below!
  • Prefer to download onto your computer? Download the repo/code onto your computer then transfer the Python code to the Pi via WinSCP (or your preferred remote file transfer program).

5. Once you've connected the hardware to the Pi's GPIO pins, read through the example code and update any file paths as needed:

  • Line 29: filepath to the Lobe TF model
  • Lines 47 and 83: filepath to captured images via Pi Camera

6. If necessary, update the model labels in the code to exactly match the labels in your Lobe model (including capitalization, punctuation, etc.):

  • Line 57: "garbage"
  • Line 60: "recycle"
  • Line 63: "compost"
  • Line 66: "hazardous waste facility"
  • Line 69: "not trash!"
# ------------------------------------------------------------------------
# Trash Classifier ML Project
# Please review ReadMe for instructions on how to build and run the program
# 
# (c) 2020 by Jen Fox, Microsoft
# MIT License
# --------------------------------------------------------------------------


#import Pi GPIO library button class
from gpiozero import Button, LED, PWMLED
from picamera import PiCamera
from time import sleep

from lobe import ImageModel

#Create input, output, and camera objects
button = Button(2)

yellow_led = LED(17) #garbage
blue_led = LED(27) #recycle
green_led = LED(22) #compost
red_led = LED(23) #hazardous waste facility
white_led = PWMLED(24) #Status light and retake photo

camera = PiCamera()

# Load Lobe TF model
# --> Change model file path as needed
model = ImageModel.load('/home/pi/Lobe/model')

# Take Photo
def take_photo():
    # Quickly blink status light
    white_led.blink(0.1,0.1)
    sleep(2)
    print("Pressed")
    white_led.on()
    # Start the camera preview
    camera.start_preview(alpha=200)
    # wait 2s or more for light adjustment
    sleep(3) 
    # Optional image rotation for camera
    # --> Change or comment out as needed
    camera.rotation = 270
    #Input image file path here
    # --> Change image path as needed
    camera.capture('/home/pi/Pictures/image.jpg')
    #Stop camera
    camera.stop_preview()
    white_led.off()
    sleep(1)

# Identify prediction and turn on appropriate LED
def led_select(label):
    print(label)
    if label == "garbage":
        yellow_led.on()
        sleep(5)
    if label == "recycle":
        blue_led.on()
        sleep(5)
    if label == "compost":
        green_led.on()
        sleep(5)
    if label == "hazardous waste facility":
        red_led.on()
        sleep(5)
    if label == "not trash!":
        white_led.on()
        sleep(5)
    else:
        yellow_led.off()
        blue_led.off()
        green_led.off()
        red_led.off()
        white_led.off()

# Main Function
while True:
    if button.is_pressed:
        take_photo()
        # Run photo through Lobe TF model
        result = model.predict_from_file('/home/pi/Pictures/image.jpg')
        # --> Change image path
        led_select(result.prediction)
    else:
        # Pulse status light
        white_led.pulse(2,1)
    sleep(1)

7. Run the program using Python3 in the terminal window:

python3 rpi_trash_classifier.py

This guide was first published on Oct 22, 2020. It was last updated on Oct 22, 2020.

This page (Code it: Software) was last updated on Oct 15, 2021.

Text editor powered by tinymce.