Solve America's Recycling Problem using Robotic ConvNets

Detection and classification of recyclable items to help recyclable facility robots identify and pick them up correctly

Hardware Design Model

Description of Files

• pics folder

  • includes the images of the recyclable and non-recycable objects used in training and validation of ConvNets.
  • The images were taken using a Mac camera and the OpenCV-Python library

• prediction_pics folder

  • Set of original images and the image recorded by the camera and sent to the ConvNet model for prediction

• cnn_resnet_load_camera.py

  • Python main program
  • Loaded in Jetson Nano
  • Using OpenCV-Python, opens connection to camera. Takes and reads the picture
  • Load the model on the Jetson Nano kit.
  • Invoke the trained model to read the image and predict the class
  • Based on the prediction, send a signal to the Nano Microcontroller to bin the item correctly

• cnn_resnet_h5model.ipynb

  • Program created using Jupyter notebook in Google Colab to create the trained ConvNet model in h5 format

• cnn_resnet.ipynb, cnn_inception.ipynb, cnn_vgg.ipynb

  • Programs created using Jupyter notebook in Google Colab to create the trained ConvNet models

• picmac.py

  • Program to take images of recyclable and non-recyclable objects using Mac's integrated camera and OpenCV-Python library

• testarduino/djhost/djhost.ino

  • Program loaded in nano microcontroller
  • Moves Smart Servo Motors
  • If value received is 'Y' (i.e. recyclable), it moves them to one place.
  • If value received is 'N' (i.e. non-recyclable), it moves them to the other place.