Recycle Sort

• 

  Screen shot of application running from command line

Inspiration

Since climate change is one of the rising problems in the 21st century, one of the leading ideas that is taught to children is to recycle however incorrect recycling has led much of the process to become ineffective and inefficient.

What it does

Our project strives to create a computer vision system that would help visually determine what product you were intending on recycling and the proper process of recycling it. Other uses from the vision system could be implemented in a physical sorting device .

How we built it

We build it by creating a image classifier for 5 main types (metals, cardboard, paper, plastic, glass) of recycling. We trained the model on a custom cnn for image classification. Using the output graph trained over 10,000 epochs. We then created a python script to read frames from a USB camera stream where a user can select a frame to test the model on and determine the correct process for recycling the material as well as creating a platform for where a hardware system could be implemented to sort through house hold trash to maximize recycling efficiency.

Challenges we ran into

Setting up an image classifier since our original setup was to use coreML however due to issues with xCode we swapped to a pc based system. Later on collecting over 2000 images for a training set on 5 different types of recyclable materials was a race against time as the ability to train on mobile hardware vs. a desktop or server could affect our accuracy in the final product.

Accomplishments that we're proud of

Accomplishments that our group are proud of include training a image classifier on multiple objects that contain many visual similarities for example being able to differentiate glass from plastic as well as the fact that all of this was done on tensorflow-cpu rather than gpu due to a lack of resources creating a larger time gap for training and setup. Another accomplishment we are proud of is our ability to classifiy frames from a live usb camera and classify frames of the 5 types of recyclables in real time.

What we learned

We learned about creating a image classification system using tensor flow. We also learned to apply numpy, tensorflow and opencv in order to process frames live.

What's next for Recycle Sort using Tensorflow and Image Classification

In the future, if given more time to work on this project, our group would most likely work on object detection rather than just image classification creating more possibilities for hardware and software applications as with a object detection algorithm we would be able to track position as well.

Built With

• opencv
• python
• tensorflow