Overall Process Video

https://youtu.be/mxnrM5A6Auc

Images taken in the test

Paper1.jpeg

Juice.jpeg

Plastic.jpeg

Metal.jpeg

Backend and Embedded communication

Backend.jpeg

Embedded.jpeg

Image of the sorted waste

Untitled

Video of the internal structure in work

https://youtu.be/DqwAZq2mESA

Video of credits redeeming

Since scanning the QR Code from the display proved to be a problem in the last minute we are scanning the same QR Code from the PC screen.

QR Code Read DEMO

Neural Network datasets

To train the neural network, 1866 images divided in 6 classes were taken. We also used image augmentation to increase the total number os images to 5638. For training 4416 images were used, 1102 for test and 120 for validation in a total of 50 epochs. Below are the accuray and loss curves.

ACC.jpeg

LOSS.jpeg