The MAJURCA company is developing an automated recycling machine. When plastic waste is given to said machine, it will attempt to recognize the type of plastic by first taking two picture of it, and then feeding said pictures to a deep learning image classification model, pretrained for the task of plastic recognition.
The task of the team I was part of consisted in exploring different image classification models to accomplish the task of plastic type recognition. Several CNNs (Convolution Neural Networks) where tested on the 64,000 image dataset that was provided. Some examples are VGG16, InceptionV3, ResNet50, EfficientNetB0, Xception. As well as YOLOv3 for object detection.
This repo shows some of my work on this task. Each of the jupyter notebooks presented here have the following description:
- majurca.ipynb: This is the notebook in which the CNN models where trained.
- load-data.ipynb: Used to download the image dataset from the Azure Database.
- extract.ipynb: Used to index images for Content-Based image retreival.
- query.ipynb: Used to cleanup the dataset with Content-Based image retrieval.
- visualize.ipynb: Used to visualize the learning curves for some of the tested models
Reports detailing the work that was performed for the given task in both English and French can be found in:
- English: https://github.com/botiose/majurca/blob/master/doc/english.pdf
- French: https://github.com/botiose/majurca/blob/master/doc/french.pdf
As instructed the first page for the French report is in English.