An analysis on the usefulness of various typical classification methods when applied to predicting wether images of common household processed materials classify as biomass. Classifiers used:
- VGG16
- Resnet50
- Xception
- KNN
- Random Forest
Our dataset was small, but was expanded by augmentation. A quick look into a small subset of our dataset:
As predicted, when using simple traditional model fitting techniques, convolutional neural networks performed the best by a good margin.
A simple GUI based application was coded that allows the loading of the trained models to pedict the class on camera taken pictures. The application code can be found in the "app" folder.
Large files can be found at: https://drive.google.com/open?id=1JPRsoK4WOJitXLvhHDIsXDuCVernQHh0