Detecting Leaves

To be able to detect coffee rust on leaves, our robot must first be able to detect leaves in general. As such, a model was trained on a dataset of labelled leaves. The model used was the YOLOv5 model, and the result of the model after running 200 epochs on roughly 1000 labelled images is shown below. The precision and recall metrics plateau around 0.7, which can be improved through more data and more training epochs. One promising aspect of this trained model is that it is able to detect leaves with spots on them, which is similar to what coffee rust on leaves look like.