This repository contains an implementation of a UNet-based model for segmentation of Nutrient Deficiency Stress (see paper).
The dataset is available at the link.
The model is based on the custom UNet architecture (see model.py). The input to the model represents a combination of three RGB images.
The model was trained using Google Colab, all outputs can be found in the colab_notebook. The pretrained model is avalible in the "pretrained" directory.
The dataset containing 386 flights was randomly separated on 328 train and 58 validation flights. The model was trained for only 26 epochs (see train.py and config.py for detail).
The evalutaion results on validation data:
Inference examples can be found in the notebook inference_examples.
Example 1 (validation data):
Example 2 (validation data):
