Remote-sensing-based crop mapping implemented using UNet architecture
The UNet architecture takes 7 bands of Landsat as input and output a one-band image with the same size (width*height). Every pixel in the images are translated to a crop pixel with a value indicating its type.
The training images and the corresponding output array were reshaped into a specific matrix format, e.g., (batch_size, band_count, width, height, classes) and then mapped pixel by pixel.
- config.py file includes some properties like dataset directory, test directory and the colormap.
- make_dataset.py file is used for feature extraction and creating dataset.
- train_model.py contains the UNet Architecture
- utils.py file contains the helper functions