ajigeo/temporal-sar-transformer

Crop classification in the Cauvery Delta Zone using a multichannel based transformer model

Transformer based temporal SAR and multispectral data fusion for crop classification

Requirements

python=3.9.7
tensorflow=2.5.0
keras=2.4.3
sklearn=0.24.2
imblearn=0.8.0
numpy=1.21.2
pandas=1.3.3
matplotlib=3.4.3
seaborn=0.11.2
gdal=3.3.2

Installation

Run the code below to clone the repository to your local machine and change it to your working directory.

git clone https://github.com/ajigeo/convlstm-classification.git
cd convlstm-classification

All the libraries required for executing this code is shared as a yaml file. It can be installed by running the command,

conda env create -f environment.yml

Usage

Run the first 5 cells of training.py, to import the libraries, read the training data, preprocess it, building the DL model and fit the model.

Building the model

To call the model and train the data,

from models import fusion_transformer

model_history = my_fused_model.fit(
	x=[X_train_vv, X_train_vh, X_train_mss], y=y_train,
	validation_data=([X_val_vv, X_val_vh, X_val_mss], y_val),
	epochs=EPOCHS,batch_size=BATCH_SIZE,
    callbacks=[model_checkpoint])

Plotting the model performance

To plot the training accuracy and loss,

from utils import plot_performance

plot_performance(model_history, 'Title of the performance plot')

Model Evaluation Metrics

The Overall Accuracy, Kappa and class-wise F1 scores can be calculted by running the following code.

from utils import model_metrics

accuracy, kappa, f1_scores = model_metrics(predictions,y_test)

The confusion matrix for the classifier can be constructed by running,

from utils import confusion_matrix

confusion_matrix(class_labels, y_test, predictions, 'Title of the Confusion Matrix')

Reading the Image file

To read the image file, run the following code.

from raster import read_image

image_data = read_image('X:/location/of/the/image.tif')

Plotting the final thematic map

After reshaping and classifying the pixels in the image, the final thematic map can be plotted, assigned projections and extent by,

from raster import plot_output

plot_output(reshaped_classifed_matrix,transformations,'X:/location/of/classifed_map.tif')