CropClassification-TCN

Models

1. Simple TCN

2. TCN with Residual Blocks

Both models can be trained either using weighted loss or not. We also implemented random prediction as a simple baseline.

Train

Please go to CodeRepo/MODEL_NAME/train_config.py to change the training configuration. Then run CodeRepo/MODEL_NAME/train.py

The code supports wandb.ai to monitor the training procedure online.

Automatic Hyperparameter Tuning

We tuned hyperparameters on simple TCN. Please run CodeRepo/simple_tcn/hyperpara_opt.py.

Test

Please go to CodeRepo/MODEL_NAME/test_config.py to change the testing configuration. Then run CodeRepo/MODEL_NAME/predict.py. It will save the confusion matrix and metrics (i.e., f1, precision and recall per crop type) as .csv file.

Generate crop type map and error map

Please go to CodeRepo/MODEL_NAME/test_config.py to change the testing configuration. Then run CodeRepo/MODEL_NAME/predict_map.py.

Evaluations

Visualize crop type map and error map

Please go to CodeRepo/utils/visualize_tf_pred.py

Result evaluation and visualization

The analysis can be found in ResultEvaluation.ipynb