Example of the aitlas toolbox for crop type prediction

This notebook shows a sample implementation for crop type prediction using the aitlas toolbox and the Netherlands 2019 dataset.

import os

from aitlas.datasets import EOPatchCrops
from aitlas.models import LSTM
from aitlas.visualizations import display_eopatch_predictions

Load the dataset

dataset_config = {
    "root": "/home/dkocev/data/CropTypeNetherlands/2019/",
    "csv_file_path": "index.csv",
    "batch_size": 128,
    "shuffle": True,
    "num_workers": 4,
    "regions":["train", "test", "val",],
}
dataset = EOPatchCrops(dataset_config)

Visualize sample time series from the dataset

fig1 = dataset.show_timeseries(0)
fig2 = dataset.show_timeseries(100)
fig3 = dataset.show_timeseries(1000)

Inspect the data

dataset.show_samples()

	id	path	eopatch	polygon_id	CODE_CULTU	sequencelength	classid	classname	region
0	0	eopatch_11173/1666071	eopatch_11173	1666071	J0000	0	0	Permanent grassland	train
1	1	eopatch_10137/1894170	eopatch_10137	1894170	J0000	0	0	Permanent grassland	train
2	2	eopatch_6711/1875243	eopatch_6711	1875243	J0000	0	0	Permanent grassland	train
3	3	eopatch_6590/1924836	eopatch_6590	1924836	G1000	0	1	Temporary grasses and grazings	train
4	4	eopatch_3731/2150854	eopatch_3731	2150854	NUAA+WA+FA9	0	6	Other farmland	train
5	5	eopatch_939/2018637	eopatch_939	2018637	NUAA+WA+FA9	0	6	Other farmland	train
6	6	eopatch_3999/1573533	eopatch_3999	1573533	G3000	0	2	Green maize	train
7	7	eopatch_9569/1709233	eopatch_9569	1709233	J0000	0	0	Permanent grassland	train
8	8	eopatch_2152/2286699	eopatch_2152	2286699	NUAA+WA+FA9	0	6	Other farmland	train
9	9	eopatch_10867/1620648	eopatch_10867	1620648	J0000	0	0	Permanent grassland	train
10	10	eopatch_4205/1674540	eopatch_4205	1674540	G3000	0	2	Green maize	train
11	11	eopatch_3465/2142385	eopatch_3465	2142385	R1000	0	3	Potatoes (including seed potatoes)	train
12	12	eopatch_2504/2167864	eopatch_2504	2167864	J0000	0	0	Permanent grassland	train
13	13	eopatch_4567/1530529	eopatch_4567	1530529	G3000	0	2	Green maize	train
14	14	eopatch_8115/1825570	eopatch_8115	1825570	G3000	0	2	Green maize	train
15	15	eopatch_8220/2108154	eopatch_8220	2108154	R2000	0	5	Sugar beet (excluding seed)	train
16	16	eopatch_7998/2183462	eopatch_7998	2183462	G3000	0	2	Green maize	train
17	17	eopatch_10489/2192293	eopatch_10489	2192293	J0000	0	0	Permanent grassland	train
18	18	eopatch_7950/1974596	eopatch_7950	1974596	C1111	0	4	Common winter wheat and spelt	train
19	19	eopatch_8836/1677178	eopatch_8836	1677178	J0000	0	0	Permanent grassland	train

dataset.data_distribution_table()

	Label	Region	Number of parcels
0	Common winter wheat and spelt	test	3152
1	Common winter wheat and spelt	train	15467
2	Common winter wheat and spelt	val	2778
3	Flowers and ornamental plants (excluding nurse...	test	1055
4	Flowers and ornamental plants (excluding nurse...	train	5143
5	Flowers and ornamental plants (excluding nurse...	val	911
6	Green maize	test	6841
7	Green maize	train	32271
8	Green maize	val	5614
9	Onions	test	1136
10	Onions	train	5368
11	Onions	val	948
12	Other farmland	test	1566
13	Other farmland	train	7843
14	Other farmland	val	1333
15	Permanent grassland	test	24051
16	Permanent grassland	train	116485
17	Permanent grassland	val	20502
18	Potatoes (including seed potatoes)	test	4758
19	Potatoes (including seed potatoes)	train	22920
20	Potatoes (including seed potatoes)	val	4044
21	Spring barley	test	906
22	Spring barley	train	4422
23	Spring barley	val	773
24	Sugar beet (excluding seed)	test	2391
25	Sugar beet (excluding seed)	train	11490
26	Sugar beet (excluding seed)	val	2032
27	Temporary grasses and grazings	test	7894
28	Temporary grasses and grazings	train	37484
29	Temporary grasses and grazings	val	6752

fig = dataset.data_distribution_barchart()

Load train and validation splits

train_dataset_config = {
    "root": "/home/dkocev/data/CropTypeNetherlands/2019/",
    "csv_file_path": "index.csv",
    "batch_size": 128,
    "shuffle": True,
    "num_workers": 4,
    "regions":["train", ],
}

train_dataset = EOPatchCrops(train_dataset_config)

validation_dataset_config = {
    "batch_size": 32,
    "shuffle": False,
    "num_workers": 4,
    "root": "/home/dkocev/data/CropTypeNetherlands/2019/",
    "csv_file_path": "index.csv",
    "regions":["val", ],
}

validation_dataset = EOPatchCrops(validation_dataset_config)
len(train_dataset), len(validation_dataset)

(258893, 45687)

Setup and create the model for training

epochs = 100
model_directory = "./experiments/LSTM"
model_config = {
    "input_dim":11,
    "num_classes": 10,
    "learning_rate": 0.001,
    "dropout" : 0.2,
    "metrics":["accuracy","f1_score", "kappa"]
}
model = LSTM(model_config)
model.prepare()

Training and validation

model.train_and_evaluate_model(
    train_dataset=train_dataset,
    epochs=epochs,
    model_directory=model_directory,
    val_dataset=validation_dataset,
    run_id='notebook-1',
)

Test the model

labels = ["Permanent grassland", "Temporary grasses and grazings", "Green maize", "Potatoes (including seed potatoes)",
          "Common winter wheat and spelt", "Sugar beet (excluding seed)", "Other farmland", "Onions",
          "Flowers and ornamental plants (excluding nurseries)", "Spring barley",]

test_dataset_config = {
    "batch_size": 32,
    "shuffle": False,
    "num_workers": 4,
    "root": "/home/dkocev/data/CropTypeNetherlands/2019/",
    "csv_file_path": "index.csv",
    "regions":["test", ],
}

test_dataset = EOPatchCrops(test_dataset_config)

model_path = "/home/dkocev/aitlas/experiments/LSTM/checkpoint.pth.tar"
model.metrics = ["accuracy", "f1_score", "kappa"]
model.running_metrics.reset()
model.evaluate(dataset=test_dataset, model_path=model_path)
model.running_metrics.get_scores(model.metrics)

Running and visualizing predictions

y_true, y_pred, y_prob = model.predict(dataset=test_dataset,)

eopatches_path = "/home/dkocev/data/CropTypeNetherlands/2019/eopatches/"
patch = "eopatch_7495"

fig = display_eopatch_predictions(
    eopatches_path,
    patch,
    y_pred,
    test_dataset.index,
    y_true,
    test_dataset.mapping,
)