Unsupervised Domain Adaptation for SAR Target Recognition from Simulated to Measured Data in Time, Frequency and Scattering Domains

PyTorch implementation of the paper "Unsupervised Domain Adaptation for SAR Target Recognition from Simulated to Measured Data in Time, Frequency and Scattering Domains".

Install

Please see INSTALL.md

SAMPLE dataset Scene1

	Simulated	Measured
	training	training	testing	Total
	14° to 17°	14° to 16°	17°
2s1	174	116	58	348
bmp2	107	55	52	214
btr70	92	43	49	184
m1	129	78	51	258
m2	128	75	53	256
m35	129	76	53	258
m548	128	75	53	256
m60	176	116	60	352
t72	108	56	52	216
zsu23	174	116	58	348

SAMPLE dataset Scene2

	Simulated	Measured
	training	training	testing	Total
	14° to 17°	16°	17°
2s1	174	50	58	282
bmp2	107	55	52	214
btr70	92	43	49	184
m1	129	52	51	232
m2	128	52	53	233
m35	129	52	53	234
m548	128	52	53	232
m60	176	51	60	288
t72	108	56	52	216
zsu23	174	50	58	282

SAMPLE dataset Scene3

	Simulated	Measured
	training	training	testing	Total
	14° to 16°	16°	17°
2s1	116	50	58	224
bmp2	55	55	52	162
btr70	43	43	49	135
m1	78	52	51	181
m2	75	52	53	180
m35	76	52	53	181
m548	75	52	53	179
m60	116	51	60	228
t72	56	56	52	164
zsu23	116	50	58	224

Classification models on the three Scenes of SAMPLE dataset

Scene	Accuracy	Precision	Recall	F1
Scene1	$ 99.33\pm 00.19$%	$ 99.40\pm 00.15$%	$ 99.33\pm 00.19$%	$ 99.35\pm 00.18$%
Scene2	$ 98.89\pm 00.15$%	$ 98.98\pm 00.07$%	$ 98.89\pm 00.16$%	$ 98.88\pm 00.14$%
Scene3	$ 98.33\pm 00.32$%	$ 98.44\pm 00.24$%	$ 98.26\pm 00.41$%	$ 98.26\pm 00.37$%

Training

# Experiment Scene1
python tools/train.py --config configs/Scene1/TFSNet.yaml --data_dir datasets/Scene1 --src_domain Simulation --tgt_domain Real --checkpoints configs/Scene1

# Experiment Scene2
python tools/train.py --config configs/Scene2/TFSNet.yaml --data_dir datasets/Scene2 --src_domain Simulation --tgt_domain Real --checkpoints configs/Scene2

# Experiment Scene3
python tools/train.py --config configs/Scene3/TFSNet.yaml --data_dir datasets/Scene3 --src_domain Simulation --tgt_domain Real --checkpoints configs/Scene3

Testing and Visualizing

# Experiment Scene1
python tools/visualize_results.py --config configs/Scene1/TFSNet.yaml --data_dir datasets/Scene1 --tgt_domain Real --checkpoints configs/Scene1

# Experiment Scene2
python tools/visualize_results.py --config configs/Scene2/TFSNet.yaml --data_dir datasets/Scene2 --tgt_domain Real --checkpoints configs/Scene2

# Experiment Scene3
python tools/visualize_results.py --config configs/Scene3/TFSNet.yaml --data_dir datasets/Scene3 --tgt_domain Real --checkpoints configs/Scene3

Warning

The paper is currently in the review stage, so the important codes has not been uploaded yet.