HSI域自适应

目录

• 数据集描述
• 支持的模型
• 用法
  • 训练
  • 测试
• 结果
• 许可证

数据集描述

Houston数据集

类别	名称	Houston13	Houston18
1	Grass healthy	345	1353
2	Grass stressed	365	4888
3	Trees	365	2766
4	Water	285	22
5	Residential buildings	319	5347
6	Non-residential buildings	408	32459
7	Road	443	6365
total	total	2530	53200
shape	N * H * C	210 * 954 * 48	210 * 954 * 48

HyRANK数据集

类别	名称	Dioni	Loukia
1	Dense urban fabric	1262	288
2	Mineral extraction sites	204	67
3	Non irrigated land	614	542
4	Fruit trees	150	79
5	Olive Groves	1768	1401
6	Coniferous Forest	361	500
7	Dense Vegetation	5035	3793
8	Sparce Vegetation	6374	2803
9	Sparce Areas	1754	404
10	Rocks and Sand	492	487
11	Water	1612	1393
12	Coastal Water	398	451
total	total	20024	12208
shape	N * H * C	250 * 1376 * 176	249 * 945 * 176
RGB bands: 23, 11, 07

ShanghaiHangzhou数据集

类别	名称	Hangzhou	Shanghai
1	Water	18043	123123
2	Land/Building	77450	161689
3	Plant	40207	83188
total	total	135700	368000
shape	N * H * C	1600 * 260 * 198	590 * 230 * 198

Pavia数据集

类别	名称	PaviaU	PaviaC
1	Tree	3064	7598
2	Asphalt	6631	9248
3	Brick	3682	2685
4	Bitumen	1330	7287
5	Shadow	947	2863
6	Meadow	18649	3090
7	Bare soil	5029	6584
total	total	39332	39355
shape	N * H * C	610 * 340 * 102	1096 * 715 * 102

Indiana数据集

类别	名称	Source	Target
1	Concrete/ Asphalt	4867	2942
2	Corn cleanTill	9822	6029
3	Corn cleanTill EW	11414	7999
4	Orchard	5106	1562
5	Soybeans cleanTill	4731	4792
6	Soybeans cleanTill EW	2996	1638
7	Wheat	3223	10739
total	total	42159	35701
shape	N * H * C	300 * 400 * 200	300 * 400 * 200

支持的模型

• DDC
• DAN
• DeepCORAL
• JAN
• DSAN
• DANN
• ADDA
• CDAN
• MCD
• ParetoDA
• Self-training
• DST
• TSTNet

用法

训练

1. 运行 train/[model]/[dataset].bat文件
2. 或者运行如下命令

python train/ddc/train.py configs/houston/dan_1800_average.yaml ^
       --path ./runs/houston/dan-train ^
       --nodes 1 ^
       --gpus 1 ^
       --rank-node 0 ^
       --backend gloo ^
       --master-ip localhost ^
       --master-port 8886 ^
       --seed 30 ^
       --opt-level O2

测试

验证集等于测试集，无需再另行测试

结果

Dataset	Model	OA-best	backbone	sample-num	sample-order	loss	loss-ratio	kernel	batch-size
Houston	DNN	0.686±0.035	fe	-	-	softmax+ce	1	-	64
Houston	DDC	0.705±0.027	fe	-	-	softmax+ce, mmd loss	1:1	g1	64
Houston	DAN	0.694±0.048	fe	-	-	softmax+ce, mmd loss	1:1	g5	64
Houston	JAN	0.694±0.033	fe	-	-	softmax+ce, joint mmd loss	1:1	g5	64
Houston	DSAN	0.664±0.108	fe	-	-	softmax+ce, local mmd loss	1:1	g5	64
Houston	DANN	0.620±0.060	fe	-	-	softmax+ce	1	-	64
Houston	MCD	0.632±0.033	fe	-	-	softmax+ce, l1 loss	1:1	-	64
Houston	Self-training	0.652±0.003	fe	-	softmax+ce, cbst loss	1:1	-	100
Houston	DST	0.597±0.018	fe	-	-	softmax+ce, wcec loss, cbst loss	1:1:1	-	100
Houston	DADST	0.593±0.013	fe	-	-	softmax+ce, wcec loss, adv loss, cbst loss	1:1:1:1	-	100
Houston	TSTNet	0.762	fe	-	-	softmax+ce, mmd loss, got loss	1:1:0.1	g5	100
Houston	DNN	0.671±0.042	fe	1260	average	softmax+ce	1	-	64
Houston	DDC	0.676±0.053	fe	1260	average	softmax+ce, mmd loss	1:1	g1	64
Houston	DAN	0.686±0.058	fe	1260	average	softmax+ce, mmd loss	1:1	g5	64
Houston	JAN	0.677±0.058	fe	1260	average	softmax+ce, joint mmd loss	1:1	g5	64
Houston	DSAN	0.643±0.050	fe	1260	average	softmax+ce, local mmd loss	1:1	g5	64
Houston	DANN	0.590±0.060	fe	1260	average	softmax+ce	1	-	64
Houston	MCD	0.618±0.027	fe	1260	average	softmax+ce, l1 loss	1:1	-	64
Houston	Self-training	0.631±0.011	fe	1260	average	softmax+ce, cbst loss	1:1	-	64
Houston	DST	0.576±0.015	fe	1260	average	softmax+ce, wcec loss, cbst loss	1:1:1	-	64
Houston	DADST	0.575±0.015	fe	1260	average	softmax+ce, wcec loss, adv loss, cbst loss	1:1:1:1	-	64
HyRANK	DNN	0.507±0.023	fe	-	-	softmax+ce	1	l	64
HyRANK	DDC	0.523±0.030	fe	-	-	softmax+ce, mmd loss	1:1	g1	64
HyRANK	DAN	0.504±0.039	fe	-	-	softmax+ce, mmd loss	1:3	g5	64
HyRANK	JAN	0.516±0.026	fe	-	-	softmax+ce, joint mmd loss	1:1	g5	64
HyRANK	DANN	0.582±0.038	fe	-	-	softmax+ce	1	-	64
HyRANK	MCD	0.561±0.026	fe	-	-	softmax+ce, l1 loss	1:1	-	64
HyRANK	Self-training	0.514±0.009	fe	-	-	softmax+ce, cbst loss	1:1	-	64
HyRANK	DST	0.558±0.021	fe	-	-	softmax+ce, wcec loss, cbst loss	1:1:1	-	64
HyRANK	DADST	0.558±0.015	fe	-	-	softmax+ce, wcec loss, adv loss, cbst loss	1:1:1:1	-	64
HyRANK	DADST	0.572±0.023	fe	-	-	softmax+ce, wcec loss, adv loss, cbst loss	1:1:2:1	-	64
HyRANK	TSTNet	0.633	fe	-	-	softmax+ce, mmd loss, got loss	1:1:0.1	l	100
HyRANK	DNN	0.492±0.029	fe	1800	average	softmax+ce	1	l	64
HyRANK	DDC	0.491±0.028	fe	1800	average	softmax+ce, mmd loss	1:1	g1	64
HyRANK	DAN	0.496±0.021	fe	1800	average	softmax+ce, mmd loss	1:3	g5	64
HyRANK	JAN	0.485±0.022	fe	1800	average	softmax+ce, joint mmd loss	1:1	g5	64
HyRANK	DANN	0.473±0.036	fe	1800	average	softmax+ce	1	-	64
HyRANK	MCD	0.552±0.027	fe	1800	average	softmax+ce, l1 loss	1:1	-	64
HyRANK	Self-training	0.514±0.006	fe	1800	average	softmax+ce, cbst loss	1:1	-	64
HyRANK	DST	0.478±0.034	fe	1800	average	softmax+ce, wcec loss, cbst loss	1:1:1	-	64
HyRANK	DADST	0.478±0.028	fe	1800	average	softmax+ce, wcec loss, adv loss, cbst loss	1:1:1:1	-	64
ShanghaiHangzhou	DNN	0.909±0.002	fe	-	-	softmax+ce	1	-	64
ShanghaiHangzhou	DDC	0.887±0.008	fe	-	-	softmax+ce, mmd loss	1:1	g1	64
ShanghaiHangzhou	DAN	0.904±0.011	fe	-	-	softmax+ce, mmd loss	1:1	g5	64
ShanghaiHangzhou	JAN	0.903±0.011	fe	-	-	softmax+ce, joint mmd loss	1:1	g5	64
ShanghaiHangzhou	DSAN	0.907±0.005	fe	-	-	softmax+ce, local mmd loss	1:1	g5	64
ShanghaiHangzhou	DANN	0.905±0.016	fe	-	-	softmax+ce	1	-	64
ShanghaiHangzhou	MCD	0.717±0.105	fe	-	-	softmax+ce, l1 loss	1:1	-	64
ShanghaiHangzhou	Self-training	0.915±0.000	fe	-	-	softmax+ce, cbst loss	1:1	-	64
ShanghaiHangzhou	DST	0.933±0.012	fe	-	-	softmax+ce, wcec loss, cbst loss	1:1:1	-	64
ShanghaiHangzhou	DADST	0.927±0.007	fe	-	-	softmax+ce, wcec loss, adv loss, cbst loss	1:1:1:1	-	64
ShanghaiHangzhou	TSTNet	0.801	fe	-	-	softmax+ce, mmd loss, got loss	1:1:0.1	l	100
ShanghaiHangzhou	DNN	0.911±0.020	fe	540	average	softmax+ce	1	-	64
ShanghaiHangzhou	DDC	0.928±0.004	fe	540	average	softmax+ce, mmd loss	1:1	g1	64
ShanghaiHangzhou	DAN	0.913±0.011	fe	540	average	softmax+ce, mmd loss	1:1	g5	64
ShanghaiHangzhou	JAN	0.905±0.014	fe	540	average	softmax+ce, joint mmd loss	1:1	g5	64
ShanghaiHangzhou	DSAN	0.901±0.013	fe	540	average	softmax+ce, local mmd loss	1:1	g5	64
ShanghaiHangzhou	DANN	0.910±0.010	fe	540	average	softmax+ce	1	-	64
ShanghaiHangzhou	MCD	0.930±0.004	fe	540	average	softmax+ce, l1 loss	1:1	-	64
ShanghaiHangzhou	Self-training	0.925±0.000	fe	540	average	softmax+ce, cbst loss	1:1	-	64
ShanghaiHangzhou	DST	0.927±0.015	fe	540	average	softmax+ce, wcec loss, cbst loss	1:1:1	-	64
ShanghaiHangzhou	DADST	0.933±0.004	fe	540	average	softmax+ce, wcec loss, adv loss, cbst loss	1:1:1:1	-	64

结果

This project is released under the MIT(LICENSE) license.