Hyperspectral Anomaly Detection Based on 3D Convolutional Autoencoder and Low Rank Representation

Bangyong SUN; Zhe ZHAO; Bingliang HU; Tao YU

doi:10.3788/gzxb20215004.0410003

Journals >Acta Photonica Sinica >Volume 50 >Issue 4 >Page 254 > Article

Acta Photonica Sinica
Vol. 50, Issue 4, 254 (2021)

Hyperspectral Anomaly Detection Based on 3D Convolutional Autoencoder and Low Rank Representation

Bangyong SUN^1、2, Zhe ZHAO¹, Bingliang HU², and Tao YU^2、*

Author Affiliations

¹College of Printing， Packaging and Digital Media， Xi'an University of Technology， Xi'an70048， China

²Key Laboratory of Spectral Imaging Technology CAS， Xi'an Institute of Optics and Precision Mechanics， Chinese Academy of Sciences， Xi'an710119， China

show less

DOI: 10.3788/gzxb20215004.0410003 Cite this Article

Bangyong SUN, Zhe ZHAO, Bingliang HU, Tao YU. Hyperspectral Anomaly Detection Based on 3D Convolutional Autoencoder and Low Rank Representation[J]. Acta Photonica Sinica, 2021, 50(4): 254 Copy Citation Text

show less

Fig. 1. Framework of the proposed method

Download full size | View in the Article

Fig. 2. The AVIRIS-1 dataset

Download full size | View in the Article

Fig. 3. The AVIRIS-2 dataset

Download full size | View in the Article

Fig. 4. The AUC value of two datasets under

Download full size | View in the Article

Fig. 5. The AUC value of two datasets under

Download full size | View in the Article

Fig. 6. The AUC value of two datasets under different eps and MinSample when p equal 8

Download full size | View in the Article

Fig. 7. The AUC value of two datasets under different eps and MinSample when p equal 9

Download full size | View in the Article

Fig. 8. The AUC value of two datasets under different eps and MinSample when p equal 10

Download full size | View in the Article

Fig. 9. The AUC value of two datasets under different eps and MinSample when p equal 11

Download full size | View in the Article

Fig. 10. The AUC value of two datasets under different eps and MinSample when p equal 12

Download full size | View in the Article

Fig. 11. The AUC value of two datasets under different p when eps equal 0.012 and MinSample equal 10

Download full size | View in the Article

Fig. 12. Detection results of various detection algorithms in AVIRIS-1 dataset

Download full size | View in the Article

Fig. 13. Detection results of various detection algorithms in AVIRIS-2 dataset

Download full size | View in the Article

Fig. 14. ROC curves of two datasets

Download full size | View in the Article

Block

Layer

Input size

Kernel size

Strides

Output size

Encoder

Conv1

BN+LReLu

189 $\times$ 5 $\times$ 5 $\times$ 1

1 $\times$ 3 $\times$ 3

1 $\times$ 1 $\times$ 1

189 $\times$ 3 $\times$ 3 $\times$ 12

Conv2

BN+LReLu

189 $\times$ 3 $\times$ 3 $\times$ 12

3 $\times$ 1 $\times$ 1

63 $\times$ 3 $\times$ 3 $\times$ 24

Conv3

BN+LReLu

63 $\times$ 3 $\times$ 3 $\times$ 24

1 $\times$ 3 $\times$ 3

1 $\times$ 1 $\times$ 1

63 $\times$ 1 $\times$ 1 $\times$ 36

Conv4

BN+Sigmoid

63 $\times$ 1 $\times$ 1 $\times$ 36

3 $\times$ 1 $\times$ 1

21 $\times$ 1 $\times$ 1 $\times$ 48

Decoder

Deconv1

BN+LReLu

21 $\times$ 1 $\times$ 1 $\times$ 48

3 $\times$ 1 $\times$ 1

63 $\times$ 1 $\times$ 1 $\times$ 36

Deconv2

BN+LReLu

63 $\times$ 1 $\times$ 1 $\times$ 36

1 $\times$ 3 $\times$ 3

1 $\times$ 1 $\times$ 1

63 $\times$ 3 $\times$ 3 $\times$ 24

Deconv3

BN+LReLu

63 $\times$ 3 $\times$ 3 $\times$ 24

3 $\times$ 1 $\times$ 1

189 $\times$ 3 $\times$ 3 $\times$ 12

Deconv4

189 $\times$ 3 $\times$ 3 $\times$ 12

1 $\times$ 3 $\times$ 3

1 $\times$ 1 $\times$ 1

189 $\times$ 5 $\times$ 5 $\times$ 1

Table 1. The Architecture of 3D-CAE

View in the Article

Dataset

RXW

WLRX₄

CRD

LRASR

DAEAD

RC-LRaSMD

Proposed

AVIRIS-1

AVIRIS-2

0.911 1

0.944 5

0.972 5

0.989 6

0.977 4

0.989 9

0.993 2

0.940 3

0.967 5

0.951 2

0.909 6

0.957 3

0.990 1

0.991 4

Table 2. The AUC value of different methods in two datasets

View in the Article

	Method	AVIRIS-1(AUC)	AVIRIS-2(AUC)
①	3D-Conv+RX	0.954 0	0.962 2
②	3D-Conv+LRR	0.991 6	0.988 9
③	3D-Conv+LRR(K-means)+RE	0.990 8	0.990 1
④	3D-Conv(MSE) +LRR+RE	0.977 2	0.977 4
⑤	3D-Conv +LRR+RE	0.993 2	0.991 4

Table 3. Analysis of the results of different steps of the Proposed algorithm

Bangyong SUN, Zhe ZHAO, Bingliang HU, Tao YU. Hyperspectral Anomaly Detection Based on 3D Convolutional Autoencoder and Low Rank Representation[J]. Acta Photonica Sinica, 2021, 50(4): 254

Download Citation

Tools

Save the article for my favorites

Paper Information