Data Augmentation for Remote Sensing Image Based on Generative Adversarial Networks Under Condition of Few Samples

Yuchen Jiang; Bin Zhu

doi:10.3788/LOP202158.0810022

Laser & Optoelectronics Progress, Volume. 58, Issue 8, 0810022(2021)

Data Augmentation for Remote Sensing Image Based on Generative Adversarial Networks Under Condition of Few Samples

Yuchen Jiang^* and Bin Zhu

State Key Laboratory of Pulsed Power Laser Technology, College of Electronic Countermeasures, National University of Defense Technology, Hefei, Anhui 230009, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(8)

Fig. 1. Basic structural frame of GAN

Download full size

Fig. 2. Data augmentation process of remote sensing image

Download full size

Fig. 3. Structural diagram of discrimination model

Download full size

Fig. 4. Examples of generated samples. (a)(b) Chimney; (c)(d) basketball court; (e)(f) stadium

Download full size

Table 1. Comparison of FID values of generated target on DIOR datasets

View table

Table 1. Comparison of FID values of generated target on DIOR datasets

Category	Airport	Basketball court	Bridge	Chimney	Overpass	Stadium
StyleGAN2	73.80	70.38	102.69	125.37	83.98	56.17
Ours	67.35	52.70	95.96	78.51	55.38	53.25
Category	Dam	Expressway-service-area	Golf field	Ground track field	Train station
StyleGAN2	123.04	62.42	67.03	45.20	113.78
Ours	103.00	54.22	57.98	39.87	93.00

Table 2. Comparison of FID values of generated target on RSOD datasets
View table
Table 2. Comparison of FID values of generated target on RSOD datasets
Category Overpass Playground
StyleGAN2 88.52 127.86
Ours 110.42 168.30

Table 3. Comparison of detection accuracy for adding different percent of enhancement data on DIOR dataset%

View table

Table 3. Comparison of detection accuracy for adding different percent of enhancement data on DIOR dataset%

Method	Percent of added data /Accuracy (mAP)
Affine transformation	0/46.06	100/47.69	200/48.45	300/47.74	400/49.14
Ours	0/46.06	5/48.14	10/46.97	15/45.60	20/46.17
Combined method	0/46.06	105/48.09	205/49.48	305/47.96	405/49.51

Table 4. Comparison of detection accuracy for adding different percent of enhancement data on RSOD dataset%

View table

Table 4. Comparison of detection accuracy for adding different percent of enhancement data on RSOD dataset%

Method
Affine transformation	0/70.78	100/74.22	200/79.28	300/76.76	400/77.84
Ours	0/70.78	5/75.37	10/74.83	15/76.52	20/77.25
Combined method	0/70.78	120/79.34	220/80.13	320/77.77	420/78.35

Tools

Get Citation

Copy Citation Text

Yuchen Jiang, Bin Zhu. Data Augmentation for Remote Sensing Image Based on Generative Adversarial Networks Under Condition of Few Samples[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0810022

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Jul. 30, 2020

Accepted: Sep. 22, 2020

Published Online: Apr. 12, 2021

The Author Email: Yuchen Jiang (jyc2647118@126.com)

DOI:10.3788/LOP202158.0810022

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of FID values of generated target on DIOR datasets

Table 1. Comparison of FID values of generated target on DIOR datasets

Table 2. Comparison of FID values of generated target on RSOD datasets

Table 2. Comparison of FID values of generated target on RSOD datasets

Table 3. Comparison of detection accuracy for adding different percent of enhancement data on DIOR dataset%

Table 3. Comparison of detection accuracy for adding different percent of enhancement data on DIOR dataset%

Table 4. Comparison of detection accuracy for adding different percent of enhancement data on RSOD dataset%

Table 4. Comparison of detection accuracy for adding different percent of enhancement data on RSOD dataset%