Semi-supervised Scene Classification of Remote Sensing Images Based on GAN

Ying XIA; Junyao LI; Dongen GUO

doi:10.3788/gzxb20225103.0310003

Journals >Acta Photonica Sinica >Volume 51 >Issue 3 >Page 0310003 > Article

Acta Photonica Sinica
Vol. 51, Issue 3, 0310003 (2022)

Semi-supervised Scene Classification of Remote Sensing Images Based on GAN

Ying XIA^1、*, Junyao LI¹, and Dongen GUO^1、2

Author Affiliations

¹Chongqing University of Posts and Telecommunications，Chongqing Engineering Research Center of Spatial Big Data Intelligent Technology，Chongqing 400065，China

²School of Computer and Software，Nanyang Institute of Technology，Nanyang ，Henan 473000，China

show less

DOI: 10.3788/gzxb20225103.0310003 Cite this Article

Ying XIA, Junyao LI, Dongen GUO. Semi-supervised Scene Classification of Remote Sensing Images Based on GAN[J]. Acta Photonica Sinica, 2022, 51(3): 0310003 Copy Citation Text

show less

Fig. 1. SFGAN network structure

Download full size | View in the Article

Fig. 2. RAGAN network structure

Download full size | View in the Article

Fig. 3. Basic structure of residual neural network

Download full size | View in the Article

Fig. 4. Feature fusion and GAM structure

Download full size | View in the Article

Fig. 5. Confusion matrix of different number of markers in EuroSAT dataset

Download full size | View in the Article

Fig. 6. Confusion matrix of different number of markers in UCM dataset

Download full size | View in the Article

Fig. 7. Accuracy curves of different numbers of markers in EuroSAT and UCM datasets

Download full size | View in the Article

Dataset	Images per category	Number of categories	Total images	Size
EuroSAT	2 000~3 000	10	27 000	64×64
UC Merced	100	21	2100	256×256

Table 1. Dataset information

View in the Article

Method	Numbers of label M on EuroSAT （10 class）				Time/h	Numbers of label M on Ucm （21 class）				Time/h
	100	1 000	2 000	21 600		100	200	400	1 680
CNN^［6］	29.3%	46.1%	59.0%	83.2%	25	18.5%	32.8%	43.6%	62.1%	1
Inception V3^［6］	63.9%	84.6%	87.9%	91.5%	27	55.4%	71.1%	81.1%	85.4%	1.7
FMGAN^［2］	63.0%	75.8%	78.3%	86.9%	30	43.6%	69.2%	74.5%	80.2%	1.5
REG⁃GAN^［9］	64.7%	72.8%	76.4%	82.3%	28	40.4%	55.4%	63.6%	72.3%	1.3
SFGAN^［6］	68.6%	86.1%	89.0%	93.2%	31.5	43.9%	52.1%	60.6%	79.5%	2
SAGGAN^［10］	76.8%	88.1%	90.7%	94.3%	33	54.1%	69.7%	83.3%	90.5%	2
RAGAN（ours）	71.5%	88.2%	93.3%	97.4%	37.5	55.2%	71.4%	85.7%	91.0%	2.25

Table 2. Classification results on EuroSAT and UCM datasets

View in the Article

Method	Numbers of label M on EuroSAT （10 class）				Time/h	Numbers of label M on UCM （21class）				Time/h
	100	1 000	2 000	21 600		100	200	400	1 680
SFGAN	68.6%	86.1%	89.0%	93.2%	31.5	43.9%	52.1%	60.6%	79.5%	2
+SNRB	73.0%	88.1%	91.6%	95.9%	34.8	45.7%	52.9%	68.3%	82.4%	2.1
+GAM	70.7%	87.3%	92.6%	97.1%	33.5	54.8%	58.2%	77.8%	90.7%	2.1
+Fusion	65.4%	86.3%	90.4%	93.4%	32	41.9%	50.7%	65.6%	80.9%	2
+All	71.5%	88.2%	93.3%	97.4%	37.5	55.2%	71.4%	85.7%	91.0%	2.25

Table 3. Influence of each module on classification accuracy

Ying XIA, Junyao LI, Dongen GUO. Semi-supervised Scene Classification of Remote Sensing Images Based on GAN[J]. Acta Photonica Sinica, 2022, 51(3): 0310003

Download Citation

Tools

Save the article for my favorites

Paper Information