Scene Classification of Remote Sensing Images Guided by Fine-Grained Salient Region

Feiyang Li; Jiangtao Wang; Ziyang Wang

doi:10.3788/LOP212616

Laser & Optoelectronics Progress, Volume. 60, Issue 4, 0428001(2023)

Scene Classification of Remote Sensing Images Guided by Fine-Grained Salient Region

Feiyang Li, Jiangtao Wang^*, and Ziyang Wang

Author Affiliations

School of Physics and Electronic Information, Huaibei Normal University, Huaibei 235000, Anhui, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Flow chart of the proposed method

Download full size

Fig. 2. Structure of residual unit

Download full size

Fig. 3. Local critical area detecting

Download full size

Fig. 4. Algorithm flow of finding the largest connected region in binary image

Download full size

Fig. 5. Partial dataset display

Download full size

Fig. 6. Confusion matrix on RSSCN7 dataset when proportion of training samples is 50%

Download full size

Fig. 7. Confusion matrix on Aerial Image dataset when proportion of training samples is 50%

Download full size

Fig. 8. Confusion matrix on NWPU-RESISC45 dataset when proportion of training samples is 20%

Download full size

Table 1. OA of the proposed method under different regulation scales on RSSCN7 dataset
View table
Table 1. OA of the proposed method under different regulation scales on RSSCN7 dataset
Parameter 20% training 50% training
$λ = 0.3$ 92.66 94.63
$λ = 0.5$ 93.17 95.11
$λ = 0.7$ 92.99 94.93
$λ = 0.9$ 92.30 94.60

Table 2. Size and running time of different models
View table
Table 2. Size and running time of different models
Method Model size /MB Test time /s
Vgg16 112 0.0121
ResNet18 89 0.0096
ResNet50 195 0.0107
Proposed method 89 $\times$ 2 0.0117

Table 3. OA of different methods on RSSCN7 dataset unit:%

View table

Table 3. OA of different methods on RSSCN7 dataset unit:%

Method	20% training	50% training
ResNet18	91.76±0.25	94.36±0.20
Proposed method	93.11±0.20	95.23±0.20
Resnet50^［18］	90.23±0.43	93.12±0.55
Resnet50-TEX-Net-LF^［18］	92.45±0.45	94.00±0.57
EfficientNetB3^［17］	92.06±0.39	94.39±0.10
VGG-M-TEX-Net-EF-4ch^［18］	86.77±0.76	89.61±0.54
VGG-M-TEX-Net-EF-6ch^［18］	85.65±0.79	88.70±0.78
Deep filter banks^［19］		90.40±0.60
Gan-full pipeline^［20］	83.47±0.63	87.32±0.54
FV+HCV^［21］		86.40±0.70
CaffeNet^［22］	85.57±0.95	88.25±0.62
VGG-VD-16^［22］	83.98±0.87	87.18±0.94

Table 4. OA of different methods on Aerial Image dataset

View table

Table 4. OA of different methods on Aerial Image dataset

Method	20% training	50% training
ResNet18	92.31±0.20	95.35±0.22
Proposed method	93.90±0.25	96.06±0.20
Resnet101-FSL^［23］		95.88
Resnet50-TEX-Net-LF^［21］	93.81±0.12	95.73±0.16
EfficientNetB3^［17］	93.43±0.33	94.45±0.76
Two-Stream Fusion^［24］	92.32±0.41	94.58±0.41
Fusion by Addition^［25］		91.87±0.36
CaffeNet^［22］	86.86±0.47	89.53±0.31
TEX-TS-Net（addition）^［26］	88.56±0.25	90.29±0.19
RADC-Net^［27］	88.12±0.43	92.35±0.19
SalM3LBPCLM^［28］	86.92±0.35	89.76±0.45
VGG-VD-16^［22］	86.59±0.29	89.64±0.36

Table 5. OA of different methods on NWPU-RESISC45 dataset

View table

Table 5. OA of different methods on NWPU-RESISC45 dataset

Method	10% training	20% training
ResNet18	88.51±0.12	91.72±0.22
Proposed method	89.73±0.10	92.70±0.09
Multi-scale triplet loss^［29］	88.30±0.24	91.62±0.35
Two-Stream Fusion^［24］	80.22±0.22	83.16±0.18
RADC-Net^［27］	85.72±0.25	87.63±0.28
Fine-tuned VGGNet-16^［30］	87.15±0.45	90.36±0.18
Fine-tuned GoogLeNet^［30］	82.57±0.12	86.02±0.18
SAL-TS-Net（addition）^［26］	79.75±0.41	81.52±0.28
TEX-TS-Net（addition）^［26］	79.63±0.30	81.22±0.27
Gan-full pipeline^［20］	72.21±0.21	77.99±0.19

Tools

Get Citation

Copy Citation Text

Feiyang Li, Jiangtao Wang, Ziyang Wang. Scene Classification of Remote Sensing Images Guided by Fine-Grained Salient Region[J]. Laser & Optoelectronics Progress, 2023, 60(4): 0428001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites