Scene Classification of Remote Sensing Images Based on Wavelet-Spatial High-Order Feature Aggregation Network

Kang Ni; Mingliang Zhai; Peng Wang

doi:10.3788/AOS202242.2428005

Acta Optica Sinica, Volume. 42, Issue 24, 2428005(2022)

Scene Classification of Remote Sensing Images Based on Wavelet-Spatial High-Order Feature Aggregation Network

Kang Ni^1,2、*, Mingliang Zhai³, and Peng Wang⁴

Author Affiliations

¹School of Computer Science, Nanjing University of Posts and Telecommunications, Nanjing 210023, Jiangsu , China

²Jiangsu Key Laboratory of Big Data Security & Intelligent Processing, Nanjing 210023, Jiangsu , China

³College of Automation & College of Artificial Intelligence, Nanjing University of Posts and Telecommunications, Nanjing 210023, Jiangsu , China

⁴College of Electronic and Information Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, Jiangsu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Results of feature extraction by max-pooling, average-pooling, and wavelet transform. (a) Remote sensing images; (b) max pooling; (c) average pooling; (d) wavelet transform

Download full size

Fig. 2. Network architecture of proposed WHFA-Net

Download full size

Fig. 3. Images of several classes from AID dataset. (a) Airport; (b) bare land; (c) bridge; (d) dense residential area; (e) farmland;(f) forest; (g) railway station; (h) square

Download full size

Fig. 4. Images of several classes from NWPU45 dataset. (a) Airport; (b) beach; (c) church; (d) desert; (e) harbor; (f) lake; (g) mountain; (h) thermal power station

Download full size

Fig. 5. Confusion matrix of scene classification on NWPU45 dataset under 10% training set ratio

Download full size

Fig. 6. Confusion matrix of scene classification on NWPU45 dataset under 20% training set ratio

Download full size

Fig. 7. Network curves on NWPU45 dataset under 10% training set ratio. (a) Loss-epoch curves; (b) Top1 error-epoch curves; (c) Top5 error-epoch curves

Download full size

Fig. 8. Network curve on NWPU45 dataset under 20% training set ratio. (a) Loss-epoch curves; (b) Top1 error-epoch curves; (c) Top5 error-epoch curves

Download full size

Fig. 9. Comparison of results of three branches

Download full size

Table 1. Differences between proposed network and existing remote sensing image scene classification algorithms

View table

Table 1. Differences between proposed network and existing remote sensing image scene classification algorithms

Network	Spatial domain	Wavelet domian	High-order feature extraction	End-to-end training
MSCP^［10］	√	×	√	×
DCCNN^［19］	√	×	×	√
APDC-Net^［13］	√	×	×	√
Wavelet CNN^［20］	×	√	×	√
SCCov^［14］	√	×	√	√
WHFA-Net	√	√	√	√

Table 2. Comparison of scene classification accuracy of algorithms

View table

Table 2. Comparison of scene classification accuracy of algorithms

Network	Classification accuracy of AID		Classification accuracy of NWPU45
Network	20% scence images	50% scence images	10% scence images	20% scence images
VGG-16	86.59±0.29	89.64±0.36	76.47±0.18	79.79±0.15
DCCNN	87.37±0.41	91.49±0.22	83.97±0.19	85.63±0.18
MSCP	91.52±0.21	94.42±0.17	85.33±0.17	88.93±0.14
APDC-Net	88.56±0.29	92.15±0.29	85.94±0.22	87.84±0.26
Wavelet CNN	-	96.65±0.24	84.69±0.44	88.84±0.38
GBNet	90.16±0.24	93.72±0.34	-	-
LCNN-BFF	91.66±0.48	94.62±0.16	86.53±0.15	91.73±0.17
WHFA-Net	91.72±0.19	96.14±0.14	88.59±0.21	91.84±0.13

Table 3. Ablation experimental results
View table
Table 3. Ablation experimental results
Spatial domain Wavelet domain Higher-order feature aggregation OA /%
√ × × 83.41
× √ × 83.98
√ × √ 87.18
× √ √ 87.61
√ √ × 87.10
√ √ √ 88.59

Tools

Get Citation

Copy Citation Text

Kang Ni, Mingliang Zhai, Peng Wang. Scene Classification of Remote Sensing Images Based on Wavelet-Spatial High-Order Feature Aggregation Network[J]. Acta Optica Sinica, 2022, 42(24): 2428005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites