Laser & Optoelectronics Progress, Volume. 58, Issue 14, 1401001(2021)
Scene Classification of Remote Sensing Images Based on RCF Network
Figures & Tables(12)
Fig. 7. Accuracy and loss value versus number of cycles. (a)(b) Training proportion is 20% in AID dataset; (c)(d) training proportion is 50% in AID dataset; (e)(f) training proportion is 10% in NWPU-RESISC45 dataset; (g)(h) training proportion is 20% in NWPU-RESISC45 dataset
Table 1. Comparison of test accuracy and number of parameters using four networks on test set in the AID and NWPU-RESISC45 datasets
View tableTable 1. Comparison of test accuracy and number of parameters using four networks on test set in the AID and NWPU-RESISC45 datasets
Group Model Params /106 FLOPs/GFLOPs Accuracy /% AID NWPU-RESISC45 Experiment I Experiment II Experiment I Experiment II 1 ResNet50 25.56 4.10 92.35±0.33 95.04±0.21 88.44±0.25 91.30±0.50 2 ResNet50-CBAM 26.08 4.11 93.57±0.41 95.68±0.25 89.53±0.23 92.32±0.42 3 ResNet50-FCAM 26.57 6.84 93.21±0.21 95.73±0.15 90.11±0.20 92.52±0.43 4 RCF 27.09 6.84 93.76±0.32 96.17±0.47 90.50±0.13 93.27±0.11
Table 2. Comparison of average training time and average test time per epoch using four networks on AID and NWPU-RESISC45 datasets
View tableTable 2. Comparison of average training time and average test time per epoch using four networks on AID and NWPU-RESISC45 datasets
Group Model AID NWPU-RESISC45 Experiment I Experiment II Experiment I Experiment II Train time /s Test time /s Train time /s Test time /s Train time /s Test time /s Train time /s Test time /s 1 ResNet50 30.65 17.33 43.23 11.34 78.07 57.26 87.61 49.23 2 ResNet50-CBAM 32.70 18.15 46.10 12.21 80.27 59.48 93.51 52.75 3 ResNet50-FCAM 48.34 28.97 65.12 18.96 126.81 97.70 144.35 87.78 4 RCF 51.62 31.15 68.76 19.86 134.61 104.35 153.27 92.95
Table 3. Classification accuracy on AID dataset unit: %
View tableTable 3. Classification accuracy on AID dataset unit: %
Algorithm Experiment I Experiment II salM3LBP-CLM 86.92±0.35 89.76±0.45 GoogleNet 91.37±0.35 93.99±0.37 VGG-VD16 92.06±0.29 94.74±0.51 ARCNet 88.75±0.40 93.10±0.55 scale-attention network 92.53±0.33 95.72±0.27 RCF 93.76±0.32 96.17±0.47
Table 4. Classification accuracy on NWPU-RESISC45 dataset unit: %
View tableTable 4. Classification accuracy on NWPU-RESISC45 dataset unit: %
Algorithm Experiment I Experiment II BoCF 82.65±0.31 84.32±0.17 GoogleNet 86.74±0.39 89.67±0.27 VGG-VD16 88.02±0.14 91.07±0.12 Integrated CNN 88.47±0.00 92.53±0.00 scale-attention network 88.92±0.29 92.25±0.18 RCF 90.50±0.13 93.27±0.11
Tools
Get Citation
Copy Citation Text
Shuxin Zhu, Zijun Zhou, Xingjian Gu, Shougang Ren, Huanliang Xu. Scene Classification of Remote Sensing Images Based on RCF Network[J]. Laser & Optoelectronics Progress, 2021, 58(14): 1401001
Paper Information
Category: Atmospheric Optics and Oceanic Optics
Received: Sep. 15, 2020
Accepted: Nov. 14, 2020
Published Online: Jun. 30, 2021
The Author Email: Shougang Ren (rensg@njau.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20