Heterogeneous Remote Sensing Image Matching Algorithm Based on Residual Pseudo-Siamese Convolution Cross-Correlation Network

Rongping Zou; Bin Zhu; Chenyang Wang; Yaoxuan Zhu; Yangdi Hu

doi:10.3788/LOP202259.1228002

Laser & Optoelectronics Progress, Volume. 59, Issue 12, 1228002(2022)

Heterogeneous Remote Sensing Image Matching Algorithm Based on Residual Pseudo-Siamese Convolution Cross-Correlation Network

Rongping Zou^1,2, Bin Zhu^1,2、*, Chenyang Wang¹, Yaoxuan Zhu^1,2, and Yangdi Hu³

Author Affiliations

¹College of Electronic Engineering, National University of Defense Technology, Hefei 230037, Anhui , China

²Key Laboratory of Infrared and Low Temperature Plasma of Anhui Province, Hefei 230037, Anhui , China

³Army 32256 of PLA, Guiling541000, Guangxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(6)

Fig. 1. Structure of residual pseudo-siamese convolution cross-correlation network

Download full size

Fig. 2. Heat map of inference results for different networks in different terrains

Download full size

Fig. 3. Heat map of middle layer features in RPSCCNet and heat map of result before Softmax operation for two networks

Download full size

Table 1. Quality of corresponding normalized heat map and L2 distance error of inference result in Fig. 2

View table

Table 1. Quality of corresponding normalized heat map and L2 distance error of inference result in Fig. 2

Terrain	Evaluating indicator	PSiamNet（c）	Adapted HardNet（d）	CorASLNet（e）	RPSCCNet（f）
Forest	Heatmap quality	2862.88	11939.88	0.27	28.14
（1）	L2 distance /pixel	1.00	6.08	0.00	2.00
Urban Area	Heatmap quality	11379.40	13830.62	7.08	11.78
（2）	L2 distance /pixel	3.00	4.24	0.00	0.00
Farmland	Heatmap quality	17970.94	15000.8073	60.99	27.42
（3）	L2 distance /pixel	4.47	5.00	45.69	2.00
River	Heatmap quality	18175.43	9731.15	1813.56	32.71
（4）	L2 distance /pixel	51.24	78.81	64.28	2.00
Water Surface	Heatmap quality	11373.06	14464.50	136.06	41.58
（5）	L2 distance /pixel	80.60	66.24	67.67	4.47
Gobi（flat）	Heatmap quality	17252.40	9498.53	474.13	54.71
（6）	L2 distance /pixel	45.18	32.57	69.07	53.66
Gobi	Heatmap quality	16796.47	7643.43	27.01	27.87
（7）	L2 distance /pixel	7.07	87.69	2.00	2.00
Hilly Area	Heatmap quality	13250.74	16251.06	35.37	2.66
（8）	L2 distance /pixel	3.16	66.67	10.00	0.00
River valley	Heatmap quality	4118.53	12901.60	4.73	1.53
（9）	L2 distance /pixel	3.00	8.60	0.0	0.00
Highland （10）	Heatmap quality	9106.88	16033.42	8.69	4.92
Highland （10）	L2 distance /pixel	1.00	29.83	0.00	0.00
Highland	Heatmap quality	11021.97	12959.89	30.17	33.03
（11）	L2 distance /pixel	9.21	63.97	2.00	2.00
Highland	Heatmap quality	7603.97	16355.28	68.42	1.44
（12）	L2 distance /pixel	5.00	86.37	84.85	0.00
Highland	Heatmap quality	8796.72	18123.40	0.39	2.86
（13）	L2 distance /pixel	2.23	69.77	0.00	0.00
Highland	Heatmap quality	13597.99	12910.65	0.93	4.24
（14）	L2 distance /pixel	3.60	83.63	0.00	0.00
Highland	Heatmap quality	5318.50	15361.87	0.48	2.00
（15）	L2 distance /pixel	70.17	74.81	0.00	0.00
Mean Value of Heatmap quality		11247.00	13532.41	177.88	18.46
Mean Value of L2 distance /pixel		19.33	50.95	23.03	4.54
Median Value of Heatmap quality		11373.49	13830.62	27.01	11.78
Median Value of L2 distance /pixel		4.47	66.24	2.00	0.00

Table 2. Matching accuracy and mean value of L2 distance of four methods in test set

View table

Table 2. Matching accuracy and mean value of L2 distance of four methods in test set

Evaluating indicator	RPSCCNet	CorASLNet	PSiamNet	Adapted HardNet
Mean value of L2 distance /pixel	5.347	12.27	17.86	35.03
ACC（L2≤3 pixel）/%	91.03	79.65	55.32	16.72
ACC（L2≤2 pixel）/%	82.49	74.53	23.04	6.01
ACC（L2≤1 pixel）/%	55.49	52.47	6.50	1.75
ACC（L2＝0 pixel）/%	17.37	17.02	0.00	0.00

Table 3. Performance comparison of four methods in Nvidia RTX 2080Ti
View table
Table 3. Performance comparison of four methods in Nvidia RTX 2080Ti
Parameter RPSCCNet CorASLNet PSiamNet Adapted HardNet
Size of Network /MB 85.40 4.96 41.00 5.16
FPS（Network） 82.88 183.37 165.46 364.82
FPS（Single match） 82.88 183.37 0.01 0.02

Tools

Get Citation

Copy Citation Text

Rongping Zou, Bin Zhu, Chenyang Wang, Yaoxuan Zhu, Yangdi Hu. Heterogeneous Remote Sensing Image Matching Algorithm Based on Residual Pseudo-Siamese Convolution Cross-Correlation Network[J]. Laser & Optoelectronics Progress, 2022, 59(12): 1228002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: Apr. 28, 2021

Accepted: Jun. 2, 2021

Published Online: May. 23, 2022

The Author Email: Zhu Bin (zhubin@nudt.edu.cn)

DOI:10.3788/LOP202259.1228002

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Quality of corresponding normalized heat map and L2 distance error of inference result in Fig. 2

Table 1. Quality of corresponding normalized heat map and L2 distance error of inference result in Fig. 2

Table 2. Matching accuracy and mean value of L2 distance of four methods in test set

Table 2. Matching accuracy and mean value of L2 distance of four methods in test set

Table 3. Performance comparison of four methods in Nvidia RTX 2080Ti

Table 3. Performance comparison of four methods in Nvidia RTX 2080Ti