AOSN: alpha optimal structure network for height estimation from a single SAR image in mountain areas

doi:10.11947/j.AGCS.2025.20230431

Abstract

Abstract:

Height estimation from a single synthetic aperture radar (SAR) image is a possible way to estimate height in all day and all weather conditions with real time capabilities. However, it is an ill-posed problem since the same 2D images may be projected from multiple 3D images. The development of deep learning provides a possible solution for it. The problems of the current deep learning methods are lack of detail feature information and the accuracy of the estimated height is not high enough. In order to address the above issue, this paper proposes alpha optimal structure network (AOSN), utilizing the characteristics of various feature extraction capabilities from the convolution kernels with different sizes. A structural parameter named α is proposed and it searches the optimal combination of convolution kernels with different sizes, and the residual block is introduced and the transposed convolution operation is used instead of the unpooling operation, which improves the final accuracy of height estimation. The experiments carried on the datasets on Guiyang, Huangshan, Geermu demonstrate that the proposed method outperforms the state-of-the-art in mountain areas.

Key words: SAR image, height estimation, deep learning, structural parameter

CLC Number:

P237

Qingli LUO, Xueyan LI, Guoman HUANG, Honghui CHEN, Minglong XUE, Jian LI. AOSN: alpha optimal structure network for height estimation from a single SAR image in mountain areas[J]. Acta Geodaetica et Cartographica Sinica, 2025, 54(5): 888-898.

Figures/Tables 15

Fig. 1

Fig. 2

Tab. 1

Fig. 3

Fig. 4

Fig. 5

Fig. 6

Fig. 7

Tab. 2

Fig. 8

Fig. 9

Tab. 3

Fig. 10

Tab. 4

Fig. 11

References 25

[1]	罗卿莉, 崔峰志, 魏钜杰, 等. SAR影像变化检测的前景特征流形排序法[J]. 测绘学报, 2022, 51(11): 2365-2378. DOI: . doi: 10.11947/j.AGCS.2024.20240040
	LUO Qingli, CUI Fengzhi, WEI Jujie, et al. Foreground feature manifold ranking method for SAR image change detection[J]. Acta Geodaetica et Cartographica Sinica, 2022, 51(11): 2365-2378. DOI: . doi: 10.11947/j.AGCS.2024.20240040
[2]	STUBBS D C, SILWAL L, THUROW B S, et al. Three-dimensional measurement of the crater formation during plume-surface interactions using stereo-photogrammetry[J]. AIAA Journal, 2022, 60(3): 1316-1331.
[3]	WU Songbo, ZHANG Bochen, DING Xiaoli, et al. A hybrid method for MT-InSAR phase unwrapping for deformation monitoring in urban areas[J]. International Journal of Applied Earth Observation and Geoinformation, 2022, 112: 102963.
[4]	LUO Qingli, LI Jian, ZHANG Yuanzhi. Monitoring subsidence over the planned Jakarta-Bandung (Indonesia) high-speed railway using Sentinel-1 multi-temporal InSAR data[J]. Remote Sensing, 2022, 14(17): 4138.
[5]	DANA K J, VAN GINNEKEN B, NAYAR S K, et al. Reflectance and texture of real-world surfaces[J]. ACM Transactions on Graphics, 1999, 18(1): 1-34.
[6]	WU B, LIU W C, GRUMPE A, et al. Construction of pixel-level resolution DEMs from monocular images by shape and albedo from shading constrained with low-resolution DEM[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 140: 3-19.
[7]	张艳超, 庄载椿, 肖宇钊, 等. 基于运动恢复结构算法的油菜NDVI三维分布[J]. 农业工程学报, 2015, 31(17): 207-214.
	ZHANG Yanchao, ZHUANG Zaichun, XIAO Yuzhao, et al. Rape plant NDVI 3D distribution based on structure from motion[J]. Transactions of the Chinese Society of Agricultural Engineering, 2015, 31(17): 207-214.
[8]	MOULON P, MONASSE P, MARLET R. Adaptive structure from motion with a Contrario model estimation[C]//Proceedings of 2012 Asian Conference on Computer Vision. Berlin: Springer, 2013: 257-270.
[9]	SARITAS M M, YASAR A. Performance analysis of ANN and naive Bayes classification algorithm for data classification[J]. International Journal of Intelligent Systems and Applications in Engineering, 2019, 7(2): 88-91.
[10]	ZHOU Tinghui, BROWN M, SNAVELY N, et al. Unsupervised learning of depth and ego-motion from video[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 6612-6619.
[11]	LIU B Y, GOULD S, KOLLER D. Single image depth estimation from predicted semantic labels[C]//Proceedings of 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Francisco: IEEE, 2010: 1253-1260.
[12]	SHI Wenzhe, CABALLERO J, HUSZAR F, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 1874-1883.
[13]	FAROOQ BHAT S, ALHASHIM I, WONKA P. AdaBins: depth estimation using adaptive bins[C]//Proceedings of 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville: IEEE, 2021: 4008-4017.
[14]	REZA M A, KOSECKA J, DAVID P. FarSight: long-range depth estimation from outdoor images[C]//Proceedings of 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems. Madrid: IEEE, 2018: 4751-4757.
[15]	CHO J, MIN Dongbo, KIM Y, et al. Deep monocular depth estimation leveraging a large-scale outdoor stereo dataset[J]. Expert Systems with Applications, 2021, 178: 114877.
[16]	FENG Qi, SHUM H P H, MORISHIMA S. 360 depth estimation in the wild - the Depth360 dataset and the SegFuse network[C]//Proceedings of 2022 IEEE Conference on Virtual Reality and 3D User Interfaces. Christchurch: IEEE, 2022: 664-673.
[17]	YUAN Haobo, LI Xiangtai, YANG Yibo, et al. PolyphonicFormer: unified query learning for depth-aware video panoptic segmentation[C]//Proceedings of 2022 European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 582-599.
[18]	KIM J, LEE J K, LEE K M. Accurate image super-resolution using very deep convolutional networks[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 1646-1654.
[19]	GUIZILINI V, AMBRUS R, PILLAI S, et al. 3D packing for self-supervised monocular depth estimation[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 2482-2491.
[20]	SRIVASTAVA S, VOLPI M, TUIA D. Joint height estimation and semantic labeling of monocular aerial images with CNNS[C]//Proceedings of 2017 IEEE International Geoscience and Remote Sensing Symposium. Fort Worth: IEEE, 2017: 5173-5176.
[21]	MOU Lichao, ZHU Xiaoxiang. IM2HEIGHT: height estimation from single monocular imagery via fully residual convolutional-deconvolutional network[EB/OL]. [2024-02-04]. https://arxiv.org/abs/1802.10249v1.
[22]	GHAMISI P, YOKOYA N. IMG2DSM: height simulation from single imagery using conditional generative adversarial net[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(5): 794-798.
[23]	XUE Minglong, LI Jian, ZHAO Zheng, et al. SAR2HEIGHT: height estimation from a single SAR image in mountain areas via sparse height and proxyless depth-aware penalty neural architecture search for Unet[J]. Remote Sensing, 2022, 14(21): 5392.
[24]	RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]//Proceedings of 2015 International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer International Publishing, 2015: 234-241.
[25]	XUE Minglong, LI Jian, LUO Qingli. A geometry-aware consistent constraint for height estimation from a single SAR imagery in mountain areas[J]. IEEE Geoscience and Remote Sensing Letters, 2023, 20: 1-5.

模块	层	操作	输出尺寸
Input		拼接	（256，256，3）
	Down1	3×3Conv+5×5Conv+Pool	（128，128，64）
Down Block	Down2	3×3Conv+5×5Conv+Pool	（64，64，128）
	Down3	3×3Conv+5×5Conv+Pool	（32，32，256）
	[Conv1	Conv	（32，32，512）
5×[Down Resblocks]	Conv2	Conv	（32，32，512）
	Conv3］	Conv	（32，32，256）
	[Conv1	Conv	（32，32，512）
5×[Up Resblocks]	Conv2	Conv	（32，32，512）
	Conv3］	Conv	（32，32，512）
	Up1	3×3TransConv+5×5 TransConv+Bilinear	（32，32，256）
Up Block	Up2	3×3TransConv+5×5 TransConv+Bilinear	（64，64，128）
	Up3	3×3TransConv+5×5 TransConv+Bilinear	（128，128，64）
Output	Out	TransConv	（256，256，1）

数据集	残差模块和转置卷积模块	结构参数α	RMSE/m	MAE/m	SSIM
贵阳	N	N	43.18	29.63	0.33
	Y	N	36.72	24.51	0.47
	N	Y，Trainable	34.81	23.77	0.54
	Y	Y，Fixed	28.93	20.08	0.56
	Y	Y，Trainable	24.54	16.37	0.68
黄山	N	N	49.05	33.56	0.37
	Y	N	39.92	31.09	0.42
	N	Y，Trainable	39.47	30.26	0.47
	Y	Y，Fixed	30.10	27.64	0.49
	Y	Y，Trainable	26.29	21.64	0.61
格尔木	N	N	51.02	44.57	0.28
	Y	N	46.19	39.28	0.44
	N	Y，Trainable	44.16	35.80	0.52
	Y	Y，Fixed	36.69	32.94	0.56
	Y	Y，Trainable	29.19	24.16	0.58

数据集	网络	RMSE/m	MAE/m	SSIM
贵阳	PackNet	28.67	19.73	0.54
	VDSR	31.93	23.59	0.49
	AOSN	24.54	16.37	0.68
黄山	PackNet	29.48	23.19	0.57
	VDSR	34.38	26.11	0.52
	AOSN	26.29	21.64	0.61
格尔木	PackNet	35.58	28.53	0.51
	VDSR	41.51	34.89	0.42
	AOSN	29.19	24.16	0.58

数据集	RMSE/m	SSIM
格尔木测试集	29.19	0.58
黄山数据集	192.25	0.31