卷积循环神经网络的高光谱图像解混方法

doi:10.19665/j.issn1001-2400.20241009

摘要/Abstract

摘要：

针对传统的解混方法和基于自编码器的解混网络方法,利用空间信息提升了解混性能,但未深入挖掘和利用光谱特征,而光谱特征和空间信息的有效结合能够进一步提高解混性能,因此,提出了基于双向卷积长短期记忆网络的解混框架。该框架采用创新性的网络结构设计,通过卷积层深入挖掘空间特征,同时利用卷积长短期记忆单元充分挖掘波段间的光谱变异性及其光谱相关性,有效处理光谱维度的序列信息,从而实现对高光谱数据更加精准和高效的分析。为了更加细致地区分和利用高光谱数据中不同谱段的特异性,采用深度光谱分区方法优化网络输入,通过自适应学习机制对不同光谱区域精细化处理,增强了模型对高光谱数据中复杂光谱关系的捕捉能力,进一步提升网络的解混性能。在模拟和多个真实高光谱数据集上的对比实验表明,该方法在解混精度和模型鲁棒性等方面均优于现有方法,特别是在处理复杂地物光谱特征时,表现出良好的泛化能力和稳定性,能够准确估计端元和丰度。

关键词: 高光谱图像, 循环神经网络, 自编码器, 卷积长短期记忆网络, 深度光谱分区

Abstract:

While traditional unmixing methods and autoencoder-based unmixing networks have improved the unmixing performance by utilizing spatial information,they have not fully explored and leveraged spectral features.The effective integration of spectral features with spatial information could further enhance the unmixing performance.Therefore,an unmixing framework based on a Bidirectional Convolutional Long Short-Term Memory Autoencoder Network(CLAENet) with an innovative network architecture design is proposed.This framework deeply mines spatial features through convolutional layers,while convolutional long short-term memory units are used to fully explore spectral variability and the correlations between bands,effectively processing the sequential information on the spectral dimension for a more accurate and efficient analysis of hyperspectral data.To further distinguish and utilize the specificity of different spectral bands in hyperspectral data,a deep spectral partitioning method is adopted to optimize the network input.An adaptive learning mechanism is employed for refined processing of different spectral regions,enhancing the model's capability to capture complex spectral relationships within hyperspectral data and further improving unmixing performance.Comparative experiments conducted on simulated and multiple real hyperspectral datasets demonstrate that this method outperforms existing methods in terms of unmixing accuracy and model robustness.Notably,it exhibits good generalization and stability when handling complex spectral features of land cover,thus accurately estimating endmembers and abundances.

Key words: hyperspectral imaging, recurrent neural networks, autoencoders, convlstm, deep spectral partitioning

中图分类号:

TP751/P2

孔繁锵, 余圣杰, 王坤, 方煦, 吕志杰. 卷积循环神经网络的高光谱图像解混方法[J]. 西安电子科技大学学报, 2025, 52(1): 142-151.

KONG Fanqiang, YU Shengjie, WANG Kun, FANG Xu, LV Zhijie. Hyperspectral image unmixing method based on convolutional recurrent neural networks[J]. Journal of Xidian University, 2025, 52(1): 142-151.

图/表 11

图1

图2

图3

图4

表1

表2

表3

图5

表4

表5

图6

参考文献 28

[1]	蔡明娇, 蒋俊正, 蔡万源, 等. 张量分解和自适应图全变分的高光谱图像去噪[J]. 西安电子科技大学学报, 2024, 51(2):157-169.
	CAI Mingjiao, JIANG Junzheng, CAI Wanyuan, et al. Hyperspectral Image Denoising Based on Tensor Decomposition and Adaptive Weight Graph Total Variation[J]. Journal of Xidian University, 2024, 51(2):157-169.
[2]	DENG C, CHEN Y, ZHANG S, et al. Robust Dual Spatial Weighted Sparse Unmixing for Remotely Sensed Hyperspectral Imagery[J]. Remote Sensing, 2023, 15(16):4056.
[3]	DUAN Y, XU X, LI T, et al. UnDAT:Double-Aware Transformer for Hyperspectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61:5522012,1-12.
[4]	REN L, HONG D, GAO L, et al. Orthogonal Subspace Unmixing to Address Spectral Variability for Hyperspectral Image[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61:5501713,1-13.
[5]	CELIK B. QLSU(QGIS Linear Spectral Unmixing) Plugin:An Open Source Linear Spectral Unmixing Tool for Hyperspectral & Multispectral Remote Sensing Imagery[J]. Environmental Modelling & Software, 2023, 168:105782.
[6]	FENG X R, LI H C, WANG R, et al. Hyperspectral Unmixing Based on Nonnegative Matrix Factorization:A Comprehensive Review[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022, 15:4414-4436.
[7]	LI J, AGATHOS A, ZAHARIE D, et al. Minimum Volume Simplex Analysis:A Fast Algorithm for Linear Hyperspectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2015, 53(9):5067-5082.
[8]	FANG Y, WANG Y, XU L, et al. BCUN:Bayesian Fully Convolutional Neural Network for Hyperspectral Spectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60:5523714,1-14.
[9]	CUI C, WANG X, WANG S, et al. Unrolling Nonnegative Matrix Factorization with Group Sparsity for Blind Hyperspectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61:1-12.
[10]	IORDACHE M D, BIOUCAS-DIAS J M, PLAZA A. Sparse Unmixing of Hyperspectral Data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2011, 49(6):2014-2039.
[11]	CHEN B, LOU Y, BERTOZZI A L, et al. Graph-Based Active Learning for Nearly Blind Hyperspectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61:5523716,1-16.
[12]	RASTI B, HONG D, HANG R, et al. Feature Extraction for Hyperspectral Imagery:The Evolution from Shallow to Deep:Overview and Toolbox[J]. IEEE Geoscience and Remote Sensing Magazine, 2020, 8(4):60-88.
[13]	苏远超, 许若晴, 高连如, 等. 基于深度学习的高光谱遥感图像混合像元分解研究综述[J]. 遥感学报, 2024, 28(1):1-19.
	SU Yuanchao, XU Ruoqing, GAO Lianru, et al. Development of Deep Learning-Based Hyperspectral Remote Sensing Image Unmixing[J]. National Remote Sensing Bulletin, 2024, 28(1):1-19.
[14]	CHEN J, ZHAO M, WANG X, et al. Integration of Physics-Based and Data-Driven Models for Hyperspectral Image Unmixing:A Summary of Current Methods[J]. IEEE Signal Processing Magazine, 2023, 40(2):61-74.
[15]	SU Y, LI J, PLAZA A, et al. Daen:Deep Autoencoder Networks for Hyperspectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(7):4309-4321.
[16]	PALSSON B, SIGURDSSON J, SVEINSSON J R, et al. Hyperspectral Unmixing Using a Neural Network Autoencoder[J]. IEEE Access, 2018, 6:25646-25656.
[17]	GAO L, HAN Z, HONG D, et al. Cycu-Net:Cycle-Consistency Unmixing Network by Learning Cascaded Autoencoders[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60:1-14.
[18]	SU Y, ZHU Z, GAO L, et al. DAAN:A Deep Autoencoder-Based Augmented Network for Blind Multilinear Hyperspectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62:1-15.
[19]	管业鹏, 苏光耀, 盛怡. 双向长短期记忆网络的时间序列预测方法[J]. 西安电子科技大学学报, 2024, 51(3):103-112.
	GUAN Yepeng, SU Guangyao, SHENG Yi. Time Series Prediction Method Based on the Bidirectional Long Short-Term Memory Network[J]. Journal of Xidian University, 2024, 51(3):103-112.
[20]	ZHAO M, YAN L, CHEN J. LSTM-DNN Based Autoencoder Network for Nonlinear Hyperspectral Image Unmixing[J]. IEEE Journal of Selected Topics in Signal Processing, 2021, 15(2):295-309.
[21]	LIU Y, LI J, PLAZAA, et al. Spectral Partitioning for Hyperspectral Remote Sensing Image Classification[C]//2014 IEEE Geoscience and Remote Sensing Symposium. Piscataway:IEEE, 2014:3434-3437.
[22]	CAVALLI R M. Spatial Validation of Spectral Unmixing Results:A Systematic Review[J]. Remote Sensing, 2023, 15(11):2822.
[23]	ELKHOLY M M, MOSTAFA M, EBIED H M, et al. Hyperspectral Unmixing Using Deep Convolutional Autoencoder[J]. International Journal of Remote Sensing, 2020, 41(12):4799-4819.
[24]	PALSSON B, ULFARSSON M O, SVEINSSON J R. Convolutional Autoencoder for Spectral-Spatial Hyperspectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 59(1):535-549.
[25]	OZKAN S, KAYA B, AKAR G B. Endnet:Sparse Autoencoder Network for Endmember Extraction and Hyperspectral Unmixing[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 57(1):482-496.
[26]	YU Y, MA Y, MEI X, et al. Multi-Stage Convolutional Autoencoder Network for Hyperspectral Unmixing[J]. International Journal of Applied Earth Observation and Geoinformation, 2022, 113:102981.
[27]	CLARK R N, SWAYZE G A, WISE R, et al. Usgs Digital Spectral Library Splib06a[R]. U.S. Geological Survey,Open File Report, 2003:3395.
[28]	DAVIS C O, KAVANAUGH M, LETELIERR, et al. Spatial and Spectral Resolution Considerations for Imaging Coastal Waters[C]// Coastal Ocean Remote Sensing.Washington,D.C.:SPIE, 2007, 6680:196-207.

SNR/dB		CyCU	DAEU	CNNAEU	MSNeT	EndNet	CLAENet
5	RMSE	0.198 2	0.147 0	0.223 4	0.167 1	0.124 5	0.090 0
	SAD	0.147 6	0.123 6	0.173 0	0.159 2	0.116 1	0.087 3
10	RMSE	0.186 7	0.136 4	0.207 7	0.144 8	0.097 1	0.086 5
	SAD	0.123 3	0.112 5	0.152 6	0.141 6	0.089 0	0.078 6
20	RMSE	0.157 5	0.109 5	0.189 9	0.112 3	0.087 2	0.068 2
	SAD	0.105 4	0.093 7	0.125 8	0.098 8	0.078 8	0.061 0
30	RMSE	0.140 1	0.095 5	0.171 3	0.103 6	0.085 9	0.062 3
	SAD	0.093 0	0.082 3	0.111 4	0.067 2	0.072 1	0.056 1

算法		CyCU	DAEU	CNNAEU	MSNeT	EndNet	CLAENet
SAD	土壤	0.048 5	0.050 1	0.043 1	0.033 4	0.045 3	0.029 3
	树木	0.038 0	0.061 0	0.046 4	0.036 7	0.041 5	0.034 6
	水	0.187 1	0.049 1	0.121 4	0.051 1	0.066 0	0.051 2
平均SAD		0.091 2	0.053 4	0.070 3	0.040 3	0.050 9	0.038 4

算法		CyCU	DAEU	CNNAEU	MSNeT	EndNet	CLAENet
RMSE	土壤	0.191 1	0.088 6	0.235 9	0.074 5	0.080 7	0.062 3
	树木	0.209 0	0.085 6	0.215 9	0.044 0	0.072 0	0.039 2
	水	0.151 6	0.033 3	0.226 3	0.027 8	0.047 9	0.041 5
平均RMSE		0.183 9	0.069 2	0.226 1	0.048 7	0.066 8	0.047 6

算法		CyCU	DAEU	CNNAEU	MSNeT	EndNet	CLAENet
SAD	土壤	0.138 4	0.026 8	0.158 5	0.042 6	0.060 4	0.034 6
	树木	0.117 4	0.049 1	0.085 0	0.029 2	0.051 6	0.041 2
	水	0.177 5	0.053 4	0.059 5	0.061 5	0.134 1	0.050 6
	道路	0.054 7	0.061 3	0.110 2	0.070 8	0.076 5	0.049 3
平均SAD		0.122 0	0.047 7	0.103 3	0.051 1	0.080 6	0.043 9

算法		CyCU	DAEU	CNNAEU	MSNeT	EndNet	CLAENet
RMSE	土壤	0.145 3	0.081 2	0.215 2	0.088 5	0.105 7	0.075 4
	树木	0.100 4	0.067 3	0.187 6	0.070 6	0.096 5	0.061 6
	水	0.101 6	0.132 1	0.173 1	0.112 3	0.125 6	0.092 4
	道路	0.117 6	0.181 4	0.236 3	0.090 5	0.137 9	0.061 2
平均RMSE		0.116 2	0.115 5	0.203 1	0.090 5	0.116 4	0.072 7