融合谱-空域信息的DBM高光谱图像分类方法

doi:10.19665/j.issn1001-2400.2019.03.017

摘要/Abstract

摘要：

在高光谱图像分类问题中,提取能够有效表达地物特征的信息是分类方法中的关键问题。为了提高高光谱图像分类精度,提出一种基于深度玻尔兹曼机的高光谱图像分类方法。该方法首先对高光谱图像数据进行主成分分析法白化处理,并提取像元的空域信息,与像元光谱信息组成综合的谱-空域信息;然后通过多层深度玻尔兹曼机模型从像元的谱-空域信息中提取深层次类别特征;最后通过逻辑回归模型对所提取特征进行分类。这种深度玻尔兹曼机模型能够利用数据的先验知识对高维数据进行特征提取,并且所提取的特征内在地表示了地物的空间结构和光谱特征。实验结果表明,这种方法能够有效地提高高光谱图像的分类精度。

关键词: 高光谱图像, 特征提取, 深度学习, 深度玻尔兹曼机

Abstract:

In the classification of hyperspectral images, extracting more expressive information on the ground objects from the data is a key problem in the classification method. For the purpose of improving classification accuracy, a classification method based on the Deep Boltzmann Machine (DBM) is proposed. First, PCA whitening is performed on the hyperspectral image data, and spatial information on pixels is extracted, followed by the combination with the spectral information on the pixel to construct hybrid spectral-spatial information on pixels; Second, deep features are extracted from the spectral-spatial information on pixels by the multi-layer DBM model; finally, the extracted features are classified based on the logistic regression model. The Deep Boltzmann Machine can extract features from high-dimensional data using prior knowledge, and the extracted features inherently represent the spatial structure and spectral characteristics of objects. Experimental results show that the proposed method can effectively improve the classification accuracy of hyperspectral images.

Key words: hyperspectral image, feature extraction, deep learning, deep Boltzmann machine

中图分类号:

TP751.1

杨建功,汪西莉,刘侍刚. 融合谱-空域信息的DBM高光谱图像分类方法[J]. 西安电子科技大学学报, 2019, 46(3): 109-115.

YANG Jiangong,WANG Xili,LIU Shigang. Spectral-spatial classification of hyperspectral images using deep Boltzmann machines[J]. Journal of Xidian University, 2019, 46(3): 109-115.

图/表 9

图1

图2

图3

图4

表1

图5

图6

图7

表2

参考文献 15

[1]	DU B, ZHANG Y X, ZHANG L P , et al. Beyond the Sparsity-based Target Detector: a Hybrid Sparsity and Statistics-based Detector for Hyperspectral Images[J]. IEEE Transactions on Image Processing, 2016,25(11):5345-5357. doi: 10.1109/TIP.2016.2601268
[2]	BEDINI E, RASMUSSEN T M . Use of Airborne Hyperspectral and Gamma-ray Spectroscopy Data for Mineral Exploration at the Sarfartoq Carbonatite Complex, Southern West Greenland[J]. Geosciences Journal, 2018,22(4):641-651. doi: 10.1007/s12303-017-0078-5
[3]	DU B, ZHANG L P . A Discriminative Metric Learning Based Anomaly Detection Method[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014,52(11):6844-6857. doi: 10.1109/TGRS.2014.2303895
[4]	DU B, ZHANG L P . Random-selection-based Anomaly Detector for Hyperspectral Imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2011,49(5):1578-1589. doi: 10.1109/TGRS.2010.2081677
[5]	张姝茵, 侯彪 . 高概率选择和自适应MRF的极化SAR分类[J]. 西安电子科技大学学报, 2017,44(6):59-64.
	ZHANG Shuyin, HOU Biao . POLSAR Image Classification via High-probability Selection and Adaptive MRF[J]. Journal of Xidian University, 2017,44(6):59-64.
[6]	GHAMISI P, PLAZA J, CHEN Y , et al. Advanced Spectral Classifiers for Hyperspectral Images: A Review[J]. IEEE Geoscience and Remote Sensing Magazine, 2017,5(1):8-32.
[7]	SCHMIDHUBER J . Deep Learning in Neural Networks: an Overview[J]. Neural Networks, 2015,61(1):85-117. doi: 10.1016/j.neunet.2014.09.003
[8]	HINTON G E . Training Products of Experts by Minimizing Contrastive Divergence[J]. Neural Computation, 2002,14(8):1771-1800. doi: 10.1162/089976602760128018
[9]	PAN B, SHI Z W, XU X . MugNet: Deep Learning for Hyperspectral Image Classification Using Limited Samples[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018,145(11):108-119. doi: 10.1016/j.isprsjprs.2017.11.003
[10]	陈希闯 . 基于深度学习的高光谱图像分类[D]. 西安: 西安电子科技大学, 2017.
[11]	CHEN Y S, LIN Z H, ZHAO X , et al. Deep Learning-based Classification of Hyperspectral Data[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2014,7(6):2094-2107. doi: 10.1109/JSTARS.2014.2329330
[12]	CHEN Y S, ZHAO X, JIA X P . Spectral-spatial Classification of Hyperspectral Data Based on Deep Belief Network[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015,8(6):2381-2392. doi: 10.1109/JSTARS.2015.2388577
[13]	DING J, CHEN B, LIU H W , et al. Convolutional Neural Network with Data Augmentation for SAR Target Recognition[J]. IEEE Geoscience and Remote Sensing Letters, 2016,13(3):364-368.
[14]	SALAKHUTDINOV R, HINTON G. Deep Boltzmann Machines [C]//Proceedings of the 2009 12th International Conference on Artificial Intelligence and Statistics. New York: ACM, 2009: 448-455.
[15]	HINTON G E, OSINDERO S, TEH Y W . A Fast Learning Algorithm for Deep Belief Nets[J]. Neural Computation, 2006,18(7):1527-1554. doi: 10.1162/neco.2006.18.7.1527

类别	Indian Pines			Pavia University
类别	Categories	Samples		Categories		Samples
1	Alfalfa	46	Asphalt		6 631
2	Corn-N	1 428	Meadows		18 649
3	Corn-M	830	Gravel		2 099
4	Corn	237	Trees		3 064
5	Grass-M	483	P-M-sheets		1 345
6	Grass-T	730	Bare Soil		5 029
7	Grass-P-M	28	Bitumen		1 330
8	Hay-W	478	S-B-Bricks		3 682
9	Oats	20	Shadows		947
10	Soybean-N	972
11	Soybean-M	2 455
12	Soybean-C	593
13	Wheat	205
14	Woods	1 265
15	Build-G-T-D	386
16	Stone-S-T	93
	总数	10 249	总数		42 776

数据集	度量	RBF-SVM	SAE	DBN	深度玻尔兹曼机
India Pines	OA	71.31±0.061 3	92.39±0.082 4	95.72±0.051 4	96.25±0.022 3
	AA	70.19±0.211 2	92.23±0.241 1	95.63±0.094 3	96.29±0.035 2
	Kappa	67.82±0.002 5	91.75±0.001 3	95.28±0.003 2	96.08±0.004 3
Pavia University	OA	94.82±0.062 1	91.68±0.051 5	96.12±0.062 2	96.81±0.024 1
	AA	94.17±0.043 2	91.31±0.074 4	95.88±0.036 3	96.55±0.048 2
	Kappa	93.19±0.006 2	90.75±0.023 1	95.82±0.016 7	96.36±0.003 8