高级搜索

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于多尺度非对称密集网络的高光谱图像分类

蔡轶珩 谭美伶 潘建军 何楷祺

蔡轶珩, 谭美伶, 潘建军, 何楷祺. 基于多尺度非对称密集网络的高光谱图像分类[J]. 电子与信息学报, 2024, 46(4): 1448-1457. doi: 10.11999/JEIT230651
引用本文: 蔡轶珩, 谭美伶, 潘建军, 何楷祺. 基于多尺度非对称密集网络的高光谱图像分类[J]. 电子与信息学报, 2024, 46(4): 1448-1457. doi: 10.11999/JEIT230651
CAI Yiheng, TAN Meiling, PAN Jianjun, HE Kaiqi. Hyperspectral Image Classification Based on Multi-scale Asymmetric Dense Network[J]. Journal of Electronics & Information Technology, 2024, 46(4): 1448-1457. doi: 10.11999/JEIT230651
Citation: CAI Yiheng, TAN Meiling, PAN Jianjun, HE Kaiqi. Hyperspectral Image Classification Based on Multi-scale Asymmetric Dense Network[J]. Journal of Electronics & Information Technology, 2024, 46(4): 1448-1457. doi: 10.11999/JEIT230651

基于多尺度非对称密集网络的高光谱图像分类

doi: 10.11999/JEIT230651
基金项目: 国家重点研发计划(2017YFC1703302)
详细信息
    作者简介:

    蔡轶珩:女,副教授,研究方向为图像处理和模式识别

    谭美伶:女,硕士生,研究方向为图像和信息处理

    潘建军:男,硕士生,研究方向为图像和信息处理

    何楷祺:男,硕士生,研究方向为图像和信息处理

    通讯作者:

    蔡轶珩 caiyiheng@bjut.edu.cn

  • 中图分类号: TN911.73; TP751

Hyperspectral Image Classification Based on Multi-scale Asymmetric Dense Network

Funds: The National Key Research and Development Program of China (2017YFC1703302)
  • 摘要: 近年来,基于有限标记样本的高光谱图像(HSI)分类方法取得了重大进展。然而,由于高光谱图像的特殊性,冗余的信息和有限的标记样本给提取强判别特征带来了巨大挑战。此外,由于各类别像素分布不均,如何强化中心像素的作用,减弱不同类别的周围像素的负面影响也是提高分类性能的关键。为了克服上述局限性,该文提出一种基于多尺度非对称密集网络(MS-ADNet)的高光谱图像分类方法。首先,提出一个多尺度样本构建模块,通过在每个像素周围提取多个尺度的图像块,并进行反卷积和拼接以构建输入样本,使其既包含详细的结构区域,又包含较大的同质区域;然后,提出一个非对称密集连接结构,在空间和光谱特征联合提取中实现核骨架增强,即增强了方形卷积核的中心十字区域部分提取的特征,有效地促进了特征重用。此外,为了提高光谱特征的鉴别性,提出一种精简的元素光谱注意力机制,并将其置于密集连接网络的前端和后端。在每类仅采用5个样本进行网络训练的情况下,该方法在Indiana Pines, Pavia University和Salinas数据集上的总体准确率分别达到了77.66%, 84.54%和92.39%,取得了极具竞争力的分类结果。
  • 图  1  MS-ADNet结构

    图  2  多尺度样本构建模块

    图  3  非对称密集连接结构

    图  4  元素光谱注意力模块

    图  5  不同空间大小对3个数据集分类准确性的影响

    图  6  不同多尺度方案对3个数据集分类准确性的影响

    图  7  各结构的有效性验证

    表  1  UP数据集上的分类结果对比(%)

    类别3D-CNNFDSSCDFSL+SVMDMVLDCFSLMS-ADNet
    159.8292.0373.4362.6982.2097.04
    263.0596.4489.2596.0087.7498.19
    368.9159.3048.0984.2667.4669.77
    477.3167.7584.7236.7093.1693.49
    590.7796.8299.6586.2299.4998.56
    663.4077.8867.8190.8677.3261.08
    787.6469.5364.4882.1181.1879.13
    857.2773.3667.3779.1066.7374.91
    995.5792.1892.9222.5698.6698.75
    OA65.74±1.7781.10±6.9379.63±1.0973.81±5.4083.65±1.7784.54±0.05
    AA73.72±1.0180.59±5.3176.41±1.3971.17±3.2083.77±1.7485.65±0.03
    Kappa57.37±1.9776.07±8.4873.05±1.6066.87±6.1678.70±2.0180.46±0.06
    下载: 导出CSV

    表  2  SA数据集上的分类结果对比(%)

    类别3D-CNNFDSSCDFSL+SVMDMVLDCFSLMS-ADNet
    195.2998.1673.9297.0699.40100
    297.2099.0196.8597.8399.76100
    391.4595.5996.2897.7991.9695.45
    497.3195.9499.1154.3999.5595.76
    591.2494.5680.7287.3492.7099.62
    698.8099.5691.6390.8399.5299.98
    799.6999.2397.7394.9598.8898.89
    866.4081.4582.3393.3574.5790.43
    996.2599.2994.4498.2799.5999.30
    1070.7294.7780.9697.3686.4295.42
    1193.1594.7993.3876.9196.6195.63
    1299.6599.3297.9496.0699.9398.70
    1392.6396.5395.7976.2499.3096.64
    1493.5685.9798.8771.9698.8591.10
    1568.0265.3471.1382.5375.3873.04
    1681.4199.8790.57100.092.2299.84
    OA84.20±2.6288.62±4.0386.95±1.3089.16±1.6389.34±2.1992.39±0.03
    AA89.56±1.7993.71±1.8290.08±1.4488.30±1.4294.04±1.1495.61±0.01
    Kappa82.46±2.9087.37±4.4485.51±1.4287.98±1.7988.17±2.4091.56±0.03
    下载: 导出CSV

    表  3  IN数据集上的分类结果对比(%)

    类别3D-CNNFDSSCDFSL+SVMDMVLDCFSLMS-ADNet
    195.1267.7996.7524.3395.3777.38
    237.7071.0336.3872.3843.2681.35
    319.7763.6938.3466.7757.9565.49
    432.5163.7077.1663.7380.6057.33
    588.4586.8773.9274.9172.9189.73
    673.6595.0886.2566.4987.9698.03
    781.8232.9297.1029.7999.5743.75
    853.3599.3181.8286.3986.2699.60
    9100.017.2475.5613.7499.3353.62
    1041.3553.2152.2277.6762.4461.22
    1166.7186.9859.9686.6862.7584.58
    1237.4069.7336.5683.2648.7267.83
    1385.7172.7198.0047.0099.3595.43
    1462.5793.4984.6391.1685.4097.01
    1556.4268.5774.1066.4166.6978. 03
    1690.3653.84100.031.1197.6183.27
    OA54.76±0.0372.49±4.1261.69±1.8570.26±4.8666.81±2.3777.66±0.06
    AA63.93±0.0268.51±3.0573.05±0.8461.36±3.0077.89±0.8677.10±0.04
    Kappa48.72±0.0369.38±4.4256.78±1.9066.92±5.1662.64±0.8674.89±0.07
    下载: 导出CSV

    表  4  网络参数量和训练时间对比

    方法INSAUP
    参数量(M)训练时间(min)参数量(M)训练时间(min)参数量(M)训练时间(min)
    3D-CNN0.0961.0390.0981.0510.0480.898
    FDSSC1.2300.9901.2501.1800.6400.260
    DFSL+SVM0.0337.9220.0338.0840.03310.848
    DMVL24.620141.20024.620243.10024.620197.700
    DCFSL4.07020.9204.07021.1204.06020.670
    MS-ADNet2.33018.4302.38018.8901.2105.090
    下载: 导出CSV

    表  5  TeaFarm数据集上的分类结果对比(%)

    类别FDSSCDMVLDCFSLMS-ADNet
    191.6092.2892.6495.59
    253.1881.0074.7962.51
    399.3496.9888.3099.63
    486.9942.3010091.90
    597.7591.2298.6898.88
    647.4824.2489.6391.83
    782.5325.3799.0677.94
    851.9236.2360.5957.71
    910099.4698.8399.99
    1075.0546.0110094.83
    OA86.45±0.0782.38±3.8090.47±1.0991.82±0.01
    AA78.58±0.0663.51±3.5390.25±0.7787.08±0.03
    Kappa81.83±0.0975.47±4.6386.51±1.3288.40±0.02
    下载: 导出CSV
  • [1] YAO Ding, ZHANG Zhili, ZHAO Xiaofeng, et al. Deep hybrid: Multi-graph neural network collaboration for hyperspectral image classification[J]. Defence Technology, 2023, 23: 164–176. doi: 10.1016/j.dt.2022.02.007.
    [2] WANG Xue, TAN Kun, DU Peijun, et al. A unified multiscale learning framework for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 4508319. doi: 10.1109/tgrs.2022.3147198.
    [3] CHEN Yushi, JIANG Hanlu, LI Chunyang, et al. Deep feature extraction and classification of hyperspectral images based on convolutional neural networks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2016, 54(10): 6232–6251. doi: 10.1109/TGRS.2016.2584107.
    [4] LI Ying, ZHANG Haokui, and SHEN Qiang. Spectral-spatial classification of hyperspectral imagery with 3D convolutional neural network[J]. Remote Sensing, 2017, 9(1): 67. doi: 10.3390/rs9010067.
    [5] 刘娜, 李伟, 陶然. 图信号处理在高光谱图像处理领域的典型应用[J]. 电子与信息学报, 2023, 45(5): 1529–1540. doi: 10.11999/JEIT220887.

    LIU Na, LI Wei, and TAO Ran. Typical Application of graph signal processing in hyperspectral image processing[J]. Journal of Electronics & Information Technology, 2023, 45(5): 1529–1540. doi: 10.11999/JEIT220887.
    [6] SHAHRAKI F F and PRASAD S. Graph convolutional neural networks for hyperspectral data classification[C]. 2018 IEEE Global Conference on Signal and Information Processing, Anaheim, USA, 2018: 968–972. doi: 10.1109/GlobalSIP.2018.8645969.
    [7] ZHU Lin, CHEN Yushi, GHAMISI P, et al. Generative adversarial networks for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(9): 5046–5063. doi: 10.1109/TGRS.2018.2805286.
    [8] MOU Lichao, GHAMISI P, and ZHU Xiaoxiang. Deep recurrent neural networks for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(7): 3639–3655. doi: 10.1109/TGRS.2016.2636241.
    [9] ZHU Kaiqiang, CHEN Yushi, GHAMISI P, et al. Deep convolutional capsule network for hyperspectral image spectral and spectral-spatial classification[J]. Remote Sensing, 2019, 11(3): 223. doi: 10.3390/rs11030223.
    [10] ZHONG Zilong, LI J, LUO Zhiming, et al. Spectral-spatial residual network for hyperspectral image classification: A 3-D deep learning framework[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(2): 847–858. doi: 10.1109/TGRS.2017.2755542.
    [11] WANG Wenju, DOU Shuguang, JIANG Zhongmin, et al. A fast dense spectral-spatial convolution network framework for hyperspectral images classification[J]. Remote Sensing, 2018, 10(7): 1068. doi: 10.3390/rs10071068.
    [12] CAI Yiheng, GUO Yajun, LANG Shinan, et al. Classification of hyperspectral images by spectral-spatial dense-residual network[J]. Journal of Applied Remote Sensing, 2020, 14(3): 036513. doi: 10.1117/1.JRS.14.036513.
    [13] FANG Bei, LI Ying, ZHANG Haokui, et al. Hyperspectral images classification based on dense convolutional networks with spectral-wise attention mechanism[J]. Remote Sensing, 2019, 11(2): 159. doi: 10.3390/rs11020159.
    [14] YAN Huaiping, WANG Jun, TANG Lei, et al. A 3D cascaded spectral-spatial element attention network for hyperspectral image classification[J]. Remote Sensing, 2021, 13(13): 2451. doi: 10.3390/rs13132451.
    [15] PAN Jianjun, CAI Yiheng, TAN Meiling, et al. Multiscale residual weakly dense network with attention mechanism for hyperspectral image classification[J]. Journal of Applied Remote Sensing, 2022, 16(3): 034504. doi: 10.1117/1.Jrs.16.034504.
    [16] ZHU Minghao, JIAO Licheng, LIU Fang, et al. Residual spectral-spatial attention network for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(1): 449–462. doi: 10.1109/TGRS.2020.2994057.
    [17] SUN Le, ZHAO Guangrui, ZHENG Yuhui, et al. Spectral-spatial feature tokenization transformer for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5522214. doi: 10.1109/TGRS.2022.3144158.
    [18] CAI Yiheng, XIE Jin, LANG Shinan, et al. Hyperspectral image classification using multi-branch-multi-scale residual fusion network[J]. Journal of Applied Remote Sensing, 2021, 15(2): 024512. doi: 10.1117/1.JRS.15.024512.
    [19] LIU Bing, YU Anzhu, YU Xuchu, et al. Deep multiview learning for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(9): 7758–7772. doi: 10.1109/TGRS.2020.3034133.
    [20] LI Zhaokui, LIU Ming, CHEN Yushi, et al. Deep cross-domain few-shot learning for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5501618. doi: 10.1109/TGRS.2021.3057066.
    [21] XUE Zhixiang, YU Xuchu, LIU Bing, et al. HresNetAM: Hierarchical residual network with attention mechanism for hyperspectral image classification[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 3566–3580. doi: 10.1109/JSTARS.2021.3065987.
    [22] HE Xin, CHEN Yushi, and GHAMISI P. Dual graph convolutional network for hyperspectral image classification with limited training samples[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5502418. doi: 10.1109/TGRS.2021.3061088.
    [23] LI Rui, ZHENG Shunyi, DUAN Chenxi, et al. Classification of hyperspectral image based on double-branch dual-attention mechanism network[J]. Remote Sensing, 2020, 12(3): 582. doi: 10.3390/rs12030582.
    [24] XIE Jie, HE Nanjun, FANG Leyuan, et al. Multiscale densely-connected fusion networks for hyperspectral images classification[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 31(1): 246–259. doi: 10.1109/TCSVT.2020.2975566.
    [25] HU Jie, SHEN Li, and SUN Gang. Squeeze-and-excitation networks[C]. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 7132–7141. doi: 10.1109/CVPR.2018.00745.
    [26] WANG Qilong, WU Banggu, ZHU Pengfei, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks[C]. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 11531–11539. doi: 10.1109/CVPR42600.2020.01155.
    [27] LIU Bing, YU Xuchu, YU Anzhu, et al. Deep few-shot learning for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(4): 2290–2304. doi: 10.1109/TGRS.2018.2872830.
  • 加载中
图(7) / 表(5)
计量
  • 文章访问数:  125
  • HTML全文浏览量:  73
  • PDF下载量:  27
  • 被引次数: 0
出版历程
  • 收稿日期:  2023-06-30
  • 修回日期:  2023-12-05
  • 网络出版日期:  2023-12-14
  • 刊出日期:  2024-04-24

目录

    /

    返回文章
    返回