Objective Hyperspectral imaging technology captures image data across a wide range of wavelengths, providing rich spectral information for each pixel in the scene. This detailed spectral information enables precise identification and classification of various materials and land cover types, making HSIC a critical task in remote sensing. The vast amount of data contained in hyperspectral images presents significant challenges and opportunities for advanced image processing techniques, particularly those involving deep learning. Over the past decade, deep learning has revolutionized numerous fields, including image processing and classification. In the context of HSIC, deep learning techniques, especially those utilizing convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have demonstrated remarkable improvements in performance. These methods effectively capture the complex and high-dimensional nature of hyperspectral data, extracting both spatial and spectral features to enhance classification accuracy. Methods based on spatial-spectral features have gained substantial attention due to their ability to integrate spatial context with spectral information. These methods can better discriminate between different classes by combining these two types of features. However, a limitation arises when applying sample partitioning strategies from natural image classification directly to hyperspectral images. This approach often leads to unintended overlap between training and test samples, particularly in spatially contiguous regions, which can artificially inflate classification accuracy and reduce the model's generalization capability. Methods This study proposed a unique approach that reduced the size of input image patches in spatial-spectral-based methods, enabling an effective separation of the training and test sets. When the image patches were reduced to a specific size, their features gradually transformed into pure spectral features. Considering this observation, this study refocused on the HSIC method based on spectral features and proposed an improved method that relied on deep pixel pair features for faster and more robust hyperspectral image classification. The proposed method improved the original pixel pair feature method in two main aspects. The first key improvement introduced was the distance-constrained pixel pair generation method. Traditional pixel pair methods often suffered from inefficiencies and the inclusion of redundant or irrelevant pixel pairs. The proposed method ensured that only meaningful and diverse pixel pairs were selected by incorporating distance constraints, enhancing the training process and the robustness of the model. This method not only efficiently constructed a large-scale pixel pair training set but also eliminated redundant and unreasonable pixel pairs, leading to more accurate and efficient model training. The second major enhancement was the result-reuse voting strategy combined with a classification method. This strategy optimized the classification process by reusing intermediate results, reducing the computational burden, and accelerating the overall classification time. This approach enhanced efficiency while maintaining high classification accuracy by avoiding redundant model calls. The result-reuse voting strategy ensured that the classification process remained both effective and efficient, making it suitable for large-scale hyperspectral datasets. Results and Discussions Extensive experiments were conducted on public hyperspectral image datasets to validate the effectiveness of the proposed method. The results demonstrated that the proposed method outperformed existing spatial-spectral feature-based methods in terms of classification accuracy. In addition, compared to standard pixel pair methods, the proposed approach significantly reduced computational time, making it more suitable for practical applications. The experimental results confirmed that the proposed method achieved higher classification accuracy in practical applications and significantly improved time efficiency compared to existing methods. Conclusions Accordingly, this study addresses a critical challenge in hyperspectral image classification by proposing a novel approach that shifts the emphasis from spatial-spectral features to purely spectral features through the reduction of the input image block size. The introduction of a distance-constrained pixel-pair generation method and a result-reuse voting strategy significantly enhances classification efficiency and robustness. The experimental results validate the superiority of the proposed method, highlighting its potential for broader applications in remote sensing. This research contributes to the development of more accurate and generalizable HSIC models, paving the way for future advancements in hyperspectral image analysis by addressing the overlap issue between training and test samples. The study highlights the importance of considering the unique characteristics of hyperspectral data and provides a robust framework for leveraging spectral features to achieve high-precision classification. The proposed method represents a significant advancement in HSIC, providing a practical and efficient solution to challenges inherent in existing methods. Future research will explore the integration of additional constraints and optimizations to improve the performance and applicability of hyperspectral image classification techniques.
本文方法采用3个公共的高光谱图像分类数据集来评估提出的FasterDPPF方法的有效性和效率,分别是印第安纳松木数据集(Indiana Pines dataset, IP数据集)、萨利纳斯谷数据集(Salinas Valley dataset, SV数据集)和肯尼迪航天中心数据集(Kennedy Space Centre dataset, KSC数据集)。
LandgrebeD.Hyperspectral image data analysis[J].IEEE Signal Processing Magazine,2002,19(1):17‒28. doi:10.1109/79.974718
[2]
MalthusT J, MumbyP J.Remote sensing of the coastal zone:An overview and priorities for future research[J].International Journal of Remote Sensing,2003,24(13):2805‒2815. doi:10.1080/0143116031000066954
[3]
Bioucas-DiasJ M, PlazaA, Camps-VallsG,et al.Hyperspectral remote sensing data analysis and future challenges[J].IEEE Geoscience and Remote Sensing Magazine,2013,1(2):6‒36. doi:10.1109/mgrs.2013.2244672
[4]
DuBo, ZhangLiangpei.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]
BandosT V, BruzzoneL, Camps-VallsG.Classification of hyperspectral images with regularized linear discriminant analysis[J].IEEE Transactions on Geoscience and Remote Sensing,2009,47(3):862‒873. doi:10.1109/tgrs.2008.2005729
[6]
MelganiF, BruzzoneL.Classification of hyperspectral remote sensing images with support vector machines[J].IEEE Transactions on Geoscience and Remote Sensing,2004,42(8):1778‒1790. doi:10.1109/tgrs.2004.831865
[7]
VillaA, BenediktssonJ A, ChanussotJ,et al.Hyperspectral image classification with independent component discriminant analysis[J].IEEE Transactions on Geoscience and Remote Sensing,2011,49(12):4865‒4876. doi:10.1109/tgrs.2011.2153861
[8]
ZhongYanfei, ZhangLiangpei.An adaptive artificial immune network for supervised classification of multi-/hyperspectral remote sensing imagery[J].IEEE Transactions on Geoscience and Remote Sensing,2012,50(3):894‒909. doi:10.1109/tgrs.2011.2162589
[9]
AudebertN, Le SauxB, LefevreS.Deep learning for classification of hyperspectral data:A comparative review[J].IEEE Geoscience and Remote Sensing Magazine,2019,7(2):159‒173. doi:10.1109/mgrs.2019.2912563
[10]
GhamisiP, MaggioriE, LiShutao,et al.New frontiers in spectral-spatial hyperspectral image classification:The latest advances based on mathematical morphology,Markov random fields,segmentation,sparse representation,and deep learning[J].IEEE Geoscience and Remote Sensing Magazine,2018,6(3):10‒43. doi:10.1109/mgrs.2018.2854840
[11]
LiShutao, SongWeiwei, FangLeyuan,et al.Deep learning for hyperspectral image classification:An overview[J].IEEE Transactions on Geoscience and Remote Sensing,2019,57(9):6690‒6709. doi:10.1109/tgrs.2019.2907932
[12]
LicciardiG, MarpuP R, ChanussotJ,et al.Linear versus nonlinear PCA for the classification of hyperspectral data based on the extended morphological profiles[J].IEEE Geoscience and Remote Sensing Letters,2012,9(3):447‒451. doi:10.1109/lgrs.2011.2172185
[13]
HeKaiming, ZhangXiangyu, RenShaoqing,et al.Deep residual learning for image recognition[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Las Vegas:IEEE,2016:770‒778. doi:10.1109/cvpr.2016.90
[14]
SzegedyC, LiuWei, JiaYangqing,et al.Going deeper with convolutions[C]//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Boston:IEEE,2015:1‒9. doi:10.1109/cvpr.2015.7298594
[15]
KrizhevskyA, SutskeverI, HintonG E.ImageNet classification with deep convolutional neural networks[J].Communications of the ACM,2017,60(6):84‒90. doi:10.1145/3065386
[16]
Ben HamidaA, BenoitA, LambertP,et al.3‒D deep learning approach for remote sensing image classification[J].IEEE Transactions on Geoscience and Remote Sensing,2018,56(8):4420‒4434. doi:10.1109/tgrs.2018.2818945
[17]
RoyS K, KrishnaG, DubeyS R,et al.HybridSN:Exploring 3‒D‒2‒D CNN feature hierarchy for hyperspectral image classification[J].IEEE Geoscience and Remote Sensing Letters,2020,17(2):277‒281. doi:10.1109/lgrs.2019.2918719
[18]
ZhongZilong, LiJ, LuoZhiming,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
[19]
PaolettiM E, HautJ M, Fernandez‒BeltranR,et al.Deep pyramidal residual networks for spectral-spatial hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing,2019,57(2):740‒754. doi:10.1109/tgrs.2018.2860125
[20]
RoyS K, MannaS, SongTiecheng,et al.Attention-based adaptive spectral-spatial kernel ResNet for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing,2021,59(9):7831‒7843. doi:10.1109/tgrs.2020.3043267
[21]
ZhuMinghao, JiaoLicheng, LiuFang,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
[22]
FangLeyuan, LiShutao, DuanWuhui,et al.Classification of hyperspectral images by exploiting spectral-spatial information of superpixel via multiple kernels[J].IEEE Transactions on Geoscience and Remote Sensing,2015,53(12):6663‒6674. doi:10.1109/tgrs.2015.2445767
[23]
LiShutao, LuTing, FangLeyuan,et al.Probabilistic fusion of pixel-level and superpixel-level hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing,2016,54(12):7416‒7430. doi:10.1109/tgrs.2016.2603190
[24]
ZhongZilong, LiYing, MaLingfei,et al.Spectral-spatial transformer network for hyperspectral image classification:A factorized architecture search framework[J].IEEE Transactions on Geoscience and Remote Sensing,2021,60:5514715. doi:10.1109/tgrs.2021.3115699
[25]
SongRuoxi, FengYining, ChengWei,et al.BS2T:Bottleneck spatial-spectral transformer for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing,2022,60:5532117. doi:10.1109/tgrs.2022.3185640
[26]
SunLe, ZhaoGuangrui, ZhengYuhui,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
[27]
RoyS K, DeriaA, HongDanfeng,et al.Multimodal fusion transformer for remote sensing image classification[J].IEEE Transactions on Geoscience and Remote Sensing,2023,61:5515620. doi:10.1109/tgrs.2023.3286826
[28]
HongDanfeng, HanZhu, YaoJing,et al.SpectralFormer:Rethinking hyperspectral image classification with transformers[J].IEEE Transactions on Geoscience and Remote Sensing,2021,60:5518615. doi:10.1109/tgrs.2021.3130716
[29]
HeJi, ZhaoLina, YangHongwei,et al.HSI‒BERT:Hyperspectral image classification using the bidirectional encoder representation from transformers[J].IEEE Transactions on Geoscience and Remote Sensing,2020,58(1):165‒178. doi:10.1109/tgrs.2019.2934760
[30]
YuHaoyang, XuZhen, ZhengKe,et al.MSTNet:A multilevel spectral‒spatial transformer network for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing,2022,60:5532513. doi:10.1109/tgrs.2022.3186400
[31]
GaoLiang, LiuHui, YangMinhang,et al.STransFuse:Fusing swin transformer and convolutional neural network for remote sensing image semantic segmentation[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2021,14:10990‒11003. doi:10.1109/jstars.2021.3119654
[32]
LangeJ, CavallaroG, GötzM,et al.The influence of sampling methods on pixel-wise hyperspectral image classification with 3D convolutional neural networks[C]//Proceedings of the IGARSS 2018‒2018 IEEE International Geoscience and Remote Sensing Symposium.Valencia:IEEE,2018:2087‒2090. doi:10.1109/igarss.2018.8518671
[33]
ChenYushi, LinZhouhan, ZhaoXing,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
[34]
ZhaoChunhui, WanXiaoqing, YanYiming.Spectral-spatial classification of hyperspectral images based on joint bilateral filter and stacked sparse autoencoder[C]//Proceedings of the 2017 First International Conference on Electronics Instrumentation & Information Systems(EIIS).Harbin:IEEE,2017:1‒5. doi:10.1109/eiis.2017.8298563
[35]
YangXiaofei, YeYunming, LiXutao,et al.Hyperspectral image classification with deep learning models[J].IEEE Transactions on Geoscience and Remote Sensing,2018,56(9):5408‒5423. doi:10.1109/tgrs.2018.2815613
[36]
HangRenlong, LiuQingshan, HongDanfeng,et al.Cascaded recurrent neural networks for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing,2019,57(8):5384‒5394. doi:10.1109/tgrs.2019.2899129
[37]
ZhouFeng, HangRenlong, LiuQingshan,et al.Hyperspectral image classification using spectral-spatial LSTMs[J].Neurocomputing,2019,328:39‒47. doi:10.1016/j.neucom.2018.02.105
[38]
LiWei, WuGuodong, ZhangFan,et al.Hyperspectral image classification using deep pixel-pair features[J].IEEE Transactions on Geoscience and Remote Sensing,2017,55(2):844‒853. doi:10.1109/tgrs.2016.2616355