改进Res2Net和注意力机制的高光谱图像分类

doi:10.3778/j.issn.1002-8331.2211-0021

摘要/Abstract

摘要： 针对目前以卷积神经网络（CNN）为框架的高光谱图像分类模型参数量多，训练时间长，对样本数量依赖性大的问题，提出了一种改进Res2Net和注意力机制的高光谱图像分类模型。该模型首先使用主成分分析（PCA）对原始图像的通道维度进行降维，将降维后的数据输入三维空洞卷积层，并添加空间注意力模块以强化空间纹理特征；将所得特征映射输入两组空间-深度可分离残差结构结合通道注意力模块中，使用全局平均池化层将输出映射转换成一维向量；经过Softmax分类器获得分类标签。实验结果显示，该模型参数数量少，收敛速度快，使用少量训练样本在Indian Pines和Pavia University数据集上总体分类精度（OA）分别为98.95%和99.46%。

关键词: 高光谱图像, 注意力机制, 残差网络, 可分离卷积

Abstract: Aiming at the problems of large number of parameters, long training time and great dependence on the number of samples in the current hyperspectral image classification model based on convolutional neural network（CNN）, a hyperspectral image classification model based on Res2Net and attention mechanism is proposed. The model firstly uses PCA to conduct channel dimensionality reduction for spectral dimensions of original images, and inputs the data after dimensionality reduction into 3-D dilated convolution layer, and adds SAM to enhance spatial texture features. Then, the obtained feature maps are input into two groups of spatial-depth separable residual structure combined with CAM. The output feature data is converted into one dimensional vector by global average pooling layer. Finally, classification labels are obtained by Softmax classifier. Experimental results show that the model has fewer parameters and fast convergence speed. OA of the Indian Pines and Pavia University datasets with a small number of training samples are 98.95% and 99.46%.

Key words: hyperspectral image, attention mechanism, residual networks, separable convolution

王燕, 王振宇. 改进Res2Net和注意力机制的高光谱图像分类[J]. 计算机工程与应用, 2023, 59(19): 151-158.

WANG Yan, WANG Zhenyu. Hyperspectral Image Classification Based on Res2Net and Attention Mechanism[J]. Computer Engineering and Applications, 2023, 59(19): 151-158.

参考文献

[1] SCHIMLECK L，MATOS J，HIGA A，et al.Classifying wood properties of loblolly pine grown in southern Brazil using NIR-hyperspectral imaging[J].Forests，2020，11（6）：1-13.
[2] DING J L.Exploring the potential of UAV hyperspectral image for estimating soil salinity：effects of optimal band combination algorithm and random forest[J].Spectrochimica Acta Part A—Molecular and Biomolecular Spectroscopy，2022，279：12416.
[3] JING H Y，WANG Y Y，DU Z H，et al.Hyperspectral image classification with a multiscale fusion-evolution graph convolutional network based on a feature-spatial attention mechanism[J].Remote Sensing，2022，14（11）：2633.
[4] KIM J，KAWAMURA Y，NISHIKAWA O，et al.A system of the granite weathering degree assessment using hyperspectral image and CNN[J].International Journal of Mining，Reclamation and Environment，2022，36（5）：368-380.
[5] GROTTE M E，BIRKELAND R，HONIRE-LIVERMORE E，et al.Ocean color hyperspectral remote sensing with high resolution and low latency—the HYPSO-1 CubeSat mission[J].IEEE Transactions on Geoscience and Remote Sensing，2022，60：1-19.
[6] TYLER N，GABRIEL D P，DAVID J M，et al.The influence of aerial hyperspectral image processing workflow on nitrogen uptake prediction accuracy in maize[J].Remote Sensing，2022，14（1）：132.
[7] ZHANG L L，ZHAO C H.Sparsity divergence index based on locally linear embedding for hyperspectral anomaly detection[J].Journal of Applied Remote Sensing，2016，10（2）：25-26.
[8] LI W，ZHANG P，ZHANG L F，et al.GPU parallel implementation of isometric mapping for hyperspectral classification[J].Materials Today：Proceedings，2017，14（9）：1532-1539.
[9] MOU L C，ZHU X X.Learning to pay attention on spectral domain：a spectral attention module-based convolutional network for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing，2020，58（1）：110-122.
[10] ZHOU P C，HAN J W，CHENG G，et al.Learning compact and discriminative stacked autoencoder for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing，2019，57（7）：4823-4833.
[11] ZHANG P，GONG Z Q，LI S T，et al.Learning to diversify deep belief networks for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing，2017，55（6）：3516-3530.
[12] ZHU M H，JIAO L C，LIU F，et al.Residual spectral-spatial attention network for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing，2021，59（1）：449-462.
[13] MEN H，YUAN H C，SHI Y，et al.A residual network with attention module for hyperspectral information of recognition to trace the origin of rice[J].Spectrochimica Acta Part A—Molecular and Biomolecular Spectroscopy，2021，263：120155.
[14] ZHANG T Y，SHI C P，LIAO D L，et al.A spectral spatial attention fusion with deformable convolutional residual network for hyperspectral image classification[J].Remote Sensing，2021，13：3590.
[15] WANG Y，LI K，XU L，et al.A depth-wise separable fully convolutional ResNet with ConvCRF for semi-supervised hyperspectral image classification[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing，2021，14：4621-4632.
[16] LI N Y，WANG Z H.Spatial attention guided residual attention network for hyperspectral image classification[J].IEEE Access，2022，10：9830-9847.
[17] DENG Z Q，WANG Y，LI L W，et al.An attention in volved network stacked by dual-channel residual block for hyperspectral image classification[J].Infrared Physics & Technology，2022，122：104083.
[18] GAO S H，CHENG M M，ZHAO K，et al.Res2Net：a new multi-scale backbone architecture[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2021，43（2）：652-662.
[19] HU W S，LI H C，PAN L，et al.Spatial-spectral feature extraction via deep ConvLSTM neural networks for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing，2020，58（6）：4237-4250.
[20] PAOLETTI M E，MARIO H J 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.
[21] PAOLETTI M E，MORENO A S，HAUT J M.Multiple attention-guided capsule networks for hyperspectral image classification[J].IEEE Transactions on Geoscience and Remote Sensing，2022，60：5520420.
[22] LI L，YIN J H.Joint spatial-spectral attention network for hyperspectral image classification[J].IEEE Geoscience and Remote Sensing Letters，2021，18（10）：1816-1820.