基于连续尺度特征融合与VAE注意力机制网络的高光谱图像分类方法探析  

Method Exploration of the Classification of Hyperspectral Images Based on Continuous-Scale Feature Fusion with a VAE Attention Mechanism Network

作　　者：袁奇 路绍军 YUAN Qi;LU Shaojun(School of Physics and Electronic Information,Yan’an University,Yan’an Shaanxi 716000,China)

机构地区：[1]延安大学物理与电子信息学院,陕西延安716000

出　　处：《信息与电脑》2024年第16期25-28,共4页Information & Computer

摘　　要：本文提出了基于连续尺度特征融合与变分自编码器(Variational Auto Encoder,VAE)注意力机制网络(Continuous Scale Feature Fusion-VAE Attention Mechanism Network,CSFF-VAEAMN)的高光谱图像分类方法,该方法采用连续尺度特征融合块来捕捉连续层级特征,使模型可以充分利用不同层级的特征关系,同时应用VAE注意力机制在加权时生成更连贯的特征表示,以提高模型在交界区域的分类能力。通过在两个公开数据集上与其他现有方法进行对比实验,本文所提方法被证明在各项性能指标上均优于其他方法。This paper proposed the hyperspectral image classification method based on continuous scale feature fusion and Variational Auto Encoder(VAE)attention mechanism network(CSFF-VAEAMN)to capture continuous scale feature fusion block to enable the model to make full use of the feature relationships of different levels.The VAE attention mechanism is applied to generate more coherent feature representation to improve the classification ability of the model in the boundary area.Through comparative experiments with other existing methods on two public datasets,the performance indicators of the proposed method are proven to be better than those of other methods.

关 键 词：变分自编码器 高光谱图像 特征融合 注意力机制 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]