融合Transformer和VGG网络的高光谱图像分类  

Fusion of Transformer and VGG networks for hyperspectral image classification

作　　者：张明慧 周浩[1] 王先旺 ZHANG Minghui;ZHOU Hao;WANG Xianwang(School of Information,Yunnan University,Kunming 650500,China)

机构地区：[1]云南大学信息学院,云南昆明650500

出　　处：《传感器与微系统》2023年第12期142-145,150,共5页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(11663007,61802337);云南省自然科学基金资助项目(202001BB050032)。

摘　　要：在高光谱图像(HSI)光谱数据中,相邻波段间信息的相关性对光谱特征近似的不同地物的分析具有重要意义。然而在传统卷积神经网络(CNN)的HSI光谱数据处理方法中,所提取的特征忽略了不同波段间信息的关联性。提出了一种融合Transformer和VGG网络的高光谱图像分类方法(SST_Like)。采用3D卷积核的VGG网络提取空间光谱特征,基于多头自注意力(MSA)机制的Transformer网络提取连续光谱间信息,形成空谱联合特征,最终通过多层感知机(MLP)完成地物分类任务。本文提出的SST_Like网络模型在3个HSI开放数据集上的实验结果表明,与传统基于CNN的HSI分类算法相比,可以提取更加深层的、判别性的特征,具有较高的分类性能。In hyperspectral image(HSI)spectral data,the correlation of information between adjacent bands is important for the analysis of different features with similar spectral characteristics.However,in traditional convolutional neural networks(CNN)method of processing HSI spectral data,the extracted features ignore the correlation of information between different bands.A hyperspectral image classification method that fuses Transformer and VGG networks,SST_Like is proposed,which uses a VGG network with 3D convolution kernel to extract spatial spectral features and a Transformer network based on a self-attentive mechanism to extract continuous inter-spectral information to form joint spatial-spectral features,and finally completes the feature classification task by a multilayer perceptron(MLP).The experimental results of the SST_Like network model proposed in the paper on three HSI open datasets show that it can extract deeper and discriminative features with higher classification performance compared with the traditional CNN-based HSI classification algorithm.

关 键 词：VGG网络 高光谱图像分类 TRANSFORMER 空谱联合特征提取 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]