基于三维卷积神经网络的高光谱图像分类
Hyperspectral Image Classification Based on Three-dimensional Convolutional Neural Network
张岩 1沈金悦 1邵钰奕 1卢瑶1
作者信息
- 1. 青岛科技大学机电工程学院 青岛 266061
- 折叠
摘要
针对传统高光谱图像分类方法难以有效提取空间-光谱联合信息以及卷积神经网络难以有效关注到重要特征的现象,论文提出了基于卷积神经网络的3D-CNN结合CBAM注意力机制的高光谱图像分类方法.高光谱数据由于其高维性,给特征提取方面带来了困难,因此,论文采用三维卷积神经网络(3D-CNN)提取特征,结合视觉注意力机制使卷积神经网络更好地关注重要特征.通过在两个公共数据集上验证,证明论文方法能够取得较好的分类精度.
Abstract
Aiming at the phenomenon that traditional hyperspectral image classification methods are difficult to effectively ex-tract space-spectral joint information and convolutional neural networks are difficult to effectively pay attention to important fea-tures,this paper proposes a hyperspectral image based on convolutional neural network 3D-CNN combined with CBAM attention mechanism classification.Hyperspectral data brings difficulties to feature extraction due to its high dimensionality.Therefore,this paper uses three-dimensional convolutional neural network(3D-CNN)to extract features,combined with visual attention mecha-nism to make convolutional neural network more important.feature.Through verification on two public data sets,it is proved that the method in this paper can achieve better classification accuracy.
关键词
图像分类/卷积神经网络/注意力机制/分类精度Key words
image classification/convolutional neural network/attention mechanism/classification accuracy引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据