首页|基于多分支空谱特征增强的高光谱图像分类

基于多分支空谱特征增强的高光谱图像分类

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
为了解决高光谱图像自身及分类过程中噪声干扰大、空间-光谱特征信息提取不足以及有限样本下分类性能不佳等问题,提出一种基于多分支空谱特征增强的高光谱图像分类模型SSFE-MBACNN.首先,利用多分支特征提取模块分别提取浅层空谱特征和深层空间特征信息,并引入注意力机制抑制噪声干扰.其次,设计一种改进多尺度空谱特征提取融合模块及结合双池化和空洞卷积的空间特征增强模块实现空谱特征增强,减少模型参数量和提高分类性能.最后,用全局平均池化层代替全连接层,进一步降低参数量,缓解模型过拟合问题.实验结果表明,在Indian Pines(10%训练样本)、Pavia University(5%训练样本)和Salinas(1%训练样本)数据集分别取得了0.990 7、0.997 5和0.994 7的总体分类精度.SSFE-MBACNN不仅能充分利用空谱特征信息,而且在有限样本下也取得了优秀的分类性能,明显高于其他对比方法.
Hyperspectral image classification based on multi-branch spatial-spectral feature enhancement
To solve the problems of high noise interference in the hyperspectral image itself and the process of classification,insufficient extraction of spatial-spectral feature information,and poor classification performance under limited samples,a hyperspectral image classification model SSFE-MBACNN based on multi-branched spatial-spectral feature enhancement is proposed.First,shallow spatial-spectral feature information and deep spatial feature information are extracted separately using multi-branch feature extraction modules,and attention mechanism are introduced to suppress noise interference.Second,an improved fusion module for multi-scale spatial-spectral feature extraction and a spatial feature enhancement module combining dual pooling and dilated convolution are designed to achieve spatial-spectral feature enhancement,reduce the number of model parameters and improve classification performance.Finally,the global average pooling layer is used instead of the fully connected layer to further reduce the number of parameters and alleviate the model overfitting problem.The experimental results show that the overall classification accuracies of 0.990 7,0.997 5 and 0.994 7 are achieved for the Indian Pines(10%training sample),Pavia University(5%training sample)and Salinas(1%training sample)datasets.SSFE-MBACNN makes full use of the spatial-spectral feature information and achieves excellent classification performance with limited samples,which is significantly higher than other comparative methods.

hyperspectral image classificationfeature enhancementmulti-branch feature extractionattention mechanismmulti-scale featuresdual poolingdilated convolution

李铁、李文许、王军国、高乔裕

展开 >

辽宁工程技术大学 电子与信息工程学院,辽宁 葫芦岛 125105

高光谱图像分类 特征增强 多分支特征提取 注意力机制 多尺度特征 双池化 空洞卷积

辽宁省科技厅自然科学研究面上项目辽宁省教育厅科学研究经费项目辽宁省教育厅基本科研面上项目辽宁省教育厅基本科研面上项目

2023-MS-314LJ2020JCL007LJKMZ20220678LJKZ0357

2024

10.37188/CJLCD.2023-0158

液晶与显示
中科院长春光学精密机械与物理研究所 中国光学光电子行业协会液晶分会 中国物理学会液晶分会

液晶与显示

CSTPCD北大核心
影响因子：0.964
ISSN：1007-2780
年,卷(期)：2024.39(6)
  • 7