关键词: 高光谱遥感数据, 地物识别, 核-主成分分析, 概率神经网络, 半数重采样

Abstract: Hyperspectral remote sensing data can provide more abundant information about ground objects, but its high data dimension and strong redundancy also lead to low efficiency of traditional classification methods. To address this issue, this paper proposes an enhanced rapid hyperspectral remote sensing classification algorithm combining kernel-principal component analysis(K-PCA)and probabilistic neural network(PNN). Firstly, a nearest neighbor sample selection method is proposed to screen more representative ground object spectral data. Secondly, a principal component selection strategy based on half resampling is proposed to effectively remove noise and retain the essential spectral features, significantly- DOI:10.3969/j.issn.1000-1565.2025.04.002一种改进的K-PCA与PNN结合的快速高光谱遥感分类算法简萌¹,陈旭凤²,鲁军²,郝敏钗²( 1.北京工业大学 信息科学技术学院,北京 100124;2.河北工业职业技术大学 智能制造学院,河北 石家庄 050091)摘 要:高光谱遥感数据可以提供更加丰富的地物信息,但因其数据维度高、冗余性强等特点导致传统分类方法效率低下.针对此问题本文提出一种改进的核-主成分分析(kernel-principal component analysis,K-PCA)与概率神经网络(probabilistic neural network,PNN)结合的快速高光谱遥感分类算法.首先提出一种最近邻的样本选择方法,用以筛选更具代表性的地物光谱数据;其次提出一种基于半数重采样的主成分优选策略,有效去除噪声并保留光谱本质特征,大幅度降低数据维度;最后融合K-PCA的非线性降维特性与PNN的最优贝叶斯分类能力进行地物识别.在利用AVIRIS高光谱数据集的验证实验中,本算法不仅将分类精度提升至89.9%,较传统方法提升显著,且运算效率大幅提升.结果表明该算法在兼顾分类精度与实时性的高光谱地物识别场景中凸显优势,为遥感大数据智能处理提供了高效解决方案.关键词:高光谱遥感数据;地物识别;核-主成分分析;概率神经网络;半数重采样中图分类号:O433 文献标志码:A 文章编号:1000-1565(2025)04-0343-09A rapid hyperspectral remote sensing classification algorithm combining improved K-PCA and PNNJIAN Meng¹,CHEN Xufeng², LU Jun², HAO Minchai²(1.School of Information Science and Technology,Beijing University of Technology,Beijing 100124,China; 2.School of Intelligent Manufacturing, Hebei Vocational University of Industry and Technology, Shijiazhuang 050091, China)Abstract: Hyperspectral remote sensing data can provide more abundant information about ground objects, but its high data dimension and strong redundancy also lead to low efficiency of traditional classification methods. To address this issue, this paper proposes an enhanced rapid hyperspectral remote sensing classification algorithm combining kernel-principal component analysis(K-PCA)and probabilistic neural network(PNN). Firstly, a nearest neighbor sample selection method is proposed to screen more representative ground object spectral data. Secondly, a principal component selection strategy based on half resampling is proposed to effectively remove noise and retain the essential spectral features, significantly- 收稿日期:2024-09-11;修回日期:2025-05-06 基金项目:中央引导地方科技发展资金项目(216Z1004G);河北省高校硬质合金切削材料应用技术研发中心基金;河北工业职业技术大学博士基金项目(bz202205) 第一作者:简萌(1987—),女,蒙古族,北京工业大学教授,博士,主要从事数据挖掘、多媒体计算研究.E-mail:jianmeng648@163.com 通信作者:陈旭凤(1991—),女,河北工业职业技术大学讲师,主要从事深度学习技术研究.E-mail:xufengchen_gyznzz@163.com 第4期简萌等:一种改进的K-PCA与PNN结合的快速高光谱遥感分类算法河北大学学报(自然科学版) 第45卷reducing the data dimension. Finally, by integrating the nonlinear dimensionality reduction characteristics of K-PCA and the optimal Bayesian classification ability of PNN, in the verification experiment using the AVIRIS hyperspectral dataset, the algorithm not only improves the classification accuracy to 89.9%, a significant improvement over traditional methods, but also greatly enhances the operational efficiency. Experiments show that this algorithm demonstrates outstanding advantages in hyperspectral ground object recognition scenarios where both classification accuracy and real-time performance need to be considered, providing an efficient solution for intelligent processing of remote sensing big data.

Key words: hyperspectral remote sensing data, ground object recognition, K-PCA, PNN, half-sample resampling