融合特征优化的跨数据集高光谱图像分类  

Cross-dataset hyperspectral image classification based on fusion feature optimization

作　　者：马晓瑞[1] 哈林 谌敦斌 梅亮 王洪玉[1] Ma Xiaorui;Ha Lin;Shen Dunbin;Mei Liang;Wang Hongyu(School of Information and Communication Engineering,Dalian University of Technology,Dalian 116024,China;School of Optoelectronic Engineering and Instrumentation Science,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]大连理工大学信息与通信工程学院,大连116024 [2]大连理工大学光电工程与仪器科学学院,大连116024

出　　处：《中国图象图形学报》2024年第8期2175-2187,共13页Journal of Image and Graphics

基　　金：国家自然科学基金项目(62271102);辽宁省科技计划项目(2022JH1/10800100);大连市科技创新基金项目(2022JJ11CG002)。

摘　　要：目的高光谱图像分类可实现对地物目标的逐像素识别,是对地观测中的关键技术。由于采集环境变迁与成像设备差异等因素的影响,不同高光谱图像特征分布偏移,影响现有模型的跨数据集分类精度。针对此,提出了一种融合特征优化的无监督跨数据集高光谱图像分类方法。方法提出了基于奇异值抑制的特征均衡策略实现数据内独立优化,通过限制奇异值正则项兼顾特征的可迁移性与可鉴别性;提出了基于隐式增广的特征匹配策略实现数据间特征协同优化,引导源域特征逼近目标域提高模型的泛化性;设计了基于隐式鉴别器的对抗学习框架实现数据间特征类别级优化,提高了预测多样性,实现跨数据集分类。结果实验在Pavia数据集组和HyRANK数据集组上进行,与多种最新的跨数据高光谱图像分类方法进行了对比,在Pavia数据集组中,相比于性能第2的模型,总体精度、平均精度和к系数分别提高了1.75%、3.55%和2.17%;在HyRANK数据集组中,相比于性能第2的模型,总体精度、平均精度和к系数分别提高了6.58%、13.10%和7.96%。同时进行了消融实验,研究了各个模块对高光谱图像分类效果的影响。实验结果表明,每一模块在提高高光谱图像分类效果方面都是有效的。结论本文提出的融合特征优化的跨数据集高光谱图像分类方法可以在无监督的条件下显著提高跨数据集高光谱图像分类精度,提高分类预测的多样性,得到更好的分类效果。Objective Hyperspectral image is a 3D data cube of“space-spectrum integration”.Its high-resolution spectral information can realize fine-grained land cover identification,and its wide-coverage spatial information can complete accu⁃rate land cover mapping.Therefore,hyperspectral images with spatial information and spectral information are widely used in tasks related to land cover classification.Thus,hyperspectral image classification is an important research topic in remote sensing field and a key supporting technology for many Earth observation tasks,such as smart cities,precision agri⁃culture,and modern national defense.In recent years,many classification methods for hyperspectral images have been proposed to mine spatial and spectral information based on different deep networks,and they have achieved unprecedented high-precision classification results.However,due to various factors such as the change in acquisition environment and the difference in imaging sensors,the feature distribution of different hyperspectral images is shifted,which leads to the diffi⁃culty of cross-dataset classification.For this reason,existing classification methods usually retrain the model to deal with new hyperspectral image dataset,which is label intensive and time consuming.In the era of remote sensing big data,devel⁃oping classification methods for cross-dataset hyperspectral images is important.Therefore,this study investigates classifi⁃cation methods for cross-dataset hyperspectral images to achieve large-scale Earth observation missions.Method This study proposes an unsupervised classification method for cross-dataset hyperspectral images based on feature optimization.The proposed method consists of three main modules.First,a feature balancing strategy is proposed to optimize the intradataset features independently.During the adversarial domain adaptation process,the transferability and discriminability of features are contradictory,and most existing methods sacrifice the feature discriminability of target datas

关 键 词：高光谱图像分类 跨数据集分类 特征优化 领域自适应 无监督分类 域对抗网络 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]