基于级联网络的高光谱图像变化检测  

Hyperspectral Image Change Detection Based on Cascaded Networks

作　　者：谢福鼎[1] 录金秀 XIE Fuding;LU Jinxiu(College of Geographical Sciences,Liaoning Normal University,Dalian 116029,China)

机构地区：[1]辽宁师范大学地理科学学院,辽宁大连116029

出　　处：《徐州工程学院学报（自然科学版）》2023年第3期61-67,共7页Journal of Xuzhou Institute of Technology(Natural Sciences Edition)

基　　金：国家自然科学基金项目(41801340);辽宁省教育厅自然科学研究项目(LJ2019013)。

摘　　要：遥感变化检测是对地观测的核心问题之一,在许多重要领域都有着广泛的应用.已有的方法是对由2幅影像产生的差值图像进行分类得到变化的结果,但这样容易引起原始图像信息的部分丢失,进而影响检测的结果.为了解决这一问题,提出1种基于孪生级联网络的高光谱变化检测方法.该方法利用块操作和级联网络分别提取2幅影像的光谱特征,并通过特征融合实现高光谱图像的变化检测.相比于基于差分影像的变化检测方法,该方法的优点是在特征提取过程中融合了影像的空间信息,使得所提出的方法具有良好的检测能力.3个实际的数据集上的实验结果和比较结果验证了该方法的有效性.Remote sensing change detection is one of the core issues in Earth observation and has wide applications in many important fields.The existing methods classify the difference images generated by two images to obtain changes,but this can easily cause partial loss of original image information,thereby affecting the detection results.To address this issue,a hyperspectral change detection method based on twin cascaded networks is proposed.This method utilizes block operations and cascaded networks to extract spectral features from two images,and achieves change detection of hyperspectral images through feature fusion.Compared to the change detection method based on differential images,the advantage of this method is that it integrates the spatial information of the image during the feature extraction process,making the proposed method the good detection ability.The experimental and comparative results on three actual datasets validate the effectiveness of this method.

关 键 词：高光谱图像 变化检测 特征融合 级联网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP751[自动化与计算机技术—控制科学与工程]