光学精密工程2026，Vol.34Issue(8)：1283-1297,15.DOI:10.37188/OPE.20263408.1283
融合对象级KPCA-DSFA的高分辨率遥感影像无监督变化检测
Unsupervised change detection from high-resolution remote sensing images with object-level KPCA-DSFA fusion
摘要
Abstract
In order to address the challenges of incomplete identification and inaccurate judgment in land cover monitoring,while balancing integrity and practicality,a novel change detection method based on co-refinement of object-level fusion and graph cut with KPCA-DSFA was proposed,which used two regis-tered high-resolution remote sensing images.Firstly,relative radiometric correction and band stacking fu-sion were performed on the two-phase images.A simple non-iterative clustering algorithm was adopted for joint segmentation to generate homogeneous blocks that preserved the feature consistency of both image phases.Then the kernel PCA convolutional mapping network and deep slow feature analysis were coupled together for spatial-spectral feature extraction and deep semantic parsing respectively.Taking super pixels as the basic processing units,object-level high-dimensional spatial vectors were constructed via feature fu-sion to obtain change confidence information.Finally,an energy function model was established based on the Graph Cut,which leveraged the adjacency relationships and spatial differences of super pixel objects to achieve precise extraction of change regions through global optimization.Experimental results demonstrate that the proposed method achieves an overall accuracy of over 90%with excellent comprehensive perfor-mance.It can effectively suppress"salt-and-pepper"noise,significantly improve the recall rate of change regions,and exhibit favorable superiority and robustness.关键词
核主成分分析/变化检测/超像素/深度特征提取/图割模型Key words
kernel convolution mapping/change detection/super pixels/deep feature analysis/graph cut model分类
信息技术与安全科学引用本文复制引用
宫金杞,王宗晨,王铁,周俊亦..融合对象级KPCA-DSFA的高分辨率遥感影像无监督变化检测[J].光学精密工程,2026,34(8):1283-1297,15.基金项目
湖北珞珈实验室开放基金资助项目(No.230100023) （No.230100023）
江苏省研究生科研与实践创新计划项目(No.SJCX25_0620) （No.SJCX25_0620）
