基于慢特征分析与生成对抗网络的林业光学遥感影像薄云去除方法

打开文本图片集

中图分类号：S757 文献标志码：A 文章编号：1000-2006（2026）01-0223-08

Thin cloud removal method for forestry optical remote sensing images based on slow feature analysis and generative adversarial network

ZHU Songyu1'²，LI Chao’，JING Weipeng'

（1.CollegeofComputerandControlEngineering，NortheastForestryUniversity，Harbin150o40，China；2.Colegeof Electronicand Information Enginering，Harbin Vocationaland Technical University，Harbin15oo81，China）

Abstract：Toaddressthe isueof image distortionandreduced usabilitycaused bythincloudremoval inoptical remote sensing images，this studyproposes a novel thin cloudremoval method：SFGAN，that integrates slow feature analysis （SFA）with generative adversarial networks （GANs），aiming toenhance imagequalityand providereliabledata support for forestry remote sensing analysis.【Method】First，a slow-varying featuremoduleis designed to calculate cloud reflectanceand high-dimensional featureslownessTheslow-varying featurevectorsareconcatenated withrandominitial vectors as thegeneratorinput，improving cloud feature recognition.Second，cloudreflectance is utilized asa discrimiativeconstraintfactortoiterativelyoptimizethediscriminator，therebygenerating high-qualitycloud-freeimages through adversarial training.【Result】Experiments onpublic datasets RICE1and PRSC demonstrate thatthe SFGAN outperforms existing methods in both quantitative metrics（e.g.， PSNR=33.740 7 and SSIM=0.958？ 2onRICE1， （20 and SSIM=0.879 2 on PRSC）and visual assessments.Validation using Landsat 8 imagery shows SFGAN achieves superior cloud removal efects in both real and simulated cloud scenarios，with a processing timeof 0.98 secondsper image.【Conclusion】The SFGAN framework effectivelymitigates thincloud interference in forestry optical emotesensing imagesbysynergizing SFA and GANs，significantly improving datausabilityandanalyticalaccuracy at the source level.

Keywords：forestryoptical remote sensing images；thincloudremoval；slow featureanalysis（SFA）；generative adversarial networks（GANs）

随着对地观测技术的不断发展，林业遥感影像在森林资源监测、生态环境保护等方面被广泛应用[1]。（剩余14472字）