摘要
【目的】卫星遥感是目前作物物候监测的主要手段,具有监测面积大、数据获取便捷等多重优点,但用于精准观测的高分辨率遥感卫星回访周期较长,而且易受大气、云雾等不利条件的影响,这使得在高分辨率时序作物物候检测的数据稀疏,无法在变化较快的作物生长期提供足够的影像。本研究提出一种新的方法,以期提高时间序列监测的准确性,更好地实现物候的精准检测。【方法】以水田和旱田为研究样地,探索结合生成式图像处理技术的时序遥感数据填补方法,用于影像重建的轻量化超分辨率生成对抗网络(generative adversarial networks,GAN),并以重建数据为基础进行作物生长季密集时序监测与物候提取。【结果】①在影像的超分辨率重建方面,基于本研究提出的方法,结构相似性指数(structural similarity,SSIM)和峰值信噪比(peak signal to noise ratio,PSNR)分别达0.834和28.69,相较于目前的主流方法可以更好地重建异源遥感数据;②时序重建后,2个样地的遥感影像的重访周期分别由6.40、6.63 d降至5.70、5.88 d,空间分辨率提升至10 m;③4种平滑方法的物候提取结果,与原始数据提取结果相比具有一定的差异性,优于目前广泛用于时序填补的基于插值的方法。【结论】本研究提出的方法可以有效填补卫星影像时间序列,增强观测数据的连续性,并进行高时空分辨率精准作物物候监测。
[Objective]Satellite remote sensing has merged as the primary approach for monitoring crop phenology.This method has many advantages,including large monitoring area and convenient data acquisition.However,high-resolution remote sensing satellites,which are essential for accurate observation,have a long revisit period.Inevitably multiple dates are affected by unfavorable conditions such as atmospheric and cloud fog.As result,the data for high-resolution time-series crop phenology detection becomes sparse,failing to provide sufficient images during the rapidly changing crop growth period.A new method is proposed to enhance the accuracy of time series monitoring and achieve precse phenology detection.[Method]Paddy fields and dry fields were selected as research plots.First,a time-series remote sensing data filling method wes explored by combining with generative image processing technology.Then a lightweight super-resolution reconstruction generative adversarial networks(GAN)was proposed for image reconstruction.Finally,the reconstructed data were utilized to conduct intensive time series monitoring and phenological extraction of crop growing season.[Result](1)In terms of image super-resolution reconstruction,the proposed method achieved values of 0.834 and 28.69 in structural similarity(SSIM)and peak signal to noise ratio(PSNR),respectively.It can reconstruct heterologous remote sensing data more effectively than mainstream methods.(2)After time series reconstruction,the revisit period of remote sensing images in 2 experimental areas decreased from 6.40 d and 6.63 d to 5.70 d and 5.88 d respectively,and the spatial resolution increased to 10 m.(3)Regarding phenological extraction,the extraction results of 4 smoothing methods differed from those of original data extraction,and were superior to the interpolation-based methods commonly used for time series imputation.[Conclusion]The proposed method can effectively fill the time-series of satellite images,enhance the continuity of observation data,and enable accurate high spatiotemporal-resolution monitoring of crop phenology.
作者
韩昊
冯子怡
蔡渊吉
许童羽
HAN Hao;FENG Ziyi;CAI Yuanji;XU Tongyu(College of Information and Electrical Engineering,Shenyang Agricultural University,Shenyang 110866,Liaoning,China;High-resolution Earth Observation System,Liaoning Forest and Grass Resources and Environment Remote Sensing Research and Application Center,Shenyang Agricultural University,Shenyang 110866,Liaoning,China;National Digital Agriculture Regional Innovation Center(Northeast),Shenyang Agricultural University,Shenyang 110866,Liaoning,China)
出处
《浙江农林大学学报》
北大核心
2025年第4期657-666,共10页
Journal of Zhejiang A&F University
基金
辽宁省教育厅青年项目(JYTQN2023301)
辽宁省基础应用研究(2023JH2/101300120)。
关键词
物候提取
超分辨率重建
卫星遥感
作物物候学
phenology extraction
super-resolution reconstruction
satellite remote sensing
crop phenology
