Abstract Landsat satellite images have become the most widely used data source in large-scale ecological monitoring studies worldwide. In remote sensing application studies of large and medium scale areas, due to seasonal, lighting and climatic conditions and different satellite re-entry cycles and sensors, patchy effects and chromatic unevenness may exist after stitching the mosaic of multi-scene remote sensing images. With the rapid development of remote sensing cloud computing technology, exploring a fast and efficient method to repair Landsat chromatic stripes based on cloud platform is important. In this paper, we propose a histogram image homogenization method based on a random forest algorithm implemented on the Google Earth Engine (GEE) cloud platform, which homogenizes the Landsat Top of Atmosphere (TOA) and Surface Reflectance (SR) of Shanxi Province from 1986 to 2020 (Landsat 5 TM/7 ETM+/8 OLI) normalized vegetation index (NDVI) images after inversion were used as the study data, and MOD13Q1 (250 m resolution), MOD13A1 (500 m resolution) and MOD13A2 (1 km resolution) MODIS datasets were used as the validation data after 2000. The NDVI images of Shanxi Province from 1986 to 2020 before and after image restoration were compared separately, and the results of the study showed that (1) 20 years of the 35-year image analysis had strip color difference problems, and in 1994, for example, the restored Landsat TOA and Landsat SR images compared with those before restoration, the mean NDVI values of the restored areas increased by 32.6% and 29.03% respectively, and the profile analysis showed that the fit increased by 0.162 3 and 0.118 0 respectively; (2) The results of the trend analysis of the 1986—2020 one-dimensional linear regression showed that the fit of the restored images was high and the fluctuation of the year-by-year images was smaller after the long time series analysis. Among them, the slopes of the restored Landsat TOA and SR images decreased by 0.006 2 and 0.006 7, and theR2 improved by 0.024 8 and 0.008 4 respectively; (3) Pearson correlation analysis of Landsat and MODIS images found that the correlation coefficients of the restored Landsat SR and TOA images improved by an average of 0.049 and 0.061 (p<0.05), where the correlation coefficients of restored Landsat SR and TOA images and MOD13Q1, MOD13A1, and MOD13A2 images increased by 0.050, 0.047, 0.049, 0.066, 0.060, and 0.059, respectively; (4) 2000—2020 Landsat and MODIS image time series analysis results show that the overall trend of the restored Landsat images is more similar to MODIS images, and the fit of the restored Landsat TOA and SR images is improved by 0.058 6 and 0.031 9, respectively. The proposed GEE cloud platform-based stochastic The proposed fast image restoration method based on the GEE cloud platform random forest algorithm achieves the accurate evaluation of NDVI inversion results of long time series remote sensing images, and the application of this method can quickly and efficiently solve the chromatic patch and banding effects caused by image mosaic.
Key words：Ecological monitoring; Google Earth Engine; Image mosaic; Image restoration; Random forest; Histogram matching
Corresponding Authors:
LI Jing
E-mail: lijing@cumtb.edu.cn
Cite this article:
YAN Xing-guang,LI Jing,YAN Xiao-xiao, et al. A Rapid Method for Stripe Chromatic Aberration Correction in
Landsat Images[J]. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2023, 43(11): 3483-3491.
[1] Liu P. Frontiers in Environmental Science, 2015, 3: 45.
[2] Li J, Zipper C E, Donovan P F, et al. Environmental Monitoring and Assessment, 2015, 187(9): 557.
[3] LI Jing, DENG Xiao-juan, YANG Zhen, et al(李 晶, 邓晓娟, 杨 震). Spectroscopy and Spectral Analysis(（光谱学与光谱分析), 2019, 39(12): 3788.
[4] LI De-ren, WANG Mi, PAN Jun(李德仁, 王 密, 潘 俊). Geomatics and Information Science of Wuhan University[武汉大学学报(信息科学版)], 2006, 31: 753.
[5] CUI Hao, ZHANG Li, AI Hai-bin, et al(崔 浩, 张 力, 艾海滨, 等). Acta Geodaetica et Cartographica Sinica(测绘学报), 2017, 46(12): 1986.
[6] Lillesand Thomas M, Kiefer Ralph W, Chipman J. Remote Sensing and Image Interpretation 7th ed. American: John Wiley & Sons, Inc, 2015, 289.
[7] LI Shuo, WANG Hui, WANG Li-yong, et al(李 烁, 王 慧, 王利勇). National Remote Sensing Bulletin(遥感学报), 2018, 22(3): 450.
[8] Jiang B, Woodell G A, Jobson D J. Journal of Real-Time Image Processing, 2015, 10(2): 239.
[9] WANG Mi, PAN Jun(王 密, 潘 俊). Remote Sensing for Land and Resources(国土资源遥感), 2006, 18(4): 10.
[10] WANG Wen-tao, WEN De-bao(王文滔, 闻德保). Jiangsu Science & Technology Information(江苏科技信息), 2017, (6): 51.
[11] Liu J, Wang X, Chen M, et al. Remote Sensing, 2014, 6(2): 1102.
[12] Richter R. International Journal of Remote Sensing, 1990, 11(1): 159.
[13] Gorelick N, Hancher M, Dixon M, et al. Remote Sensing of Environment, 2017, 202: 18.
[14] Kumar Lalit, Mutanga Onisimo. Remote Sensing, 2018, 10(10): 1509.
[15] LI Jing, YAN Xing-guang, YAN Xiao-xiao, et al(李 晶, 闫星光, 闫萧萧, 等). China Coal Society(煤炭学报), 2021, 46（5）: 1439.
[16] Helmer E H, Ruefenacht B. Photogrammetric Engineering & Remote Sensing, 2005, 71(9): 1079.
[17] Chen Soong-Der, Ramli A R. IEEE Transactions on Consumer Electronics, 2003, 49(4): 1310.
[18] Breiman L. Machine Learning, 2001, 45：5.
