engleski

The effect of atmospheric corrections and satellite image fusion on radiometric indices values and classification accuracy

This research summarizes different preprocessing steps of atmospheric correction and image fusion methods to analyze radiometric indices values before and after preprocessing and the influence of radiometric indices on classification accuracy using machine learning methods. Images collected using three different satellite missions were used for data analysis: Landsat 8, Sentinel-2 and WorldView-4. To analyze the influence of atmospheric correction on radiometric indices, Dark object subtraction (DOS), Land Surface Reflectance Correction and Sentinel-2 atmospheric correction (S2AC) were compared to satellite images with top-of-the- atmosphere reflectance values. For this analysis, Landsat 8 and Sentinel-2 images were used, along with four radiometric indices namely Normalized difference vegetation index (NDVI), Normalized difference water index (NDWI), Soil-adjusted vegetation index (SAVI) and Modified soil- adjusted vegetation index (MSAVI). Additional analysis was done to observe the influence of atmospheric correction on eight different locations: water, two forest sites, two grasslands, agriculture, building and road. The result showed that NDVI presents the best potential for comparison between sensors. The effect of image fusion on radiometric indices was analyzed using Sentinel-2 and WorldView-4 satellite images. Multi-sensor image fusion was evaluated using Ehlers, Brovey, Modified Intensity-Hue-Saturation (M-IHS) and High-pass filtering image fusion methods to evaluate change of NDVI, Blue normalized difference vegetation index and Green normalized difference vegetation index values. M-IHS is chosen as the best image fusion method based on visual assessment, Cramer- Von-Misses test and difference between fused and original radiometric indices. Influence of five atmospheric corrections, namely S2AC, DOS, Image correction for atmospheric effects (iCOR), Surface reflectance and Standardized surface reflectance on five machine learning classification algorithms, Support vector machine (SVM), Random forest (RF), Extreme gradient boosting (XGB) and CatBoost (CB) was examined. SVM classification method outperformed all other methods with radiometric indices included, but also without included radiometric indices for all twelve dates. For classification with radiometric indices included SVM performed the best for S2AC atmospheric correction with the median value of 96.54%, while for classification without radiometric indices SVM performed the best for STDSREF atmospheric correction with the median value of 96.83%.

Sentinel-2 ; WorldView-4 ; Landsat 8 ; Modified Intensity-Hue-Saturation ; Support Vector Machine ; radiometric indices ; machine learning ; atmospheric correction ; image fusion ; Normalized difference vegetation index

nije evidentirano