PATCH-BASED IMAGE CLASSIFICATION FOR SENTINEL-1 AND SENTINEL-2 EARTH OBSERVATION IMAGE DATA PRODUCTS

摘要

In an era where the satellite image collections are in arncontinuous growth, Earth Observation (EO) image annotationrnand classification is becoming an important componentrnof data exploitation. In this paper we present howrnfeature extraction methods such as Gabor (G) and WeberrnLocal Descriptor (WLD) are performing in a patchbasedrnapproach in the frame of Sentinel-1 and Sentinel-2rnimage data analysis. Having the goal to develop an applicationrncapable to join feature extraction and classificationrnalgorithms, in our assessment, we performed supervisedrnsupport vector machines (SVM) and k-NearestrnNeighbors (k-NN) classifications to extract a few genericrnclasses from synthetic aperture radar (SAR), multispectralrn(MSI) and data fusion (DFI) images. The result ofrnthis study is intended to establish the optimum number ofrnclasses that can be found in the Sentinel-1 and Sentinel-2rnimages when using patch based image classification techniques.rnAlso another important objective of this paper isrnto determine the best patch sizes suitable for this classificationrntype in order to return best results for Sentinel-1rnand Sentinel-2 EO images.