IMPROVING PER-PIXEL CLASSIFICATION OF CROP-SHELTER COVERAGE BY TEXTURE ANALYSES OF HIGH-RESOLUTION SATELLITE PANCHROMATIC IMAGES

Abstract

Actual research challenges in automated recognition of crop shelters regard, among other issues, the accuracy of classification, contour detection and typology identification. In this field the use of high-resolution multispectral images has been found to improve the feature recognition in comparison to RGB images or low resolution multispectral ones. As for classification methodologies, per-pixel and object-oriented ones offer different tools to cope with image recognition and feature extraction. In this study, to improve the classification of cropshelter coverage, the per-pixel method was applied to high-resolution multispectral images, coupled with a texture analysis of high-resolution panchromatic images. In detail, the results of the classification accuracy assessment achieved by the use of native high-resolution panchromatic images and RGB-band images resampled accordingly, were compared with those found in a previous study in which panchromatic images degraded to the RGB-band image resolution were used. The results show that the proposed methodology is suitable to improve crop-shelter classification quality and contour detection of parcels.