Classification and information extraction in very high resolution satellite images for tree crops monitoring

Recent access to Very High Spatial Resolution (VHSR) Satellite Images allows vegetation monitoring at metric and sub-metric scale, with individual trees now detectable. Therefore, it discloses new applications in precision agriculture for orchards and other tree crops. In this paper, we present some methodological directions for classification, and extraction of specific agricultural information from these images. Aims are tree crop detection, plot mapping, species identification, and cropping-system characterization. This latter includes for instance row management (e. g. grid vs. line pattern, width of rows and inter-rows, row orientation), crown shape, and crown size estimation. In this paper, we skip the segmentation step and consider that we have got a precise delimitation of plots that have a homogeneous content. To classify these plots, we have used expert knowledge in agronomy combined with image information in a decision tree. Classification criteria were based on parameters resulting from the Fourier transform analysis or vegetation indices, derived as one single descriptor for the whole plot. As a conclusion, the proposed methodology was found capable of classification and characterization of tree crops, provided the trees are clearly seen from above, and their planting is regular enough to give a response with Fourier analysis.