Fusion of SAR and SPOT image data for crop mapping

This paper describes image processing and fusion for land cover mapping focusing on mapping agricultural crops. To achieve a frequent update of crop maps, classification and segmentation algorithms were applied to 13 test cases. Reference data for classification were available from 302 crop fields, and reference data for segmentation from vector data of field boundaries. The reference vector dataset was enhanced by means of visual image interpretation. Radiometric accuracy of segmentation results was evaluated using ratio images. These studies show synergy of radar and optical imagery, with dependence on the sequence of pre-processing and processing techniques in the mapping procedure. Classification accuracy of crop maps based on synthetic aperture radar (SAR), visible-infrared (VIR) and fused imagery reached 82%, 92% and 76% respectively. We conclude that majority based object classification does not improve significantly the overall accuracy. Subsets resulting from the optimum index factor (OIF) algorithm proved slightly better than principal component analysis (PCA) fusion and intensity-hue-saturation (IHS) colour transformation proved worse.