From pixel to vine parcel: A complete methodology for vineyard delineation and characterization using remote-sensing data

The increasing availability of Very High Spatial Resolution images enables accurate digital maps production as an aid for management in the agricultural domain. In this study, we develop a comprehensive and automatic tool for vineyard detection, delineation and characterization using aerial images and without any parcel plan availability. In France, vineyard training methods in rows or grids generate periodic patterns which make frequency analysis a suitable approach. The proposed method computes a Fast Fourier Transform on an aerial image, providing the delineation of vineyards and the accurate evaluation of row orientation and interrow width. These characteristics are then used to extract individual vine rows, with the aim of detecting missing vine plants and characterizing cultural practices. Using the red channel of an aerial image, 90% of the parcels have been detected; 92% have been correctly classified according to their rate of missing vine plants and 81% according to their cultural practice (weed control method). The automatic process developed can be easily integrated into the final user’s Geographical Information System and produces useful information for vineyard management.