Construction of a 3D vineyard model using very high resolution airborne images

In recent years there has been a growth in interest and number of research studies regarding the application of remote optical and thermal sensing by unmanned aerial vehicle (UAV) in agriculture and viticulture. Many papers report on the use of images to map or estimate the growth and water status of plants, or the heterogeneity of different parcels. Most often, NDVI or other similar indices are used. However, analysis of this type of image is difficult in vineyards covered with grass, because contrast between the green of the grass and the green of the vine is low and difficult to classify. This paper presents the acquisition methodology of very high-resolution (5 [cm]) images and their processing to construct a three-dimensional surface model for the creation of precise digital surface and terrain models in order to separate different strata of a vineyard.

The images were acquired with a Sensefly Swinglet CAM unmanned aerial vehicle at an altitude of 110 [m], allowing for a resolution of 5 [cm]. The images were combined using Pix4D software, with a lateral overlap of 75% and a longitudinal overlap of 60%. The produced digital terrain and surface model was subtracted and an extraction mask containing only vine pixel was created. The results show the importance of using a precise digital terrain model. The raster file obtained by subtracting the DSM and the DTM showed values between -0.1 and + 2 m. in good accordance with the average value of the vine. The great majority of pixels fell between the threshold (0.5 [m]) and the topping values 1.6[m]). Using this procedure and parameters, an extremely precise surface model is obtained, as well as the pattern of the vineyard rows and, to some extent, the location of different plants stocks. This mask could be used to analyse images of the same plot taken at different times. The extraction of only vine pixels will facilitate subsequent analyses, for example, a supervised classification of these pixels.