In viticulture, as in many other fields, the adoption of modern methods is an essential step towards ensuring compliance with quality standards. Remote sensing, which enables vineyards to be mapped, analysed and treated, is today a modern solution for efficiently managing vineyard plots. In addition, the use of multi-spectral sensors, notably infrared and near infrared, provides key indicators to help diagnose a plot, notably through the generation of a normalized difference vegetation index (NDVI) map. Finally, 3D reconstruction methods offer a real contribution to the diagnosis of a vineyard field, since they allow precise measurements to be taken directly from the model. While several solutions exploit the different strengths of these methods, they simply treat these indicators separately. Thus, it is possible to obtain a 3D map from specific sensors, a 2D NDVI map can be generated using infrared images and aerial images are regularly taken from drones or satellites to analyse the evolution of a vineyard. The aim of this thesis is to show that it is possible to combine these indicators in order to provide a 3D NDVI map of a plot from aerial images taken by a drone equipped with a pair of sensors. We propose in the first part of this manuscript the context in which our study is placed. We introduce the notion of precision viticulture, and establish a state of the art of the different remote sensing methods applied to the study of the phytosanitary state of the vine. In the second part, we are interested in 3D models reconstructed by stereo vision, and propose different methods to improve the quality of the cloud of points generated. In a third part, we will focus on the methods of 3D reconstruction using a system with only one camera. In this part, we present a method for optimising the number of images processed to lighten the solutions on embedded systems; as well as a convolutional neural network model allowing the generation of a 3D model from a single image. Finally, we highlight in a last part the use of a multi-spectral mobile stereo system. We provide a public dataset to help the development of future methods and detail the development of a multi-spectral stereo UAV prototype. Finally, we show how to use the images obtained by this prototype to build a 3D NDVI map and demonstrate its advantages over the 2D maps that are used today.