Hyperspectral imagery for precision agriculture

Abstract

The emerging theme of Precision Agriculture defines a multidisciplinary subject of investigation of soil and plant tissues, in which the radiometric thematic information acquired far beyond the visible field is essential to develop sustainable, economical and effective cultivation processes. Overcoming the well-known applications based on images acquired from a satellite platform, useful in small-scale analysis with minimum cell dimensions of a few meters (or a few tens of meters), the use of autonomous terrestrial means, properly equipped with multi/hyperspectral sensors (robotic platforms), makes it possible to acquire information on a very large scale that, also through automatic procedures, allows the development of a great variety of different approaches to the cultivated territory, promoting a precise and not generalized investigation. In this context, the described study aims to present a particular application addressed to the elaboration and interpretation of images acquired in a vineyard in the province of Asti, images derived from the use of the Rikola hyperspectral camera directly on the ground, to estimate beforehand the production.

In the pre-processing, the acquired images were calibrated in both geometric and radiometric terms, generating the calibrated images subjected to classification in order to extract the bunches of grapes using Artificial Intelligence techniques. These bunches of grapes were subsequently modeled in three-dimensional terms using digital photogrammetry methods according to the Structure from Motion approach, proceeding with the evaluation of their volume: the final estimation of the overall production of the vineyard was made through the development of am approximate law empirically verified.