Neural Networks (NNs) are used in various application areas to identify objects. Reliable behavior of NNs is an important aspect, especially for embedded systems. In this paper, we focus on the analysis of NNs to find correlations between their characteristics in order to be reliable and predictable in time, with the aim of making sugar beet recognition more transparent. Although obtaining promising results, as we are only going to focus on analysing models and finding correlations between some of their characteristics, this paper is only the first milestone towards using this correlation to optimize smart farming applications to improve the production and sustainability of the plantations around the world.