Precision agriculture can optimize the production of agricultural crops by analyzing aerial images with varying resolutions and acquired from different sources. It is widely accepted that machine learning (ML) model, especially deep neural networks (DNN), are very efficient for image segmentation. DNNs have been used to segment complex texture and planting structures, such as vine lines, due to their variations in shape, color and orientation. However existing DNNs reach their limits to segment aerial images with varying resolution and multiple instance of vine lines crossing a entire image. In this paper, we present an improvement of the generalization capabilities of ML models to segment vine lines in satellite images. An approach from a previous works that combine neural networks and other classifiers allow us to improve the classification and generalize the models that increase the f-score by 17%.