Precision agriculture can benefit from the us- age of swarms of drones to monitor a field. Crop/weed classification is a concrete applica- tion that can be efficiently carried out through collaborative approaches, whereby the infor- mation gathered by a drone can be exploited as prior to improve the classification per- formed by other drones observing the same area. In this study, we instantiate this con- cept by exploiting state-of-the-art deep learn- ing techniques. We propose the usage of a shallow convolutional neural network that re- ceives as input, besides the RGB channels of the acquired image, also an additional chan- nel that represents a probability map about the presence of weeds in the observed area. Exploiting a realistic, synthetic dataset, the performance is assessed showing a substancial improvement in the classification accuracy.
