Abstract
This paper proposes an image processing/machine learning system for estimating the amount of biomass in a field. This piece of information is precious in agriculture as it would allow a controlled adjustment of water and fertilizer. This system consists of a flying robot device which captures multiple images of the area under interest. Subsequently, this set of images is processed by means of a combined action of digital elevation models and multispectral images in order to reconstruct a three-dimensional model of the area. This model is then processed by a machine learning device, i.e. a support vector regressor with multiple kernel functions, for estimating the biomass present in the area. The training of the system has been performed by means of three modern meta-heuristics representing the state-of-the-art in computational intelligence optimization. These three algorithms are based on differential evolution, particle swarm optimization, and evolution strategy frameworks, respectively. Numerical Results derived by empirical simulations show that the proposed approach can be of a great support in precision agriculture. In addition, the most promising results have been attained by means of an algorithm based on the differential evolution framework.