Paolo Remagnino

Plants are the backbone of all life on Earth providing us with food and oxygen. A good understanding of plants can make substantial contributions in identifying new or rare plant species to improve the drug industry, balance the ecosystem as well as the agricultural productivity and sustainability. However, tasks related to the description and categorisation of plants are still tedious for botanists, due to our limited knowledge on the World's plant families, in particular near tropical areas, where flora is richer and more varied. Recent development in science and technology has enabled the computer vision approach to assist botanists with the plant classification task. In particular, using image analysis based on computational tools, raw image data are converted to a suitable internal computer representation, the feature vectors of plant characters, from which classifiers or any machine learning algorithm can be employed in the identification process. Classic image processing algorithms were very limited, requiring a partial manual extraction of features. More recently, machine learning approaches have improved the process but they are strongly reliant of the existence of large dataset used to train a computational model. In recent publications, shape, texture, and colour have been the most common features used in plant classification. In this presentation, the audience will be introduced to the plant identification problem, some of the classic methods and then to the deep-learning approach. In this talk, our involvement in the H2020 Natural Intelligence project will be described, as well as the challenges the project poses to existing solutions and the step change required to design a successful solution. The talk will also introduce how the use of a robotic platform can be employed to extract information in a given habitat for the assessment of its health, introducing complexity to the problem at hand, but also a means to extract additional information and improve the identification process.