The increase of global warming and pollution is a real danger to the survival of millions of species in natural habitats around the World. This is a real threat, for which the European Union has adopted deeply transformative policies detailed in the European Green Deal. Among these, a prominent role is given to the restoring and preservation of ecosystems by increasing the coverage of protected biodiversity rich land and sea areas building on the Natura 2000 Network.
At present, human operators are the only option to perform the monitoring of such a large area, because of their specific expertise and knowledge in identifying plants and assess threats to a habitat, and their ability to move, explore and assess in wild unstructured environments such dunes, forests, and mountains.
The reality is that the availability of botanists, as human operators, has rapidly decreased during the last decades. The artificial alternative is robotics, with an agile robotic platform carrying a varied and intelligent array of sensors. Indeed, robotics has made tremendous advancements in recent years, however robots hardly leave laboratories and factories because they are not robust and efficient to survive in the real world. Robot intelligence is also a clear limitation to what is required in botanical field work. The topic of this PhD course of studies will be mainly concerned with the development of artificial intelligence methods for the automatic recognition and identification of the health of a habitat. This will be an integrating aspect of the
H2020 Natural Intelligence European project (https://www.nih2020.eu), in collaboration with European botanists and roboticists.
Proposed Methodology
The programme of work will entail an in-depth study of the existing methods used for the identification of a small set of key indicators of the health of a natural habitat. This could be the amount of green areas or the presence of a plant, either positive (for instance a wild orchid) or negative (lack of green areas) to the habitat.
Research will then have to be carried out to fill the state of the art gaps, employing the latest machine learning and image interpretation methods. The development will exploit new large datasets collected by the botanists in four European very different habitats (dunes, prairies, forests and mountains) and the latest deep learning techniques, employed for identification, disambiguation and recognition of the sought for species. Ultimately, data from the natural habitats will be collected semi-automatically with the ANYMAL robot
(https://rsl.ethz.ch/robots-media/anymal.html), with an on-board sensor array, inclusive of a camera in the visible and non visible spectra, LiDAR, humidity and temperature sensors. The richness of the sensor array will allow the development of multimodal algorithms, making an innovative contribution to research in a number of fields, such as image and video interpretation, artificial intelligence, machine learning, pattern recognition, botany and robotics.
The project activities will include research into AI algorithm development, liaison with external partners and presentation of work at project meetings, technical sessions, and scientific meetings.
Continue with Facebook
