Automated Classification and Identification of Plant Specimens of the Royal Botanic Gardens Kew Using Imaging Processing, AI and Machine Learning

Applications are invited for a PhD opportunity in machine learning, deep learning and plant specimen classification. The position will be based in Computer Science, situated on the beautiful Egham campus close to London. The project builds collaborations between Computer Science and Health Studies in Royal Holloway, University of London (RHUL), and Royal Botanic Gardens, Kew.

The biodiversity crisis is among the most urgent global threats facing humanity. Confronted by this challenge, it is critical to understand plant diversity and assess conservation priorities. Accurate species identification is essential for the safe and sustainable use of plants. However, gaps in our taxonomic knowledge, and a shortage of trained taxonomists to fill this need (the taxonomic impediment) mean that species are going extinct before they are described. With nearly 343,000 accepted species, vascular plants are particularly susceptible to the taxonomic impediment—there are too many plant species and too few plant taxonomists. Traditionally, identification of plant species in herbaria relied heavily on morphology and required time-consuming input from expert plant taxonomists. Now, innovations in machine learning allow us to accelerate species identification by using our expert-identified collection of herbarium specimens as a reference (training) dataset.

As part of the Kew Science Strategy, we aim to tackle this challenge by developing and implementing new identification tools for recognising plant species, and more specifically by designing and implementing new machine learning algorithms to accelerate specimen-naming workflows with automated identification of herbarium specimens.

In particular, the project will explore several directions, including

·        Evolving deep neural networks for plant specimen localization and classification

·        Weakly supervised learning for plant specimen classification with limited and/or inaccurately labelled instances

·        Deep learning methods for classifying unseen specimens without any training examples

The current digitised images of Kew herbarium specimens and images available from global repositories will be used for model development. The resulting methods will be extensively evaluated using real-world images/videos and case studies. The project will foster wider multi-disciplinary national, international, academic and industrial research collaborations.

The project will be supervised jointly by top researchers from Computer Science and Health Studies in RHUL, as well as top experts from Royal Botanic Gardens, Kew.

Royal Botanic Gardens, Kew supervisors

Dr Isabel Larridon (Initiative Leader Innovating Species Identification, Accelerated Taxonomy Department)

Dr Juan Viruel (Research Leader Conservation Genetics and Molecular Ecology, Ecosystem Stewardship Department)

Eligibility

Applicants should have the equivalent of a First-Class degree in Computer Science. A relevant MSc (Merit or Distinction) is desirable. Applicants should have sufficient knowledge on machine learning with programming skills in Python, MATLAB, and C++/Java. Research expertise in deep learning and image processing and relevant track records would be advantageous. 

How to apply

Applicants are encouraged to send their CVs, abstract and outline of dissertations, and publications if any to Dr Li Zhang (li.zhang@rhul.ac.uk) for an informal discussion before the online application. Formal applications must be submitted through the University online application system after discussions with the supervisor.