COBECORE workshops

Friday 30 September 2022

Automatic stomata detection using deep learning: a hands-on introduction

Francis wijfels & Sofie Meeus

Measuring stomatal densities is relevant to many fields of research within plant science ranging from studying plant evolution and adaptation to optimizing water-use efficiency in crops. The From-leaf-to-label is a workflow that integrates the use of deep learning and was originally developed to deal with the huge number of photomicrographs - needed to investigate tropical tree adaptation to increasing atmospheric carbon dioxide levels - as an alternative to the time-consuming, repetitive, error-prone, unreproducible, manual counting of stomata. During this hands-on workshop Francis wyffels and Sofie Meeus will present the From-leaf-to-label workflow and will guide you through the different steps needed to automate stomata detection using deep neural networks.

No programming or prior experience with these techniques are required to attend the workshop, but bring your laptop to participate in the hands-on part of the workshop. The teaching materials also have been used in the context of secondary education via Dwengo vzw and have been awarded with the Queen Paola award for education.

Francis wyffels is a Full Professor at the Department of Electronics and Information Systems, IDLab-AIRO, Ghent University--imec and Sofie Meeus is a researcher and data steward at the Biodiversity Informatics team, Department Herbarium & Library at Meise Botanic Garden.

Computer-assisted timber identification: an hands-on introduction to available applications on macroscopic wood anatomy

Ruben De Blaere & Kévin Lievens

Wood identification is a key step in the enforcement of laws and regulations aiming at combatting illegal timber trade. It is a major concern especially for countries with species-rich forest resources. The most used, cheapest and most generally applicable method for wood identification is the anatomical assessment by trained experts. Such assessment includes the observation of features on tissues and cells on the transversal, tangential and radial plane and aims at scoring diagnostic features to characterize the botanical taxon. Traditionally, this process requires a laboratory setting to prepare microscopic thin sections, and large collections of reference material to identify a wood specimen unto species level. Some features do not require the use of laboratory equipment to observe them as they are visible with the unaided eye or a handheld macro-lens. Those ‘macroscopic’ features can be used to indicate the genus or species of the specimen, and thereby provide a cheap way to identify wood. Nowadays, modern technology can provide simplified ways in order to aid macroscopic wood identification such as digital identification keys. These are essentially decision trees, using large databases of textual descriptions on anatomical features or other distinguishing characteristics. They are consulted by giving the observed features as input and result in a list of possible genera or species. The main advantages of classification keys are their speed and flexibility, although they still require training of the user in recognition of macroscopic features. AI is another example of modern technology that can simplify identification, as learning to use a picture snapping app or device is easy in comparison to the long and difficult training on wood anatomy. Machine learning and specifically deep learning can be used to identify the botanical taxon of specimens by taking images of the wood surface and using Convolutional Neural networks to classify them.

This workshop is an introduction to modern applications on timber identification using macroscopic wood anatomy. Two approaches will be shown and available applications for at-home timber identification will be demonstrated on Central-African timbers. This workshop will explore the opportunities and pitfalls of macroscopic wood anatomical assessment and its applications. Participants can bring their own laptop and smartphone to actively identify specimens, but it is not required to follow the workshop.