The large-scale digitization of herbarium specimens—over 10 million in France via the Recolnat portal led by MNHN, and hundreds of millions worldwide—offers unprecedented opportunities for advancing plant biodiversity research. However, extracting reliable morphological traits from these images remains challenging due to the presence of non-vegetal elements (e.g., labels, rulers, envelopes) and visual background noise that can mislead model predictions. In this work, we introduce a comparative study of three deep learning models—YOLOv8, ViT, and ResNet101—applied to botanical trait classification from digitized herbarium images. We evaluate these models on both raw images and segmented versions where non-plant elements are removed. Although models trained on raw images can sometimes yield high accuracy, interpretability analysis (e.g., Grad-CAM) reveal that their predictions often rely on irrelevant background features. In contrast, segmentation consistently drives the models to focus on the plant itself, leading not only to improved performance for several morphological traits, but also laying the groundwork for explainable and scientifically meaningful AI. We argue that segmentation is not merely a preprocessing step but a prerequisite for trustworthy trait recognition and a necessary condition for enabling AI-driven discovery in botanical sciences. This work establishes a methodological foundation for interpretable plant-focused deep learning applied to biodiversity research.

There are a lot of data about biodiversity stored in different database models and most of them are relational. Recent research shows the importance of links and network analysis to discover knowledge in existing data. However, the relational model was not designed to address problems in which the links between data have the same importance as the data -- a common scenario in the biodiversity area. Moreover, the Linked Data and Semantic Web efforts empowered the fast growth of open knowledge repositories on the web, mainly in the RDF (Resource Description Framework) graph model. The flexible graph database model contrasts with the rigid relational model and is also suitable for data analysis focusing on links and the network topology, e.g., a connected component analysis. Our research is inspired by the data OLAP (OnLine Analytical Processing) approach of creating a special database designed for data analysis, a network-driven data analysis using graph databases, in our case. Beyond an initial ETL (Extract, Transform and Load) approach, we are facing the challenge of migrating the data from the relational to the graph database, managing a dynamic coexistence and evolution of both, not supported by related work. This work is motivated by a joint research involving network-driven data analysis over the FishBase global information system. We present a novel approach to analyzing the connections among thousands of identification keys and species and to linking local data to third party knowledge bases on the web.

Several projects aim at gathering together data concerning life around the world, in order to systematize them and produce a big, unified tree of life. Rather than a static single picture of the living world, this kind of tree: (i) is a result of a dynamic interaction among several models produced by biologists for describing life and expressing how life changes and evolves as time goes by, (ii) is not unique, since there are different competing perspectives describing life (morphology, behavior, ecology, genetics etc.) and different methods of reconstructing evolutionary trees. Our work addresses these problems by proposing a "superimposed metamodel" mechanism, which acts as a modeling skeleton, supporting a unified view and articulation of models/ontologies involved in tasks that start at collecting data from the field towards producing descriptions and evolutionary trees. It enables to externalize specific knowledge as ontologies and to trace the entire rationale from one extreme of the process to the other one. This paper shows practical experiments in which we explore such characteristics as: guiding the expression of evolutionary hypotheses from observational data, going backwards on the provenance path, or evaluating changes of the tree in front of new evidences collected in the field.

MOTIVATION: Computer Aided Identification systems provide users with the resources to relate morpho-anatomic observations with taxa names and to subsequently access other knowledge about the organisms. They have the ability to manage descriptive data and make identifications through interactive keys. They are essential for both authors and users of biodiversity information. Xper(2) version 2.0 is one of the most user-friendly tools in its category and provides a complete environment dedicated to taxonomic management. AVAILABILITY: Xper(2) software can be freely downloaded at http://lis-upmc.snv.jussieu.fr/lis/?q=en/resources/softwares/xper2