Projects
SpIDAI
Design of an intelligent tool for semantic segmentation and sketch-photo matching of spider genitalia to aid taxonomic identification
Biodiversity loss is occurring at an unprecedented rate, making it urgent to improve species identification and data retrieval from institutional collections. Spiders, key bioindicators of ecosystem health, present significant taxonomic challenges due to their reliance on complex genital morphology, traditionally documented through 2D hand-sketches that are difficult for non-experts to interpret. While museums house essential historical and contemporary collections, maintaining databases with up-to-date taxonomy is hindered by a shortage of specialized expertise and the time-consuming nature of manual verification.
To address these bottlenecks, this project propose to develop an AI-driven tool for the morphological analysis and identification of spiders, bridging the gap between natural history collections and cutting-edge technology. Focusing on ecologically significant African taxa as a case study, the project will generate thousands of high-resolution, focus-stacked composite images using specialized motorized microscopy and an ad-hoc hardware tool designed to standardize specimen and genitalia positioning.
The success of this ambitious program is anchored in a robust and highly complementary partnership between the Royal Museum for Central Africa (MRAC) and the University of Mons (UMONS). This consortium forms a cohesive team where domain expertise in spider taxonomy and collection management flows seamlessly into high-level technical innovation. The latest advances in computer vision will be leveraged to develop 2D and 3D semantic segmentation models using state-of-the-art architectures such as Convolutional Neural Networks, Vision Transformers, and U-Net models. This approach enables a "query-by-example" paradigm, using machine learning to match specimen photographs with vast archives of taxonomic literature. To overcome specific optical challenges like semitransparency in microphotographs, the project employs extensive data augmentation and photorealistic image synthesis. The partnership brings together specialized skills ranging from spider taxonomy and imaging to AI model development, 3D data processing, and digital tool deployment. This interdependency ensures that outputs from the MRAC (specimen imaging) flow directly into the technical work packages at UMONS (AI training).
By streamlining identification workflows, this collaboration not only enhances museum digitization but also empowers taxonomic revisions and conservation efforts. The resulting database of high-resolution images and metadata will support the global arachnological community, offering a methodology that can be extended to other arthropod groups and elevating the precision of natural history collections worldwide.
Principal investigator:
Dates:
2025 2030Museum staff:
External collaborators:
M. DUPONT Stéphane - Université de Mons (UMONS)Financed Belgian partner