Automatisation de la classification des spécimens d’insectes provenant de boîtes entomologiques de collections muséales

Details

Supervisors

Jodogne, Sébastien

Faculty

Ecole polytechnique de Louvain

Degree label

Master [60] en sciences informatiques

Abstract

enIdentifying and classifying insects, essential for entomology and biodiversity preservation, remain labor-intensive tasks when relying on manual methods. This project aims to automate the detection and classification of insect specimens from digital images of entomological boxes from museum collections. By integrating advanced deep learning and computer vision techniques, two complementary approaches were developed: insect detection using YOLO and Faster R-CNN models, followed by accurate classification through ResNet and EfficientNet neural networks. The results demonstrate that the developed models provide high accuracy in species recognition while significantly optimizing analysis time compared to traditional methods. However, challenges persist, particularly for morphologically similar species and small specimens. This work highlights the benefits of such automation, facilitating large-scale processing of entomological collections and paving the way for new applications in biodiversity monitoring and ecological research.