Monitoring populations of winged insects is a crucial step in vector control and biodiversity evaluation. Identifying these insects with conventional methods can be time-consuming, labour-intensive, and expensive. A computer vision-based system that uses wing interferential patterns (WIPs) could solve this problem. Existing datasets comprising close to 150 species and 5,000 images are used to train, validate, and test models. This study has contributed to the development of deep learning architectures with low computational requirements and short inference times, as well as an interface that allows users to select different model architectures for prediction. This approach enables the automatic identification of a large number of winged insect species, including cryptic species, with high accuracy (over 90% depending on the selected subset). Previous work has shown promising results on Diptera vector families, including Culicidae, Glossinidae and Psychodidae. Ongoing work is focused on extending the method to the Hemiptera and Hymenoptera orders.
- WIP-based Insect Recognition with Light Deep Networks
- Rudricile Tomouo Djiogue - Equipes Traitement de l'Information et SystèmesFrederic Bartumeus - Centre d'Estudis Avançats de BlanesJoan Garirga - Centre d'Estudis Avançats de BlanesGuillaume Renton - Equipes Traitement de l'Information et SystèmesAymeric Histace - Equipes Traitement de l'Information et SystèmesNicolas Sauvion - Université de Montpellier, PHIM - Plant Health Institute of MontpellierCamille Simon Chane - Equipes Traitement de l'Information et SystèmesDenis Sereno - Université de Montpellier, INTERTRYP - Intéractions hôtes-vecteurs-parasites-environnement maladies tropicales
- InsectAI Annual Meeting 2026 (Niš, Serbia, 22/04/2026–23/04/2026)
- INRAE
- 99231213109311
- INTERTRYP - Intéractions hôtes-vecteurs-parasites-environnement maladies tropicales; PHIM - Plant Health Institute of Montpellier
- English
- Conference proceeding
- hal-05607883