Automated mosquito classification in different image resolutions
Problem Statement:
Automated mosquito identification using AI is highly dependent on image quality, yet the effect of image resolution on identification performance across hierarchical taxonomic levels remains poorly understood. As image resolution decreases, critical morphological features may become indistinguishable, potentially limiting reliable classification and reducing model interpretability. There is therefore a need to systematically determine how resolution affects AI-based mosquito identification, identify the morphological characters that are lost or become unreliable at lower resolutions, and establish minimum imaging requirements and resolution thresholds for accurate and practical automated identification.
Task:
This assignment aims to answer the following research questions:
1. How does image resolution affect automated mosquito classification across hierarchical taxonomic levels?
2. Which morphological characters become limiting as image resolution decreases? (supported by Grad-CAM and expert interpretation)
3. What are the minimum imaging requirements for reliable AI-based mosquito identification? (resolution thresholds and practical recommendations)
Work:
10% Theory, 60% Modelling, Coding and Testing, 10% Evaluation and Validation, 20% Writing
Contact:
Andreas Kamilaris: a.kamilaris@utwente.nl