Understanding insect-habitat interactions from image-based data using computer vision and deep learning techniques

This thesis integrates insect ecology with computer vision and machine learning to analyze insect image backgrounds and their habitats. Using over 250,000 images from the Atlas of Living Australia, we developed a deep learning model to classify image backgrounds as natural or anthropogenic, introducing a "naturalness score" to quantify habitat use. Our findings highlight biases in insect photography, such as nocturnal moths photographed on artificial surfaces and aesthetically pleasing insects in natural settings, affecting habitat representation. By addressing these biases, the research provides new insights into insect-environment interactions, offering valuable tools for ecologists and conservationists to improve biodiversity monitoring and management.