Abstract
This research explores the use of deep learning to automate the identification of Collembola species, small soil arthropods that serve as bioindicators of soil health. Given their ecological importance and sensitivity to environmental changes, Collembola are valuable for assessing soil conditions, but their manual identification is labor-intensive and requires expert knowledge. The thesis develops a deep learning model to analyze images of Collembola specimens prepared on microscope slides.The study benchmarks state-of-the-art deep learning architectures, achieving accuracy levels comparable to expert taxonomists. It also addresses challenges related to data biases and model generalization across different ecological contexts. The findings highlight the potential of deep learning to streamline ecological monitoring, enabling large-scale and efficient assessments of soil biodiversity. This work provides a scalable tool for soil health evaluation, with applications in agriculture, environmental management, and biodiversity conservation.