University of Twente Student Theses
Investigating suitable vision transformer models for wildlife camera trap data
Trusca, L.M. (2025) Investigating suitable vision transformer models for wildlife camera trap data.
|
PDF
4MB |
| Abstract: | Meadow bird species in the Netherlands have been in decline for the past few decades due to various factors, including increasing predatorial pressure on dwindling populations. In this context, this study aims to utilize vision transformer models to identify species captured in camera trap images within meadow bird habitats, facilitating targeted conservation efforts. Moreover, the aim is to investigate the performance of various vision transformer models on camera trap images, along with a better understanding for what kinds of models are best suited in this context. As a tool for explaining these models, an explainable AI technique will be used to compare DETA in night vs day settings. Consequently, the aim is to achieve a better understanding for which vision transformer models are most suited towards wildlife camera trap environments. The results showed that DETA performs best for known species detection, while OWLVIT offers greater flexibility for handling new species. Feature map analysis revealed distinct differences in how these models process day and night images, focusing on more than just the object itself. Despite demonstrating the viability of vision transformers for wildlife monitoring, limitations such as a small nighttime dataset remain. Future work could explore metadata such as time data, comparing seasonal performance and feature maps, more advanced object detection architectures, and comparisons of different model backbones to enhance performance and reliability in diverse conditions. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science BSc (56964) |
| Link to this item: | https://purl.utwente.nl/essays/105228 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page