Detection and tracking of spruce seedlings in spatiospectral images
| Main Authors: | Löwbeer, Emma, Åkesson, Erik |
|---|---|
| Format: | info publication-thesis Journal |
| Bahasa: | swe |
| Terbitan: |
, 2020
|
| Subjects: | |
| Online Access: |
https://zenodo.org/record/4607191 |
Daftar Isi:
- Glana Sensors AB has developed a camera that captures spatiospectral images, which can be converted into hyperspectral images iwth multiple wavebands in the visual and near infrared parts of the spectrum. This project investigated the possibility to detect and characeterize plants directly in the sspatiospectral images, and then make the full hyperspecral image construction only on the image patches containing the relevant plants, thus reducing computational complexity and memory usage. In addition, the purpose was to automate the process in order to avoid manual work. As test data, a data set with spruce seedlings aptured from an airborne drone was used. Four methods were evaluated for detection of spruce seedlings; manual detection, detection based on segmentation, detection using SVMs, and detection using neural networks. Manual detection and neural networks yielded the best results. Detected seedlings were tracked across the spatiospectral image set, honomgraphies between the images created, and hyperspectral image patches created. Finally, the patches were segemented and representative spectral signatures extracted from each seedling.