University of Twente Student Theses
Counting wildebeest from space using deep learning
Wu, Zijing (2021) Counting wildebeest from space using deep learning.
|
PDF
6MB |
| Abstract: | Accurate, reliable, and up-to-date information on wildlife populations is crucial for species conservation planning in the face of unprecedented biodiversity loss worldwide. Deep learning techniques combined with high-resolution satellite images have proven successful in detecting medium- and large-sized animals. However, to date, no study has shown that this method can be used to detect and count animals with indistinct features (2-4 pixels in length) and low contrast to the background in satellite images across landscapes. In this study, I tested the ability of the U-Net deep convolutional neural network for detecting and counting the migrating wildebeest in the Mara Triangle in East Africa from GeoEye-1 satellite images. I also assessed the role of the near-infrared band in the accuracy of wildebeest detection. Moreover, I tested the model on a different area with varied landscapes and a temporally different satellite image to evaluate its transferability over space and time. The results showed that the U-Net model can be used to accurately detect and count a large number of wildebeest (more than 100,000 individuals) from the GeoEye-1 satellite image, with a high generalization accuracy (F1-score) of 0.87. Adding the near-infrared band to the RGB band combinations in the satellite image did not significantly improve the accuracy of wildebeest detection. In addition, the model was able to rapidly detect the animal clusters on the spatially and temporally different satellite images, suggesting that the U-Net wildebeest detection model has the potential to be applied to the entire Serengeti-Mara ecosystem. In conclusion, this study demonstrates an effective and efficient U-Net deep learning model for accurate and rapid wildebeest detection and counting from GeoEye-1 satellite imagery. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/88785 |
| 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