Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Mowing pattern recognition from Sentinel-2 images with deep sequence modeling
In this project, we will develop a new deep learning approach to predict mowing patterns of extensively used meadows from satellite images. One possible research direction for this project is to explore TCNs for temporal segmentation of time-series data and apply it to remote sensing sequence data.
This Master thesis project is a collaboration between the EcoVision Lab and the University of Bern. Farmland inspection like validating the cultivated crop type, detecting over-fertilization, and estimating the grass mowing frequency. is a time-consuming process and requires lots of human labor. A good understanding and mapping of the mowing frequency for all pastures and grassland at the country-scale is important for preserving biodiversity. Grassland that is mowed later in the year and less frequently leads to much higher biodiversity in general. It is thus essential for further biodiversity research (and for taking appropriate action at the Swiss national level), to accurately map mowing frequencies per plot. Due to the abrupt change in biomass and corresponding spectral properties of grassland after being mowed, the difference can be observed in multispectral satellite imagery of the Sentinel 2 sensor. This master thesis project aims at developing an automated system centered on deep learning for image sequence modeling to recognize mowing patterns of extensively used meadows from Sentinel-2 satellite images. In the scope of this project, the student will work on state-of-the-art time-series modeling approaches to capture the complex reflectance distribution of the vegetation and its evolution. Recently, temporal convolution networks (TCNs) (https://arxiv.org/pdf/1609.03499.pdf) have achieved impressive results on some sequence tasks like audio generation, video classification, etc. but they have not been explored for temporal segmentation of time-series or temporal change detection problems. An interesting research direction for this project is to explore TCNs for temporal segmentation of time-series data and apply it to remote sensing sequence data.
This Master thesis project is a collaboration between the EcoVision Lab and the University of Bern. Farmland inspection like validating the cultivated crop type, detecting over-fertilization, and estimating the grass mowing frequency. is a time-consuming process and requires lots of human labor. A good understanding and mapping of the mowing frequency for all pastures and grassland at the country-scale is important for preserving biodiversity. Grassland that is mowed later in the year and less frequently leads to much higher biodiversity in general. It is thus essential for further biodiversity research (and for taking appropriate action at the Swiss national level), to accurately map mowing frequencies per plot. Due to the abrupt change in biomass and corresponding spectral properties of grassland after being mowed, the difference can be observed in multispectral satellite imagery of the Sentinel 2 sensor. This master thesis project aims at developing an automated system centered on deep learning for image sequence modeling to recognize mowing patterns of extensively used meadows from Sentinel-2 satellite images. In the scope of this project, the student will work on state-of-the-art time-series modeling approaches to capture the complex reflectance distribution of the vegetation and its evolution. Recently, temporal convolution networks (TCNs) (https://arxiv.org/pdf/1609.03499.pdf) have achieved impressive results on some sequence tasks like audio generation, video classification, etc. but they have not been explored for temporal segmentation of time-series or temporal change detection problems. An interesting research direction for this project is to explore TCNs for temporal segmentation of time-series data and apply it to remote sensing sequence data.
This master thesis project aims at developing an automated system centered on deep learning for image sequence modeling to recognize mowing patterns of extensively used meadows from Sentinel-2 satellite images.
This master thesis project aims at developing an automated system centered on deep learning for image sequence modeling to recognize mowing patterns of extensively used meadows from Sentinel-2 satellite images.
Mehmet Ozgur Turkoglu (ozgur.turkoglu@geod.baug.ethz.ch)
Dr. Jan D. Wegner (jan.wegner@geod.baug.ethz.ch)
Mehmet Ozgur Turkoglu (ozgur.turkoglu@geod.baug.ethz.ch) Dr. Jan D. Wegner (jan.wegner@geod.baug.ethz.ch)