Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Deep sequence modeling for crop classification from Sentinel-2 satellite images
In this project, we will develop a new deep learning approach to predict crop types in farmlands from satellite images. Two possible focus points are i) exploring TCNs for the sequence crop classification task, ii) exploring multi-task learning approaches to benefit from the additional label set.
This Master thesis project is a collaboration between the EcoVision Lab, Nasa Harvest and GreenTriangle. Monitoring farmland is key to predict crop production and, therefore, increase food security. Farmland inspection like validating cultivated crop types, estimating crop yield, detecting over-fertilization, and estimating the grass cutting frequency is a time-consuming, laborious and costly process today. Moreover, sending human inspectors on the ground leads to very sparse data in both time and spatial extent. It is thus important to develop automated systems that monitor crops based on remotely sensed imagery densely in space and time. This master thesis project aims to develop a system centered on deep learning to predict cultivated crop types from Sentinel-2 satellite images (which are publicly available). In the scope of this project, the student will work on state-of-the-art spatiotemporal methods 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 as audio generation, video classification, etc. One interesting research direction for this project will be to explore TCNs for the sequence crop classification task. Secondly, since the dataset which will be used for this project has ground-truth for the annual yield per crop field in addition to the crop type, the student also has the opportunity to explore multi-task learning approaches to benefit from the additional label set.
This Master thesis project is a collaboration between the EcoVision Lab, Nasa Harvest and GreenTriangle. Monitoring farmland is key to predict crop production and, therefore, increase food security. Farmland inspection like validating cultivated crop types, estimating crop yield, detecting over-fertilization, and estimating the grass cutting frequency is a time-consuming, laborious and costly process today. Moreover, sending human inspectors on the ground leads to very sparse data in both time and spatial extent. It is thus important to develop automated systems that monitor crops based on remotely sensed imagery densely in space and time. This master thesis project aims to develop a system centered on deep learning to predict cultivated crop types from Sentinel-2 satellite images (which are publicly available). In the scope of this project, the student will work on state-of-the-art spatiotemporal methods 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 as audio generation, video classification, etc. One interesting research direction for this project will be to explore TCNs for the sequence crop classification task. Secondly, since the dataset which will be used for this project has ground-truth for the annual yield per crop field in addition to the crop type, the student also has the opportunity to explore multi-task learning approaches to benefit from the additional label set.
This master thesis project aims to develop a system centered on deep learning to predict cultivated crop types in farmlands from Sentinel-2 satellite images.
This master thesis project aims to develop a system centered on deep learning to predict cultivated crop types in farmlands from Sentinel-2 satellite images.