An autonomous approach to generate the land use land cover classification map from the satellite images
In the current era, a large volume of remotely sensed images is accessible from various acquisition sources like satellites, manned and unnamed aerial vehicles. These images having different spectral, temporal, and spatial resolutions are useful for the generation of land use land cover classification (LULCC) maps. These maps can be utilized for precise monitoring of Earth and the ecosystem in real-time; which in turn have a vast spectrum of applications like crop monitoring, deforestation control, soil contamination, water pollution, biogeochemical cycling, thermal mapping. In this regard, there is an increasing demand for automatic techniques, as the generation of LULCC maps by field survey is expensive, time-consuming, and mostly infeasible. Due to this, it is necessary to investigate more sophisticated techniques for producing automatic LULCC maps using remotely sensed images.
This project requires the use of high spatial-resolution satellite imagery, using data from the Pleiades satellite, provided by the European Space Agency (0.5m resolution). The images are collected over the region of Cyprus, an island located in South Europe. Based on this data, the goal is to automatically identify the land use land cover classes and obtain the classification maps with high precision and recall. To accomplish the task, deep learning-based classification and semantic segmentation strategies are likely to be utilized. It is important to mention that the project will need some manual annotation of training data in order to train the models.
20% Theory, 25% Labelling/Annotations, 35% Modelling and Programming, 20% Writing
Andreas Kamilaris (firstname.lastname@example.org)