MENADD - Land Cover Classification using Multispectral Sentinel-2 Satellite Imagery, Google Earth Engine, and TensorFlow

In my talk, I shared valuable insights and practical knowledge about remote sensing and machine learning. We started with an introduction to getting started with remote sensing, discussing the significance of remote sensing data and its diverse applications. I explained the process of creating a land cover dataset, covering data collection methods and preprocessing techniques for accurate results. I then guided the audience through the process of training machine learning models using two different approaches. First, we explored training a pixel-based dataset, understanding the intricacies of handling individual pixels as data points and designing models that operate on this level. This approach is crucial for certain remote sensing tasks, and I highlighted its potential and limitations. Moving on, I demonstrated how to train a machine learning model on a patch-based dataset, with a particular focus on the EuroSat dataset. This approach allows us to leverage spatial context, considering groups of pixels as patches for classification tasks. The audience gained hands-on experience in working with patch-based data and understanding the advantages it brings to remote sensing applications.