This repository contains the third Project of the course "Geospatial Data Analysis and Processing" in the MSc "Data Science & Machine Learning" program. This exercise focuses on designing and implementing a time series classification methodology for multispectral satellite data using Transformer-based architectures.
- Familiarization with Time Series Remote Sensing Data
- Understand and work with time series data from Sentinel-2 satellites.
- Model Design and Implementation
- Design a time series classification model using a Pixel Set Encoder and Transformer Encoder architecture.
- Study and implement parts of scientific publications related to Transformer models for time series classification.
- Data Preprocessing and Augmentation
- Preprocess and normalize data.
- Implement data feeding algorithms with random sampling techniques.
- Model Training and Evaluation
- Train and evaluate the classification model using various metrics and strategies.
- Implement a k-fold cross-validation strategy to assess model performance.