Comparative Analysis of Attention-Based and Convolution-Based Approaches for SITS Classification

This repository contains code, data, and results for work replication. Based on a research compendium, all the data is available.

All the code in this work has been created using the R SITS package in version 1.5.2. To install it, use the command

install.packages("sits")

To view notebooks:

• Notebook comparing attention and convolution-based approaches in Rondonia
• Notebook comparing attention and convolution-based approaches in Petrolina

Repository organization:

• 📁 analysis ➡️ Analysis code

• 📁 data/raw ➡️ Input data for notebooks

  • 📁 data/raw/samples/RO ➡️ Rondonia training samples
  • 📁 data/raw/samples/PE ➡️ Petrolina training samples
  • 📁 data/raw/validation/RO ➡️ Rondonia validation samples
  • 📁 data/raw/validation/PE ➡️ Petrolina validation samples

• 📁 data/output ➡️ Outputs generated by the analysis

  • 📁 data/output/RO/classifications ➡️ Pixel-based classification results for Rondonia for each model
  • 📁 data/output/PE/classifications ➡️ Pixel-based classification results for Petrolina for each model
  • 📁 data/output/RO/model ➡️ Trained models for Rondonia
  • 📁 data/output/PE/model ➡️ Trained models for Petrolina
  • 📁 data/output/RO/tune ➡️ Tuning results for Rondonia
  • 📁 data/output/PE/tune ➡️ Tuning results for Petrolina
  • 📁 data/output/RO/segment ➡️ Object-based classification results for Rondonia for each model
  • 📁 data/output/PE/segment ➡️ Object-based classification results for Petrolina for each model

• 📁 docs ➡️ Site documentation