Applying machine learning techniques for automatic classification of drone images.
-
Exploratory analysis. A initial attempt at creating a training dataset and applying a simple Random Forest classifier.
-
Refining the classification scheme. Using mosiaced images and a more comprehensive training dataset to see how much detail can be reasonably extracted using a simple random forest algorithm.
-
Initial raster processing. Combining the RGB and multispectral datasets into a single 8-band image.
-
Image classification. Training and evaluating a Random Forest model, then applying it to predict substrate classes for the full Akerøya dataset.
-
Tuning and comparing different models. Initial code for calibrating and comparing a range of supervsied classification algorithms.
