Implementation of OBIA algorithm for classification photo from unmanned aerial vehicle ("OBIA classification with DM.ipynb").
The input data is the aerial image in 3 channels RGB ("ortho.png") and digital model - DM ("dm.png").
DM = Digital Terrain Model (DTM) - Digital Surface Model (DSM).
Orthophoto, DTM and DSM are got from Agisoft PhotoScan software. Subtraction is implemented and labels are created in ArcMap software.
For training labeled image ("train_with_shadows.png") is used. In the example, I classified objects on the aerial image to 3 categories: trees, grassland and roads.
OBIA algorithm which is used here:
- Upload 2 images (aerial photo and DM) and sum it to one 4-bands image.
- Create segmentation (Quickshift and Felszenwalb segmentation are used).
- Upload labels and train. Ensemble "Random Forest Tree" is used.
- Save classification to the file.
You can read about results of the classification in my topic "Automated photointerpretation and vectorization of aerial survey materials for creation topographic plans.pdf" (in russian).