Comprehensive dataset from high resolution UAV land cover mapping of diverse natural environments in Serbia

This dataset comprises processed outputs from an unmanned aerial vehicle (UAV) image acquisition campaign conducted across 27 study sites in Serbia. Each site is organized in a separate folder, labeled by study site name, and includes the following output data for both Object-Based Image Analysis (OBIA) and Convolutional Neural Network (CNN) approaches.

S.No

Data Alias

File Type

Description

1

name_of_the_study site_multiband (OBIA)

.tif

Five band raster orthomosaic containing RGB, DSM and NDVI layers rescaled from (0-255).

2

Vectorized_r3 (OBIA+LSMS)

.shp

The output from the segmentation process and is a basis of preparation for data labeling.

3

train_val_set (OBIA+RF)

.shp

Contain labeled samples of land use classes for training and validation sets for respective study site.

4

ClassifiedVector (OBIA+RF)

.shp

Classified vectorized output in rectangle shape.

5

ClassifiedVector_fixed (OBIA+RF)

.shp

Final output of the classified orthomosaic in a vector file containing all the classes in the attribute table.

6

RandomForest (OBIA)

.txt

Represents trained model.

7

confusion_matrix (OBIA+RF)

.csv

Confusion matrix for each study site

8

number_of_polygons (OBIA+RF)

.csv

Contains the number of polygons marked for training and validation.

9

class_area_percentage (OBIA+RF)

.csv

Refers to percentage coverage of each class for a given study site.

10

name_of_the_study_site_result (OBIA)

.png

Image showing the evaluation metric values for each site.

11

train_val_set_CNN

.geojson

Bounding box labeling dataset used for CNN model training for each site.

12

train_parameter (CNN)

.csv

Hyperparameters used for training the CNN model.

13

CNN_models

.h5

CNN models for each site trained with defined hyperparameters.

14

CNN_confusion_matrix

.png

Confusion matrix for each study site.

15

Classified_rasters_CNN

.tif

Classified rasters through CNN model, both reclassified and colormapped.