This is script for dataset on Kaggle https://www.kaggle.com/datasets/yaroslavnaychuk/satelliteimagesegmentation
Source Images Description The team of authors has constructed a new large-scale land-cover dataset with Gaofen-2 (GF-2) satellite images. This new dataset, which is named as Gaofen Image Dataset with 15 categories (GID-15), has superiorities over the existing land-cover dataset because of its large coverage, wide distribution, and high spatial resolution. The large-scale remote sensing semantic segmentation set contains 150 pixel-level annotated GF-2 images, which is labeled in 15 categories. Some of the images are from the paper: Land-Cover Classification with High-Resolution Remote Sensing Images Using Transferable Deep Models. Source Link with Dataset
The resulting images are in .tif format and 256x256 resolution obtained with the patchify library. Masks contain values from 0 to 15. Dataset use 50% of all source train images
I have tested various backbones with Unet architecture. MobileNet showed the best results. (0.44 IOU-score)
- background - 0
- industrial land - 1
- urban residential - 2
- rural residential - 3
- traffic land - 4
- paddy field - 5
- irrigated land - 6
- dry cropland - 7
- garden plot - 8
- arbor woodland - 9
- shrub land - 10
- natural grassland - 11
- artificial grassland - 12
- river - 13
- lake - 14
- pond - 15