Detecting Spatial Information from Satellite Imagery using Semantic Segmentation using UNET and Inception ResUNET Models
Source Code on Google Collab https://colab.research.google.com/drive/1zjKoO6wLT6eT8kRj1k-4H-uA53b74Z8C?usp=sharing
U-Net and Inception ResnetV2U-Net with various 10 experiemtns are applied to small dataset of Dubai satellite imagery to discover the best model results and detect spatial features from this small satellite imagery.
DataSet: Dubai Semantic Segmanation Dataset Source: https://humansintheloop.org/resources/datasets/semantic-segmentation-dataset-2/
The images are labeled in 6 classes as the following
Building class: #3C1098 >> color is Dark Purple. Lands Class (unpaved area): #8429F6 >> color is Vivd Purple Roads Class: #6EC1E4 >> color is Sky Blue Vegetation Class: #FEDD3A >> color is Golden Yellow Water Class: #E2A929 >> color is Mustard Yellow Unlabeled Class: #9B9B9B >> color is Gray