Gave an oral presentation at the conference of

International Conference on Unmanned aerial systems in Geomatics, 2021

• Built CNN-based, modified UNet model, and trained using transfer learning on Geo-spatial domains of Austin, Chicago, Kitsap, and Vienna
• Benchmarked & analyzed outcomes generated by other baseline models: ResNet, VGGNet, SegNet & UNet on a similar scale of parameters
• Obtained Accuracy: 98%, F1-score: 93% by incorporating morphological techniques like Erosion & Dilation for an improved performance