Satellite Image Segmentation Using Federated Learning: A Privacy Preserving Approach
Satellite Image Segmentation Using Federated Learning: A Privacy Preserving Approach
Embedded Files
Overview:
Overview:
I have done this project under the guidance of Dr. Suja P. This project applies semantic segmentation to satellite images using a federated learning framework to address privacy concerns and reduce communication overhead. By distributing training without sharing raw data, the model preserves sensitive information while delivering strong performance. Experiments and evaluation on two datasets show that the federated model achieves superior performance compared to conventional deep learning, with loss values of 0.0503 and 0.255, highlighting its potential for privacy-preserving geospatial analysis.
Tools used:
Tools used:
Pytorch Framework
DATASET used:
DATASET used:
Two Satellite datasets are used:
The dataset consists of 803 training images. The Color mapping of the terrains is provided in Table I:
Table I: Color Mapping for Deep globe dataset
The dataset consists of 72 training images. The Color mapping of the terrains is provided in Table II:
Table II: Color Mapping for Dubai MBRSC Dataset
Page updated
Google Sites
Report abuse