•  
  •  
 
Information Sciences Letters

Information Sciences Letters

Abstract

This study proposes an automated technique for segmenting satellite imagery using unsupervised learning. Autoencoders, a type of neural network, are employed for dimensionality reduction and feature extraction. The study evaluates different segmentation architectures and encoders and identifies the best performing combination as the DeepLabv3+ architecture with a ResNet-152 encoder. This approach achieves high performance scores across multiple metrics and can be beneficial in various fields, including agriculture, land use monitoring, and disaster response.

Share

COinS