Semantic Change Detection in Multi-Temporal Remote Sensing Images Using Deep Learning

EasyChair Preprint 9122

Abstract

Semantic change detection (SCD) is derived from change detection (CD), SCD has great value in a variety of remote sensing. It provides semantic information on when, where, what objects, and what changes are occurring on the Earth’s surface. Studies have recently highlighted that SCD can be learned through a triple branch Convolutional Neural Network (CNN), which contains two temporal branches and a change detection branch, it supports finer grained and three-dimensional change analysis, However, the two temporal branches learn insufficient land-cover/land-use (LCLU) transition information. In this PhD dissertation, we will work on three problems.1) Developing semi-supervised SCD methods to alleviate the ’data-hungry’ problem; 2) Modeling the inherent mechanism in SCD task to reduce false detection and omission, thus improving the training stability; 3) Developing SCD methods for image time-series. The developed methodologies will be compared with state-of-the-art network models to test their performance.

Keyphrases: Semantic change detection, deep learning, remote sensing

@booklet{EasyChair:9122,
  author    = {Jing Zhang},
  title     = {Semantic Change Detection in Multi-Temporal Remote Sensing Images Using Deep Learning},
  howpublished = {EasyChair Preprint 9122},
  year      = {EasyChair, 2022}}