CHANGE DETECTION AND EXTRACTION OFINFORMATIONINREMOTE SENSINGIMAGES

Abstract

Change Detection is a very vital task carried out on this planet and has been colossally performed andalso researched in these recent decades. It has alwaysbeen applied in infrastructure and surface monitoringtechnique, disaster management, theuse in urban dynamics and other fields as well. Current methods always have a very simple mechanism where it has always been dependent for encoding bi-temporal independent images andobjects thus obtaining and performing on their representation vectors, but it ignores the vitality of trifling-layer informationwhich contains high-resolution and fine-grained functions and features which has often led to miss the small targets.In this paper our idea is to propose a system whichis based on densely connected Siamese networkuseful forchangedetectiontechniques.Ourmethodsoothelossoflocalization informationanddatawhichis donebyintroducingthenewmodulenamedattentionmechanismwhichhasbeenapplied atthe back ofinformationtransmissionmoduleinorderto givethat sort of attentionweightandtherequired accuracytoeachtemporal image feature and the classified extraction which eventually enhance the information that is changed for the imageor object we want to predict the change for and improves final change prediction. The idea revolves around the factthat bothquantitative and visualanalyses of the experimental results show that our method improves highly onmany evaluation criteria and the proposed method also has competitiveness and higher predictive ability amongotherchangedetectionmethods.