Performance Analysis of Different Threshold Determination Techniques for Change Vector Analysis

摘要

Change vector analysis as change detection technique is an effective tool for extracting and identifying landcover change information from satellite imagery data. Threshold determination is one of most critical task in change vector analysis (CVA) which distinguishes "change" pixels and "no-change" pixels. Many threshold determination techniques for CVA such as empirical strategies, manual trial-and-error procedures or double-window flexible pace search (DFPS) semi-automatic, have been developed since last two decades. But the selection of appropriate threshold determination technique for specific study area is a very important and difficult process because overall accuracy of CVA depends upon threshold value. In this paper, different threshold determination techniques such as empirical strategies, manual trial-and-error procedures and DFPS, have been implemented to evaluate a method for CVA which could more effectively distinguishes the "change" pixels and "no-change" pixels on snow cover area. Experimental results confirm that a semiautomatic DFPS has greater potential than trail-and-error and empirical procedure to determine the specific threshold value for CVA technique that minimizes the overall change detection error probability and maximizes the overall accuracy.