Efficient Change Map Detection from Imagery Data using Machine Learning Approach

Abstract

Change detection in Newly Constructed Areas (NCA) is the first step in the development of urban areas. In this field, remote sensing and deep learning are more efficient compared to other technologies. The process consists of analyzing multi-temporal satellite images between different time-stamps and automatic analysis of different graphs which is the change data. The difference calculated from the images is formed by the pixel-by-pixel subtraction of two satellite images which uses eigenvectors that are extracted for the difference image using Principle component analysis. Also, the pixel’s neighborhood is projected onto these vectors to arrive at the feature vector. Upon clustering the feature vectors into 2 clusters, we have changed an unchanged class, and each pixel belongs to one of these two clusters using which a change map is generated.