Change Vector Change Indices Method for Unsupervised Remote Sensing Change Detection

Unsupervised Remote sensing change detection is very important because it addresses the problem of scarcity in the data availability in training. Some of the researchers are successful in finding the change regions but are not able to classify what changes happened in those regions. Nearly every study faces challenges in generalizing the fluctuations occurring during both the growing and non-growing seasons. The current study utilizes a novel architecture consisting of two parts. The first part involves proposing change regions through the change vector calculation method and applying Otsu's Thresholding on the calculated change vectors, while the second part entails classifying these proposed change regions into vegetation change, building change, and non-change regions. non-change region classification process aims to eliminate false changes identified by the change vector analysis and Otsu's Thresholding. In the second stage, NDVI, NDWI, and NDMI indices are compared and the resulting change mask obtained is subjected for element-wise multiplication operation with the change mask obtained in the first stage. The study addressed the challenge of generalizing short-term changes, such as those occurring during the growing and non-growing seasons, by computing the median of pixels over a time range encompassing both types of changes. Additionally, the method was fine-tuned using NIT Puducherry data from two time periods. Subsequently, the entire Karaikal district was considered to examine the changes comprehensively. The present study outperformed different state of the art methods like CVI_OTSU, PCI_Kmeans, Continuous Change Detection and Classification (CCDC). The present study got better results with an accuracy of 86.07 and kappa coefficient of 0.7898.