Time Series Directional Change Vector Analysis

Detecting various types of changes in dense Satellite Image Time Series (SITS) presents a complex challenge. While Change Vector Analysis (CVA) is widely used for Change Detection (CD), it presents limitations due to a lack of prior information on changes, such as optimal spectral channels and change timing. To overcome these obstacles, the study focuses on the direction analysis of the Time Series Change Vectors (TSCV) [1], built upon CVA principles. Conducting unsupervised CD using time series magnitude information, the approach leverages multiple change dimensions in the direction analysis. A novel scheme formulates a representative change matrix within SITS temporal and spectral domains, guiding change representations and allowing segregation based on significance in both spectral and temporal dimensions. The proposed method efficacy is evaluated using Sentinel-2 time series data, with results affirming its robustness in effectively addressing multi-CD challenges within dense SITS.