Hyperspherical direction cosine change vector analysis

Abstract

Change vector analysis (CVA) partitions spectral change into magnitude and direction components. Although recognized as an effective change detection method, CVA has two limitations. First, it is usually applied to only two bands, and secondly, the selection of a threshold for the magnitude of change can be difficult. These limitations have been addressed through recent research. Hyperspherical direction cosine (HSDC) CVA is an extension of conventional two‐band CVA to many dimensions. Training areas can be used to guide the selection of the change magnitude threshold for discriminating Change and No Change classes. This paper provides a discussion on previous uses of HSDC approaches, and develops a simple procedure for selecting the change magnitude threshold automatically. The application of HSDC CVA and conventional CVA to a Landsat multiSpectral scanner (MSS) Las Vegas, Nevada, dataset is used to demonstrate the application of the approach, and quantify the accuracy gained by incorporating multiple bands in the change analysis. Conventional CVA resulted in an overall accuracy of 69% (Kappa = 0.60), compared to 74% for HSDC CVA (Kappa = 0.68).