Land-use/land-cover change detection using change-vector analysis in posterior probability space

Abstract

Land use/land cover change is an important field in global environmental change research. Remote sensing is a valuable data source from which land use/land cover change information can be extracted efficiently. A number of techniques for accomplishing change detection using satellite imagery have been formulated, applied, and evaluated, which can be generally grouped into two types. (1) Those based on spectral classification of the input data such as post-classification comparison and direct two-date classification; and (2) those based on radiometric change between different acquisition dates. The shortage of type 1 is cumulative error in image classification of an individual date. However, radiometric change approaches has a strict requirement for reliable image radiometry. In light of the above mentioned drawbacks of those two types of change detection methods, this paper presents a new method named change vector analysis in posterior probability space (CVAPS). Change-vector analysis (CVA) is one of the most successful radiometric change-based approaches. CVAPS approach incorporates post-classification comparison method and CVA approach, which is expected to inherit the advantages of two traditional methods and avoid their defects at the same time. CVAPS includes the following four steps. (1) Images in different periods are classified by certain classifier which can provide posterior probability output. Then, the posterior probability can be treated as a vector, the dimension of which is equal to the number of classes. (2) A procedure similar with CVA is employed. Compared with traditional CVA, new method analyzes the change vector in posterior probability space instead of spectral feature space. (3) A semiautomatic method, named Double-Window Flexible Pace Search (DFPS), is employed to determine the threshold of change magnitude. (4) Change category is discriminated by cosines of the change vectors. CVAPS approach was applied and validated by a case study of land use change detection in urban area of Shenzhen, China using multi-temporal TM data. Kappa coefficients of change/no-change detection and from-to types of change detection were employed for accuracy assessment. The experimental results show that CVAPS outperform than post-classification comparison method and can avoid cumulative error effectively. Besides, radiometric correction is not needed in this method compared with traditional CVA. Therefore, it is indicated that CVAPS is potentially useful in land-use/land-cover change detection.