Doubly segmented proxy images for multi-scale landscape ecology a…: Ingenta Connect

Keywords: Change detection, Clustering algorithms, Geographic information systems, Image analysis, Image compression, Land cover classification, Remote sensing. Multi-band remotely sensed image data contain information on landscape pattern and temporal changes that are greatly underutilized in this technological era when monitoring of disturbance and ecological dynamics is increasingly important to address questions regarding sustainability of ecosystem health and climate change. Among the reasons for this loss of analytical opportunity are the inadequacy of methods for systematic extraction of pattern elements, incongruity between information paradigms for remote sensing and geographic information systems (GIS), and the sheer volume of remotely sensed image data when acquired regularly over time. Long-term cooperative landscape ecological investigations concerning habitat and change detection in conjunction with remote sensing and GIS have yielded a pattern-based approach to progressively segmenting images (PSI) that culminates in a doubly segmented image representation by sets of approximating signal vectors that serve as parsimonious proxies for pixel vectors. The coarser level of segmentation is entirely congruent with raster map structures for GIS, and yet mimics the appearance of an image display by colorization using information on typical spectral properties of segments contained in attribute tables. The components of the coarser representation as spatial segments constitute explicit elements of pattern at several levels. The explicit nature of these pattern elements enables spatial pattern matching for change detection that resolves difficulties with phenological variability and continuity of sensor configurations over time. Conversion to segmented representation can be applied to multi-temporal change indices so as to elicit longer-term patterns of change from temporal sequences of images. The finer level of segmentation for spectral detail enables restoration of image bands in the manner of a low-pass filter for analysis according to the usual paradigms of remote sensing. Mapping of the residuals for the finer detail of image approximation provides further information on exceptional features of landscape ecological pattern. Source.

Яндекс.Метрика Рейтинг Free Web Counter
page counter
Last Modified: April 24, 2016 @ 2:07 pm