Abstract:
Change detection seeks to identify temporal changes in material composition within a remotely sensed scene by comparing the pixels of two images collected at different po...Show MoreMetadata
Abstract:
Change detection seeks to identify temporal changes in material composition within a remotely sensed scene by comparing the pixels of two images collected at different points in time. Anomalous change detection (ACD), in particular, emphasizes changes that are different from how other pixels might have changed. This will suppress broader but potentially uninteresting changes, e.g., seasonal or pervasive changes like snow or shadowing. ACD algorithms are typically applied to physics-based imagery such as multispectral data or synthetic aperture radar data, making use of the finer signal discrimination enabled by the multi-band nature of the imagery. Here we are interested in panchromatic imagery, which, while typically at much higher spatial resolution, comes at the cost of less signal information per pixel (only one value, a measure of brightness). This lack of signal information makes it challenging to apply traditional change detection techniques. This research explores augmenting panchromatic imagery with multiple derived texture feature layers, such as variance or entropy, to create a pseudo multi-band image. The resulting multi-band feature-augmented panchromatic images can then be exploited using traditional ACD approaches. Experiments are shown using real panchromatic imagery collected from spaceborne platforms.
Date of Conference: 29-31 March 2020
Date Added to IEEE Xplore: 18 May 2020
ISBN Information: