Research On Change Detection Of Specific Targets From High Spatial Resolution Panchromatic Remote Sensing Imagery

Change detection from remotely-sensed imagery is of an important application trend for remote sensing applications. With the increasing improvements in spatial resolution and temporal resolution, the study on change detection of ground covering emphasizes both on traditional fields for large scale environment resources study, and on emerging application fields for more detail changes at regional areas.It is traditionally a mental-work-centralized application that obtaining changed information of specified targets after an interval of time or a special event from multi-temporal remote-sensing imagery. Nowadays, along with more and more remote-sensing sources available, a lot of difficulties also arise in the analysis, comparison, and interpretation of specified targets with remotely-sensed imagery covering the same region due to factors as: It is a time-consuming and iterative repetition work on interpreting images obtained at different time. The results from images depend largely on the experience of the experts. And the experience is also difficult to be passed and handed over one another. Thus it is necessary to find automatic methods on image interpretation for change detection, which feeding the needs for rapid and timely change detection and updating of specified targets.This dissertation aims to detect change information of specified targets based on multi-temporal high spatial resolution panchromatic remote-sensing imagery. And some change detection methods are studied by use of the spatial and temporal characteristics with multi-temporal imagery theoretically and experimentally. The results are drawn as following:(1) A general modal for pixel-based change detection is supplied firstly. Some pixel-based change detection methods for specified targets with their suppression on environmental effect are discussed. Such methods as linear regression, local linear regression, modified principal component analysis, quadratic equation regression, and histogram match method are compared and experimented with specified artificial targets from a pair of remote-sensing images of a refinery. The experiment results show the advantages and disadvantages of these methods on detecting change information of specified targets through high spatial panchromatic remote-sensing imagery.(2) A feasible change detection method by use of region feature is provided, which works better on suppression of pixel value error than pixel-based change detection methods. It studies the formation of determinant function for region changes, generalization of region characteristics, joint change determinants of multi-factors for regions, minimization of the influence by pixel value variance due to different acquisition time. The method can not only weaken the false change through pixel value variance, but also suitable for detecting changes of varying region sizes.(3)A "two-step image correlation" change detection method is offered. The experimental results prove that the method can restrain extrinsic phenomena from local alignment error of several pixels and obtain the change-rates about the feature points, which is necessary for timely updating of the images with specified targets.