The SRCA Based Change Detection In Multispectral Remote Sensing Images

Change detection is the technique of identifying change informations of the ground objects in the remote sensing images acquired on the same geographic area at different dates. With the increase of available remote sensing image data and the improvement of digital image processing technique continuously, it has been much more popular to exploit multispectral remote sensing images to monitor the change of land use, land cover, forest, water resource and mineral resource than ever before owning to its receiving wavelength ranging from visible to infrared which can be used for discriminating diverse changes more likely and reliable. Due to its importance in the environmental monitoring, resources investigating, urban planning and disaster evaluating, etc., in this paper we addressed the problem that how to detect the change informations more effectively and automatically by using the multispectral remote sensing images in three parts:(1) We designed a registration algorithm in multitemporal and multispectral remote sensing images based on the edge feature as well as the match strategy of spectral reflectance curve clusters. At first, the edge information will be extracted from the two dates’ images by applying the2D discrete wavelet transform, and using the distinct edge feature to make a overall coarse registration. After using LRC to obtain the each band’s reflectance image of two dates, split them into small parts and choose the ones which have enough edge information to search their best matching parameters in the corresponding relatively big window in the other date’s image based on the criterion of difference of spectral reflectance curve cluster minimization (DSRCCM). At last, remove the parameters whose difference of spectral reflectance curve cluster are comparatively bigger, and then make a statistics of the remaining parameters to find the main values so that the rigid transform can be applied.(2) We present a method for detecting thick clouds and their shadows in Landsat TM/ETM+images by exploiting the spectral information and context information in the images. According to the spectral distribution of the thick clouds and their shadows, we exploit the mean, variance of all bands as well as saturation with thresholds to detect the thick clouds and shadows respectively. Then we a standard pair of cloud-shadow pairs should be find so that its relative position feature could be used to match clouds with their shadows to eliminate the false targets. Finally, utilizing the spectral information and context information to do the supplementary detecting task to eliminate the missed targets. (3) We proposed a new change detection method in multispectral remote sensing images based on Spectral Reflectance Change Analysis (SRCA). Firstly, the Logarithmic Residuals Correction Model (LRC) is applied to each band of multispectral images acquired at two dates to obtain the corresponding reflectance images. Then the variance of the reflectance change vector (VRCV) and the magnitude of the reflectance change vector (MRCV) of each pixel should be calculated which can be used as the descriptors of the difference between two reflective spectra of two dates. In order to eliminate the missed alarms in the map of VRCV caused by LRC and the false alarms in the map of MRCV due to its low contrast of the two classes, the intensity information in the original images is exploited to make an enhancement for the map of VRCV and MRCV in SRCA. Finally, apply the OTSU adaptive threshold algorithm to the enhanced map of VRCV and MRCV, respectively, and use an image fusion strategy to get the change mask.