Spatial-spectral Cross-correlation for Change Detection ---a Case Study for Citrus Coverage Change Detection

Citrus tree change detection is of great importance for citrus production (inventory) analysis and forecasting. The National Agricultural Statistics Service’s recent citrus GIS modernization effort built the GIS infrastructure for the entire State of Florida. To detect changes in citrus acreage, it is necessary to detect not only the land coverage changes between the citrus groves and other land uses, but also the changes in the existing citrus groves, such as tree removal, tree replacement, abandonment or new plantings. There are many change detection techniques applicable if proper imagery is available. However, the data available to our program is collected from different cooperating Florida State Agencies and they originate from different sensors with differences in radiometric, dynamic range, spatial resolution and spectral bands. Spectral signature correlation is one of the spectral based change detection methods, which can overcome radiometric and dynamic range differences. However, it does not solve the mixedpixel effect caused by resolution difference. In this paper, a spatial-spectral cross-correlation method is presented. This method generalized correlation coefficients defined in a spatial domain or a spectral domain into a spatialspectral domain. It has shown that either spatial or spectral correlation coefficients are a special case of spatialspectral cross-correlation coefficients. This method is spatial-spectral signature based; it is invariant to the dynamic range and robust to radiometric difference, the noise and the mixed-pixel effect; yet it is straightforward with minimum preprocessing required. Finally, experimental results will be presented as a comparison between spatialspectral correlation and spectral correlation methods.