A multimodal approach to high resolution image classification

As the collection of multiple modalities over a single region of interest becomes more common, users are provided with the capability to better overcome limitations of one data type by using the strengths of another. Often, when working only with hyperspectral imagery, scene classification is limited both by the generally lower spatial resolution of the hyperspectral imagery as well as the inability to distinguish classes which are spectrally similar, like asphalt roofing material and road asphalt. This paper will present and demonstrate a method to determine pure pixels in hyperspectral imagery by taking advantage of higher spatial resolution information available in color imagery fused with LIDAR return strength and elevation data. In return, the spectral information gained from hyperspectral imagery will then be used to perform image classification at the higher resolution of the color image. The result is a fully automated process for pure pixel determination and high resolution image classification.