Improved Hyperspectral Image Processing Algorithm Testing Using Synthetic Imagery and Factorial Designed Experiments

The goal of a remote-sensing system is to gather data about the geography it is imaging. In order to gain knowledge of the Earth's landscape, analysts develop postprocessing algorithms to extract information from the collected data. The algorithms are designed for a variety of application areas such as the following: the classification of various ground covers in a scene, the identification of specific targets of interest, or the detection of anomalies in an image. Traditional algorithm testing uses sets of extensively ground-truthed test images. However, the lack of well-characterized test data sets, as well as the significant cost and time issues associated with assembling the data sets, contributes to the limitations of this approach. This paper uses a synthetic-image-generation model in cooperation with a factorial-designed experiment to create a family of images with which to rigorously test the performance of hyperspectral algorithms. The factorial-designed experimental approach allowed the joint effects of the sensor's view angle, time of day, atmospheric visibility, and the size of the targets to be studied with respect to algorithm performance. A head-to-head performance comparison of the two tested spectral processing algorithms was also made. Finally, real images are processed using the algorithmic settings employed in the designed experiments to validate the approach.