Hyperspectral Data Processing: Algorithm Design and Analysis

4.5 Conclusions

Many hyperspectral imaging algorithms have been designed and developed for data exploitation in the past. It seems that there is a lack of standardized data sets that can be used to objectively compare one algorithm to another for performance evaluation and analysis. In other words, if one claims his algorithm to be better than any other algorithm, without a standardized data set it will be very difficult to substantiate such a claim and validate the results. This chapter investigates this issue and further designs six scenarios that can be used as a standardized data set to simulate various scenarios. However, it should be noted that the six scenarios serve only as a purpose of how to deign synthetic images. Many other scenarios can also be simulated on the basis of the same concept such as those explained in Chapter 18. To illustrate how these scenarios can be carried out for algorithm performance analysis, three applications are included as illustrative examples, which are endmember extraction, spectral unmixing for mixed pixel classification/quantification, and subpixel target detection, each of which requires different levels of target signature knowledge.