Performance Assessment and Dataset Description

Dimensionality reduction (DR) of hyperspectral imagery (HSI) is a significant pre-processing step that increases the efficiency of the subsequent data analysis. Classification is one of the important tasks in remote sensing where hyperspectral images are used. Hence, to assess the performance of DR methods, spectrally reduced datasets are further classified, and the overall classification accuracies are measured. Other than overall accuracy, other metrics can also be used for performance evaluation of DR methods. In this chapter, these metrics are described. Two real hyperspectral datasets have been used for conducting the experiments, one of which is acquired by the airborne sensor and the other one by the space-borne sensor. Since classification is a supervised task, it requires ground truth information or the labelled samples to perform training. Hence, in this chapter, detailed descriptions of datasets and corresponding ground truth classes are provided. The same set of data is used for conducting all the experiments that are mentioned in subsequent chapters.