Hyperspectral Data Processing: Algorithm Design and Analysis

2

Fundamentals of Subsample and Mixed Sample Analyses

The issues of subpixels and mixed pixels, which have been briefly discussed in Chapter 1, are crucial in hyperspectral data exploitation. Dealing with these issues is considered to be very challenging to hyperspectral image analysts, primarily due to the fact that techniques available in the traditional pure pixel-based image processing are generally not directly applicable or ineffective if they are blindly applied to hyperspectral signal and image processing. The main reason is that the pure pixel-based image analysis is usually carried out by hard (discrete) decisions in the sense that only a finite number of values are available for decision, while the subpixel and mixed pixel analyses are generally performed by soft decisions in the sense that a decision is specified by abundance fraction that is usually a real value. In other words, the relationship between a soft decision and a hard decision is similar to the relationship between an analog signal and a digital signal. With this interpretation, the process of converting a soft decision into a hard decision can be considered as an analog-to-digital (A/D) converter commonly used in communications and signal processing. This chapter reviews fundamentals of subpixel and mixed pixel analyses in hyperspectral data applications to detection and classification from a perspective of hard and soft decision-making processes. To provide a general context, the terms subsample and ...