Question

Divergence analysis in classification is used:

• A. to decorrelate a given set of bands used in classification
• B. to logically smooth the classified image
• C. to segregate mixed and homogeneous pixels
• D. to evaluate statistical separability amongst class pairs

Hint:

Use separability measures (Divergence, Transformed Divergence, Jeffries–Matusita, Bhattacharyya) to compare training sets before running a classifier; choose bands/classes with high separability.

Answer 1

The Correct Option is D

In supervised classification, we assess how well two classes are separated in feature space. Divergence (and its variants such as Transformed Divergence or Bhattacharyya distance) quantify the statistical separability between class pairs using class mean vectors and covariance matrices. A larger divergence indicates better separability and lower expected classification error.
(A) PCA/Decorrelation stretch is for band decorrelation, not divergence.
(B) Smoothing is a post‐classification spatial filtering step.
(C) Mixed vs homogeneous pixels are not determined by divergence directly.
Thus, divergence analysis is for evaluating separability among class pairs.
\[ \boxed{\text{(D)}} \]