Correlation Matrix

Features:

• Builds a correlation matrix to visualize the cross-correlation between bands.

• The matrix identifies groups of highly correlated bands in the dataset for potential compression of data.

• The toolbox allows manual deletion of the redundant bands.

Steps:

1.      Open a file.

2.      Select Toolboxes → Correlation Matrix (R-squared).

The calculation will start immediately and produce a correlation matrix with the histogram.

Remove Highly Correlated Bands (Optional):

1.      Press Remove Bands to launch a Table with all bands in the dataset.

2.      Select the bands with the highest values of correlation (i.e., 75-90) and click Remove.

The removal of the bands is global and will affect the original data.

3.      To see a new correlation matrix, you will need to run the Correlation Matrix toolbox again. Within the toolbox, the numbering of bands will reflect the deletion.

Additional Information:

The toolbox uses a square of a function that computes a Pearson correlation matrix.

Pearson's linear correlation coefficient is the most used linear correlation coefficient.

Values of the correlation coefficient can range from –1 to +1. A value of –1 indicates a perfect negative correlation, while a value of +1 indicates a perfect positive correlation. A value of 0 indicates no correlation between the columns. IDCube uses a square of the correlation coefficients. In this case, correlation coefficient can range from 0 to +1. A value of +1 indicates a perfect positive or negative correlation between the bands, suggesting the redundancy of the bands. A value of 0 indicates no correlation between the bands.

References:

Gibbons, J.D. Nonparametric Statistical Inference. 2nd ed. M. Dekker, 1985.

Hollander, M., and D.A. Wolfe. Nonparametric Statistical Methods. Wiley, 1973.