ABSTRACT
In supervised remote sensing classification questions may arise about which result to choose for scientific applications of results. Confusion matrix analysis in remote sensing can be used to analyze the accuracy of supervised learning applied to image classification. The objective of the homework is to perform confusion matrix analysis to determine classification quality metrics from experiment used to evaluate different class arrangements with and without multispectral band fusion. Different distance criteria to determine pixel similarity in a hybrid classification approach using segmentation are also evaluated. Solved exercises are used to illustrate basic statistical techniques for defining evaluative sample numbers for a classification scheme. Methods for determining optimal number of clusters in unsupervised classification are presented as an alternative insight. The calculation of confusion matrix is determined step by step and by computational algorithm using R packages.
