The signature file that has been specified does not have a covariance matrix. In a supervised classification, statistics are calculated for each class as identified by training samples. In an unsupervised classification, the classes are defined by statistically grouping cells into a specified number of clusters. The statistics for the classes are stored in a signature file. The signature file is then used as input to other multivariate statistics tools, generally to perform a classification. The most common reason for a covariance matrix not to be created is that, when creating the signature file, the specified training samples and input bands were insufficient.
- Open the specified signature file to see if any covariance matrices were created. If not, try re-creating the signature file.
- Check to make sure the training samples have enough cells in each class, that there is a good representation of the feature in the class, and the classes are distinct from one another.
- Also make sure the input bands are not duplicated, the bands are parametric (represent continuous values), and there are multiple bands entered into the signature file.