Study on the precision evaluation method for a specific category in the classification of remote sensing image

The Kappa value mainly reflects the overall classification accuracy, but it is hard to evaluate whether a specific classification is optimal or not. In earthquake forecast, an indicator, R-value, is widely used to evaluate the predicting effect of a forecasting method. This paper introduces the R-value to evaluate the classifying accuracy of a specific category in image classification. The range and implication of R-value for a specific category are described and compared. As an example, analyses are performed to determine the R-value of classified building damage grade extracted from RS image, compared with building damage grade determined according to ground survey in order to get the best corresponding relationship between them. The results demonstrate that the best corresponding scheme with maximum R-values is consistent with the subjective understanding. It indicates that the R-value method gives richer classification accuracy assessment information than the Kappa method. It is expected to be applied to the evaluation of classifying accuracy of a particular category and the determination of the best scheme in image classification.