Are remotely sensed image classification techniques improving ? Results of a long term trend analysis

Abstract

The long term trend in the accuracy of remotely sensed image classification has been investigated using reported results in the journal Photogrammetric Engineering and Remote Sensing in the period since 1989. The results indicate no significant improvement in the performance of classification methodologies over this period. Average classification performance across all results was found to be 72.7% with the average Kappa value being 0.64. Results also indicate no significant correlation between classification performance and number of classes. A good correlation is found between overall percentage accuracy figures and the Kappa coefficient indicating the suitability of either to categorize overall mapping performance. Only a small percentage of papers (8%) were found to provide all background information necessary to make a sophisticated inter-comparison of methods.