With the development of researches on the classification quality of remote sensing images, researchers thought that uncertainty is the main factor that influences classification quality. This study puts forward an app...With the development of researches on the classification quality of remote sensing images, researchers thought that uncertainty is the main factor that influences classification quality. This study puts forward an approach to uncertainty repre- sentation, which is developed from two aspects： formalized description and comprehensive evaluation. First, we complete the classification using fuzzy surveillance approach, taking it as a formalized description of classification uncertainty. Then we in- troduce a hybrid entropy model for classification uncertainty evaluation, which can meet the requirement of comprehensive reflection of several uncertainties, while constructing the evaluation index from pixel scale with the full consideration of the different contribution to the error rate of each pixel. Finally, an application example will be studied to examine the new method. The result shows that the evaluation results fully reflect the classification quality, when compared with the conventional evaluation method which constructs models from unitary uncertainty and category scale.展开更多
基金Supported by the Provincial Science Research Program in Hubei Province of China (No. ETZ2007A03)
文摘With the development of researches on the classification quality of remote sensing images, researchers thought that uncertainty is the main factor that influences classification quality. This study puts forward an approach to uncertainty repre- sentation, which is developed from two aspects： formalized description and comprehensive evaluation. First, we complete the classification using fuzzy surveillance approach, taking it as a formalized description of classification uncertainty. Then we in- troduce a hybrid entropy model for classification uncertainty evaluation, which can meet the requirement of comprehensive reflection of several uncertainties, while constructing the evaluation index from pixel scale with the full consideration of the different contribution to the error rate of each pixel. Finally, an application example will be studied to examine the new method. The result shows that the evaluation results fully reflect the classification quality, when compared with the conventional evaluation method which constructs models from unitary uncertainty and category scale.
