A NEW APPROACH TO APPLYING THE DISCOUNT RULE IN HYPERSPECTRAL SATELLITE IMAGE CLASSIFICATION

Abstract

Nowadays information fusion is a one of most important procedures in hyperspectral satellite image classifica-tion. The purpose of aggregation of information is to simplify data from different sources. A lot of known aggrega-tion methods include arithmetic averages, geometric averages, maximum values or minimum values. Combination rules are the special types of aggregation methods for data obtained from different sources. These sources provide different assessments for the same hypotheses. The requirement for establishing the independence of all sources of information is a very important question. The processing of the conflicting information and combining of conflict-ing data is a very difficult problem in classification tasks. But a lot of known combination rules yield illogical results, when bodies of evidence highly conflict with each other. Known combination rules emphasizes the agree-ment between multiple sources of information and ignore all the conflicting bodies of evidence. These rules can’t deal with significant conflict in the data. That’s why the combination of conflicting bodies of evidence is the most important issue in remote sensing. In this paper Discount rule is proposed to deal with conflicting sources of in-formation. Applying Discount rule, we can discount the sources first, and then combine the resulting basic proba-bility assignments with any known combination rule, using a discounting function. These discounting function ac-counts for the absolute reliability of the sources. Absolute reliability implies that we can make distinctions be-tween the reliability of sources of data and can express these distinctions between different sources mathematically. This procedure doesn’t change initial information. It also was considered an example, where proposed Discount rule was used for satellite image classification. Described Discount rule can be applied in forest classification, in remote searching for minerals and solution of different ecological and thematic tasks.