Choice under imprecision: a simple possibility-theory-based technique illustrated with a problem of weed identification

Abstract

Choosing among a predefined set of alternatives is a very common problem in many domains of human activity. This paper presents a simple tool for: • - describing a collection of objects (or alternatives) along a given set of attributes in a situation where the available knowledge is pervaded by imprecision; and • - retrieving from among them those being compatible (to some extent) with some data (possibly also imprecise) gathered along the same set of attributes and concerning an unknown object to be identified. Our approach advocates the use of possibility theory, which is a relatively recent and very powerful tool for dealing with the vagueness and uncertainty of available knowledge. The proposed technique has been implemented as an autonomous module and incorporated into a microcomputer system specifically designed for the problem of weed identification in cultures located in continental France. In this paper, the weed-identification application serves the basic purpose of illustration, and the method is described in a way which preserves its generality. Background material on possibility theory and a short discussion of the differences from probability theory are provided.