Premium
NEPER‐Weed: A Picture‐Based Expert System for Weed Identification
Author(s) -
Schulthess Urs,
Schroeder Kris,
Kamel Ahmed,
AbdElGhani AbdElGhani M.,
Hassanein ElHassanein E.,
AbdElHady Shaban Sh.,
AbdElShafi AbdElMaboud,
Ritchie Joe T.,
Ward Richard W.,
Sticklen Jon
Publication year - 1996
Publication title -
agronomy journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.752
H-Index - 131
eISSN - 1435-0645
pISSN - 0002-1962
DOI - 10.2134/agronj1996.00021962008800030010x
Subject(s) - identification (biology) , expert system , computer science , knowledge base , task (project management) , weed , artificial intelligence , object (grammar) , machine learning , engineering , agronomy , ecology , biology , systems engineering
Most current expert systems for weed identification are ruled‐based. They use text only and rely on a large number of botanical terms. In rule‐based expert systems, knowledge is not organized in a structured manner. Hence, they are difficult to create and use. This article describes the advantages of the generic task approach to building expert systems. The generic task approach is based on the assumption that certain knowledge and control structures may be common to a particular task across domains. Hence, reusable control structures, or tools, have been developed to solve problems. We developed an expert system that uses a hierarchical classification tool. Text descriptions are replaced with pictures, to minimize the use of technical terms. Hypotheses are established or ruled out on the basis of the user's choices among options presented as pictures. This approach reduces the number of characters required for weed identification and the user does not need to know technical terms. In our system, the classification of grasses is based on the morphologies of the leaf base and leaf surface. Broadleaf weed classification is based on the shapes of the cotyledon and true leaf. The system contains 51 Egyptian weeds. The hierarchical classification tool allowed for a clear separation of the knowledge from the structure in which the knowledge is organized. The object‐oriented nature of this approach simplifies adding or removing weeds. This approach can be readily applied to other domains, such as disease identification, fertilizer recommendation, or cultivar selection.