Artificial Keys for Botanical Identification using a MultilayerPerceptron Neural Network (MLP)

  • Authors:

  • Jonthan Y. Clark;Kevin Warwick

  • Affiliations:

  • Department of Cybernetics, P.O. Box 225, University of Reading, Whiteknights, Reading, RG6 6AY, UK (E-mail: cybjyc@cyber.reading.ac.uk);Department of Cybernetics, P.O. Box 225, University of Reading, Whiteknights, Reading, RG6 6AY, UK (E-mail: cybjyc@cyber.reading.ac.uk)

  • Venue:
  • Artificial Intelligence Review
  • Year:
  • 1998

Quantified Score

Hi-index 0.01

Visualization

Abstract

In this paper, practical generation ofidentification keys for biological taxa using amultilayer perceptron neural network is described.Unlike conventional expert systems, this method doesnot require an expert for key generation, but ismerely based on recordings of observed characterstates. Like a human taxonomist, its judgement isbased on experience, and it is therefore capable ofgeneralized identification of taxa. An initial studyinvolving identification of three species of Iriswith greater than 90% confidence is presented here.In addition, the horticulturally significant genusLithops (Aizoaceae/Mesembryanthemaceae),popular with enthusiasts of succulent plants, is usedas a more practical example, because of the difficultyof generation of a conventional key to species, andthe existence of a relatively recent monograph. It isdemonstrated that such an Artificial Neural NetworkKey (ANNKEY) can identify more than half (52.9%) ofthe species in this genus, after training withrepresentative data, even though data for onecharacter is completely missing.