Identification of Onopordum pollen using the extreme learning machine, a type of artificial neural network

Abstract

Pollen grains are complex three-dimensional structures, and are identified using specific distinctive morphological characteristics. An efficient automatic system for the accurate and rapid identification of pollen grains would significantly enhance the consistency, objectivity, speed and perhaps accuracy of pollen analysis. This study describes the development and testing of an expert system for the identification of pollen grains based on their respective morphologies. The extreme learning machine (ELM) is a type of artificial neural network, and has been used for automatic pollen identification. To test the equipment and the method, pollen grains from 10 species of Onopordum (a thistle genus) from Turkey were used. In total, 30 different images were acquired for each of the 10 species studied. The images were then used to measure 11 morphological parameters; these were the colpus length, the colpus width, the equatorial axis (E), the polar axis (P), the P/E ratio, the columellae length, the echinae length, and the thicknesses of the exine, intine, nexine and tectum. Pollen recognition was performed using the ELM for the 50–50%, 70–30% and 80–20% training-test partitions of the overall dataset. The classification accuracies of these three training-test partitions of were 84.67%, 91.11% and 95.00%, respectively. Therefore, the ELM exhibited a very high success rate for identifying the pollen types considered here. The use of computer-based systems for pollen recognition has great potential in all areas of palynology for the accurate and rapid accumulation of data.