Efficient Classification of Pollen Grains Using Computational Intelligence Approach

Abstract

Pollen grains play an important role in classification system of plants. Pollen morphological characters are very useful in plant classification and identification of plants. Pollen studies are widely used in allergy and epidemiology research, fossil fuel exploration, forensic science, food, pharmaceutical, and cosmetic industry, biotechnology, and so many other fields. Hence classification of pollen grains is a challenge that can be effectively done by computational intelligence approach. The traditional method of pollen classification analyses the pollen morphological characters using microscopy. This procedure is tedious and requires experts from the field of palynology. With a view to extract features from pollen images, a new classification strategy is developed which proposes Image Histogram coefficients in addition to image statistics and shape descriptor. The suitability of classifiers based on Multilayer Perceptron (MLP) Neural Network and Support Vector Machine (SVM) is explored with the optimization of their respective parameters in view of reduction in time as well as space complexity. Performance of two classifiers has been compared with respect to MSE, NMSE, and Classification accuracy. The Average Classification Accuracy of MLP comprising of two hidden layers is found to be superior (95 % on Cross Validation dataset) to SVM based classifier. Finally, optimal Classification strategy has been developed, which could be easily modified to classify more than 10 species. The proposed strategy will provide an effective alternative to traditional method of pollen image analysis for plant taxonomy and species identification, which is obvious from the cross-validation performance on the dataset containing ten different plant species.