LEAF RECOGNITION SYSTEM BASED ON MORPHOLOGICAL AND DISCRETE WAVELET TRANSFORM WITH PROBABILISTIC NEURAL NETWORK

In this study, it is aimed to classify leaf species by using morphological and wavelet features with probabilistic neural networks. The morphological features of the leaf images as eccentricity, form factor, the ratio of the secondary axis to the main axis, the ratio of the convex shell to the perimeter, rectangularity, extent and solidity is used. The mean, standard deviation, energy and entropy which are calculated by wavelet coefficients in two level sub-bands are used as features. The effects of the features used in this study on the success of classification were investigated. The radial based probabilistic neural network is used for the classification process. The results obtained from the study showed that 92.5% of the leaf species by using the morphological features were correctly classified. In case of the entropy features obtained by the wavelet transform are added to these morphological features, 97.5% of the leaves species are correctly classified. All experimental results show that the use of entropy features obtained by the discrete wavelet transform with the morphological features in the leaves species image classification provides higher accuracy classification compared with other feature combinations