A leaf-based back propagation neural network for oleander (Nerium oleander L.) cultivar identification

Oleander (Nerium oleander L.) includes many cultivars differing for the combination of a high number of characters, therefore their identification is difficult and time consuming. Nomenclature is often inaccurate and not uniform and the commercialization of material under unreliable names or even without name but stating only flower colour and type is frequent. In this paper, a Backpropagation Neural Network (BPNN) based on the image analysis of oleander leaves was developed as support tool for cultivar identification. It was built using 18 morphometric and colorimetric leaf parameters of 880 leaves collected from 22 cultivars (40 leaves per cultivar). The model resulted to be an efficient, reliable, and rapid method for distinguishing genotypes. The percentage of leaves attributed to the correct class reached 97.50% considering the single cultivars, and 54.55% on the total of the analysed leaves. Twenty-one cultivars were identified with certainty, and similarities in leaf morphology between some genotypes were highlighted, too. The method requires care in the choice of the leaves, which must be healthy and well-developed, but it is objective and, being oleander an evergreen species, not season-dependent. The model could be implemented in efficiency by introducing more leaf parameters, or in speed and computer performance by selecting the most representative. In fact, a smaller BPNN based on eight selected leaf parameters resulted slightly less sensitive (45.45% of the leaves attributed to the correct cultivar) but faster (4.4 s vs 7.7 s using a standard computer), and it could be used for very numerous collections.