A. Bari (Italy), A. Martin (Spain), B. Boulouha (Morocco), J.L. Gonzalez-Andujar, D. Barra
Olive modelling, feature extraction, fractals, moments,neural network, radial basis function
As a result of multi-location breeding and crop selection, hundreds of olive cultivars are grown in different areas of the Mediterranean region. These cultivars vary considerably, not only in their oil content but also in their morphological characteristics, including those of the endocarp. It is possible, based on the features of the stone alone, to identify a cultivar by name within each of these olive-growing areas, as the stone encompasses most of the highly discriminatory features for olives. The difficult task for a non-expert, however, is to objectively determine these features. To assist in this process, features were first extracted from the images of olive stones and then classified using a neural network classifier approach. This combination of image feature extraction and neural network approach was tested in the identification of five cultivars from the west Mediterranean region and compared to that achieved by a statistical classifier (k means clustering) and conventional descriptors. Image feature extraction in combination with neural networks achieved 96% correct cultivar classification, while the statistical classifier in combination with feature extraction achieved 60% correct classification, and the conventional descriptors varied from 70% (k means) to 90% (neural network) correct classification.
Important Links:
Go Back
