Chestnut (Castanea sativa Mill.) genotype identification: An artificial neural network approach

Summary

The potential use of the artificial neural networks (ANNs) for characterization and identification of seventeen chestnut (Castanea sativa Mill.) accessions, belonging to the ``marrone''-type and ``chestnut''-type, was investigated in genotypes originating from regions of Italy. Different back-propagation neural networks (BPNN) were built on the basis of image analysis parameters of the leaves, for two tasks of chestnut classification. In the first case a BPNN was built and trained to differentiate the 17 accessions of chestnut. In the second case a BPNN was conceived to distinguish between the ``marrone'' and ``chestnut'' types. BPNN produced a clear identification of all the accessions except in the case of `Garrone nero', `Garrone rosso' and `Tempuriva', which showed almost the same output diagram. Cluster analysis separated the 17 chestnut genotypes into four main groups whose differences were related to the original sources of the genotypes and to the type of affiliation (``marrone''-type or ``chestnut''-type). Artificial neural network technique was also able to discriminate between ``marrone''-type and ``chestnut''-type accessions. Qualitative and quantitative rules for the image analysis parameters, useful for classifying chestnut accessions into these two types, were obtained. On the whole the relative importance of the leaf parameters reveals that ``typical'' leaves for ``marrone''-type are more elongated, of a darker colour and with a higher perimeter/area ratio than the leaves of the ``chestnut''-type.