Amplography by means of machine vision

Ampelography as the field of grapevine discrimination is currently done by experts which makes it expensive, time-consuming and exhaustive task whereas simultaneous consideration of several features by experts increases the probability of misclassification. The identification procedure can be shortened to some fractions of a second if identifiable features are interpreted to machine vision routines. In this study, amplographic features of mature leaves were coded for a machine vision algorithm to facilitate the identification process. Amplographic features of leaves introduced in IBPGR descriptors were coded in Matlab including: general shape of petiole sinus, shape of upper lateral sinus, number of lobes, shape of teeth, leaf area and anthocyanin color of main veins. Consecutive machine vision procedures toward defining each of the mentioned features are depicted in this paper. To combine all of the features simultaneously in identification process, MLP neural networks were exploited. Test set data was selected from 70 leaves of 14 cultivars. Samples were randomly picked from mature leaves of the first one third of the branch length and scanned with a flatbed color scanner. Identification error was considered as the number of cultivars wrongly classified to the total number of test set samples. Overall identification accuracy of 92.37% was gained through the test set data. This accuracy was due to exact coding of the features rather than any estimation or simulation. Misclassification of the samples was mainly referred to environmental changes of the leaves or defects. Seven amplographic features of grapevine leaves were interpreted to machine vision routines. Evaluation of the proposed algorithm using test set data proved that instantaneous identification of grapevine cultivars was feasible. Achieving the identification accuracy of 92.37% was ascribed to exact definition of IBPGR features to machine vision routines