Tree species identification based on venation patterns of leaf images photographed with a mobile device in the outdoors

53: 43~52, The aim of this study was to identify tree species based on venation patterns of leaf images,which were photographed with a mobile device in the outdoors. Forty leaves (10 species) collected at the Kyoto University and Kyoto Prefectural University Campus were used as samples in this study. Seven learning patterns were determined from the differences in photography methods and conditions. The venation patterns were evaluated by histograms of oriented gradients (HOG). Two decision-tree algorithms (J48, RandomForest), a lazy learning (IBk) and a neural network (MultilayerPerceptron) were used for machine-learning classification. A performance evaluation of the proposed model was performed with Matthews correlation coefficient ( MCC ) and correct answer rate. The classification accuracy for test data was verified by the 10-fold cross-validation method. Every learning pattern resulted in classification accuracy for training data; however, the classification accuracy for test data varied greatly according to the difference in learning patterns. By considering camera-to-subject distance, the angle at which subjects were photographed, and the light environment, high classification accuracy could be obtained from the leaf images, which were photographed with a mobile device.