Verification of a Deep Learning-Based Tree Species Identification Model Using Images of Broadleaf and Coniferous Tree Leaves

The objective of this study was to verify the accuracy tree species identification using deep learning with leaf images broadleaf and coniferous trees in outdoor photographs. For each 12 eight species, we acquired 300 photographs leaves used those produce 72,000 256 × 256-pixel images. We Caffe as framework AlexNet GoogLeNet algorithms. constructed four models that combined two patterns: one for individual classification 20 other two-group (broadleaf vs. trees), without data augmentation, respectively. performance proposed model evaluated according MCC F-score. Both exhibited very high all patterns; highest 0.997 augmentation. higher when trained only; trees, both types simultaneously than it only.