Automated Feature-Specific Tree Species Identification from Natural\n Images using Deep Semi-Supervised Learning

Prior work on plant species classification predominantly focuses on building\nmodels from isolated plant attributes. Hence, there is a need for tools that\ncan assist in species identification in the natural world. We present a novel\nand robust two-fold approach capable of identifying trees in a real-world\nnatural setting. Further, we leverage unlabelled data through deep\nsemi-supervised learning and demonstrate superior performance to supervised\nlearning. Our single-GPU implementation for feature recognition uses minimal\nannotated data and achieves accuracies of 93.96% and 93.11% for leaves and\nbark, respectively. Further, we extract feature-specific datasets of 50 species\nby employing this technique. Finally, our semi-supervised species\nclassification method attains 94.04% top-5 accuracy for leaves and 83.04% top-5\naccuracy for bark.\n