Deep Learning Techniques in Leaf Image Segmentation and Leaf Species Classification: A Survey

Plants have elemental importance for all life forms. The research areas in the field of plant sciences for botanists and agriculturists include the identification of plant species, classification of weeds from crops, detection of various diseases that hamper the growth of plant, and monitoring the growth and its semantic interpretation. Trained botanists can easily identify plant species based on the leaf shape, texture, structure or arrangement of leaves, however, the recent trend in smart agriculture demands the use of intelligent systems for the same task. Last decade has seen an enormous rise in the use of deep learning in the field of automatic plant species recognition based on the leaf images. In this work, we have surveyed various state-of-the-art deep learning techniques (Convolutional Neural Networks, Mask RCNN, Recurrent Neural Networks, Generative Adversarial Networks) that have been applied in the field of leaf image segmentation (separation of leaf from the whole image) and classification of leaves into various species. This contribution will help the new researchers in the field to get a foundation on the trends being employed in deep learning for generation of synthetic leaf images, segmentation and classification of leaves into various species. Various difficulties and future scope have also been presented.