Urban Plants Classification Using Deep-Learning Methodology: A Case Study on a New Dataset

Plant classification requires the eye of an expert in botanics when subtle differences stem or petals differentiate between different species. Hence, accurate automatic plant might be great assistance to a person who studies agriculture, travels, explores rare This paper focuses on specific task urban plants classification. The possible practical application this work is tool which assists people, growing at home, recognize new species and provide relevant caring instructions. Because are barely covered by benchmark datasets, these cannot accurately recognized state-of-the-art pre-trained models. introduces dataset, Urban Planter, for with 1500 images categorized into 15 categories. dataset contains species, can grown home any climate (mostly desert) existing datasets. We performed extensive analysis aimed answering following research questions: (1) Does Planter enough information train deep learning models? (2) Can models successfully applied pre-training ImageNet beneficial comparison much smaller but more dataset? (3) two-step transfer further improve accuracy? report results experiments designed answer questions. In addition, we link installation code alpha version demo video web app based best evaluated model. To conclude, our contribution three-fold: introduce images; case study several networks configurations learning; believe that, extending future eatable assisting people grow food contributes achieve United Nations’ 2030 Agenda Sustainable Development.