CHARACTERIZATION OF NON-CONVENTIONAL EDIBLE PLANTS, GUIDED BY DEEP LEARNING

Abstract

This study aims to characterize unconventional food plants (PANC) using deep learning to promote their acceptance in the Brazilian diet. One key focus is developing a mobile app that employs Artificial Intelligence (AI) to identify and characterize these plants. The project collected 1,354 images of three PANC species: yam, ora-pro-nobis, and hibiscus. The AI, trained using Convolutional Neural Networks (CNN), achieved high classification accuracy between 91.16% and 99.91%. This demonstrates the model's effectiveness in identifying PANC despite genetic variability. The app not only aids PANC identification but also provides recipes and advice on incorporating them into daily meals. The Osiris project has potential social and environmental impacts, promoting biodiversity preservation and more sustainable agriculture. It also supports local products, creating economic opportunities and reducing reliance on imports. The research highlights the importance of data volume in AI effectiveness and suggests expanding the image database to include more PANC species. Participation from botanists and researchers is vital for improving the plant characterization process. The project underscores the nutritional relevance of PANC and the need to raise awareness of their benefits. The developed app offers an accessible platform for the public to learn about and use these plants, fostering a more diverse and healthy diet. Ultimately, the Osiris project is an innovative initiative with the potential to change the perception and consumption of PANC, strengthening the connection between people and the environment.