DEVELOPING A MUSHROOM IDENTIFICATION APPLICATION

Abstract

Mushroom identification is challenging due to visual similarities between edible and toxic varieties, often leading to misidentification and serious health risks, particularly for beginners. This thesis develops FungiScan, a beginner friendly Android mobile application that combines real time image recognition with complementary tools to promote safe foraging. The application uses a client-server architecture where a Flutter/Dart client handles user interactions, while a server-side Vision Transformer model, trained on a custom 80-class dataset of 16,000 images assembled from public sources to performs real-time classification, which achieves approximately 89% average accuracy on the test set, with per-class accuracy ranging from 65% to 100%. The application also offers keyword search via a bundled JSON dataset and an interactive forage map using the iNaturalist API. In conclusion, FungiScan provides an accessible and reliable toolset that integrates machine learning with intuitive mobile design, encouraging safe mushroom identification and self-learning among novice foragers.