Using Deep Learning for Prediction of Edible and Poisonous Mushrooms

This article presents an CNN based model to predict edible and poisonous mushrooms from image data. We have used dataset — Danish Fungi 2018 (DF18) with several unique characteristics: species-level labels, a small number of errors, and rich observation metadata. An advanced CNN model U2Net was implemented to build the model for prediction. While validating on a test data, the model could predict edible mushrooms with an accuracy of 62.5% and poisonous mushrooms with 85.45% respectively. The model has been finally deployed into a real-time mobile application front-end, to increase public interest in fungi in detecting edible and poisonous mushrooms.