Safe Mush: Edible Mushroom Identifier using Image Classification

Abstract

"Mushrooms are delicious food, rich with nutrients such as various vitamins, minerals, rich with protein, and good for the immunity system. But not all mushrooms are good for human consumption since some of them are highly poisonous, leading to serious health problems including death. Mushrooms in the market are good for consuming, but people often eat wild mushrooms which grow naturally in their home gardens, or nearby forests. It has also become a tradition and as well a hobby to some people to go mushroom hunting. These wild mushrooms can be poisonous. It is difficult to identify poisonous mushrooms except for an expert by looking at physical characteristics, since there is no way to test them at home. Existing methods are not effective, leading to a number of mushroom poisoning cases. With the research of edible mushroom classification, the author finds a solution, using Image classification and deep learning techniques to identify toxic mushrooms at home, before consuming them. The author works on creating a machine learning model to accurately classify edible & non-edible mushrooms using an image of it, to mainly serve the general public and people who are passionate about mushroom hunting. In the research, the author added 8 custom CNN layers, which consist of batch normalizations, activation, ReLu, global average pooling layers, and dense layers on top of Mobile Net to achieve better performances for this task. Additionally, dropout layers were added and fine-tuned for generalization. The model was trained with 9000 of mushroom images, and using data augmentation techniques. Throughout the report, the author critically discusses the problem domain, requirement gathering using various techniques such as conducting surveys and interviews, design diagrams used for the implementation, and selection of technologies for the implementation. In this research, the author was able to achieve an accuracy of 80% and has a recall value of 90% for the poisonous class, making the product reliable."