Identification Method of Poisonous Mushroom Based on Graph Neural Network

The escalating demand for mushroom consumption in society has led to a corresponding increase in the variety of mushroom types available in the market. However, due to the diverse characteristics among different mushroom types, accurate identification solely based on visual inspection and common knowledge becomes impractical. Thus, there is a necessity for a precise and efficient identification method. This study concentrates on nine specific mushroom types, encompassing both poisonous and edible varieties, as its research subjects. The proposed approach utilizes a graph neural network for poisonous mushroom identification. Initially, the mushroom images in the dataset undergo preprocessing to construct a graph representation of the mushroom images. Subsequently, feature selection for each mushroom image node is implemented. Following this, a weight matrix for the edge and node features in mushroom images is established. Finally, a graph neural network is employed to aggregate the graph data nodes, achieving the classification of mushroom data. Experimental results demonstrate that the implementation of the graph neural network on the mushroom dataset yields an accuracy rate of $97.6 \%$. In comparison to traditional models, this method exhibits superior accuracy and proficiently extracts node features from mushroom images. The classification efficacy observed in this method addresses the challenging task of mushroom classification to a considerable extent. Significantly, the approach distinguishes between poisonous and edible mushrooms, demonstrating practical application value.