3.4.3. GCN

GCN can simultaneously mine the attribute and structural information in the topology diagram for end-to-end learning, the mainstream GNN model. GCN has more robust feature extraction capabilities than CNNs, solving the problem that CNNs cannot maintain panning invariant in non-Euclidean data, and can effectively mine complex correlation information in data, showing promising potential in classification tasks [54]. Therefore, we will use GCN as the third baseline model in this paper to explore the effect of the GNN algorithm in the food detection task, setting up two convolutional layers, the activation function as ReLU, the optimizer as Adam, with the learning rate of 0.01 and the epoch of 200.