In this reseach, we conducted two experiments to compare the performance of deep learning models ranging from MLP to GCN, including MLP, CNN, ViT, ViG, and GCN, on two popular datasets, CIFAR10 and Caltech101. The goal was to determine which model performs best for image classification based on various performance metrics. In the second experiment, we investigated the ability of GCN and GMLP to learn from easier data clusters within the same dataset to support predictions for more challenging clusters. The results demonstrate that GCN is not only stable but also achieves high performance in most cases, primarily due to its capability to effectively extract and learn spatial relationships between features, enhancing its performance on both types of datasets. In the field of image classification, the combination of various methods provides new perspectives and solutions for advancing this area.