The goal of grasp pattern classification is to establish the grasp type for an object to be grasped, which can be used to improve prosthetic hand control and reduce the strain on amputees. We present a Convolutional Vision Transformer with two components: Convolutional Neural Network (CNN) and Vision Transformer (ViT). Convolutional operations within CNN are skilful at extracting local features but struggle to capture global representations. In contrast, cascaded self-attention modules in visual transformers capture long-distance feature dependencies but degrade local detail. The CNN extracts learnable features, while the ViT uses an attention mechanism to categorize the learnt features. The model was trained on two household object datasets: the RGBD Object Dataset and the Hit-GPRec dataset, and has achieved global accuracy of 78.57% and 84.02%, respectively, under BOC and 99% and 99.59% under WWC. Our contribution involves incorporating a CNN module into the ViT architecture, resulting in competitive results on the datasets.