Multi-channel information fusion network for 3D shape classification

In the field of 3D shape classification, view-based methods have attracted much research attention due to the development of deep learning. However, the existing view-based methods are generally limited in their ability to effectively aggregate multi-view features for enhanced 3D shape representation, thereby hindering further improvements in classification accuracy. To address this issue, this paper proposes a method based on fusing multi-channel information and similarity grouping. On the one hand, this paper employs illumination, depth and thickness images to represent a 3D model and fuses these images into a single input space. On the other hand, the similarity of multi-channel information is computed to obtain the correlation among them, thereby grouping and fusing multi-channel information. By focusing more on influential information, the classification precision is guaranteed. Experimental results demonstrate that the proposed method achieves state-of-the-art performance in 3D shape classification.