Abstract: The method of using multi-view 2D images for 3D shape recognition has achieved excellent classification and retrieval performance on many datasets. However, rendered 2D images can only capture the geometric information of the surface of 3D objects, and cannot explore the complex internal topological structures of objects, especially for models with porous structures. To address this issue, this paper proposes a classification network based on multi-perspective topological features. Firstly, the multidimensional topological information extracted from porous objects from multiple perspectives is used as the input feature of the network to achieve a more comprehensive perspective of the internal structure of objects. Then, through the topological feature aggregation layer, the fusion of multi-view multidimensional topological information is completed to generate more compact global topological features for subsequent classification and retrieval tasks. The experimental results on a 7-class TPMS synthetic porous dataset and a 192-class zeolite dataset show that compared to multi-view convolutional neural networks and geometric- or topology-based descriptors, the proposed method has certain advantages, with classification accuracy and retrieval nearest neighbor reaching 99.52% and 97.12%, respectively.