In the field of point cloud classification using neural networks, the traditional PointNet neural network only focuses on single-point feature learning and ignores local information structure features, which makes the accuracy of point cloud classification low. In this paper, we propose IF-PointNet (Pointnet of ISS3D and FPFH) point cloud classification network to fill this gap. Firstly IF-PointNet adds the module of ISS3D point cloud key point detection on top of PointNet, so as to better learn the local point cloud structure features. Then, the module of FPFH feature extraction is added to encode the features of the detected key points, which strengthens the connection between the point cloud localities and improves the classification accuracy. The experimental results show that IF-PointNet has higher classification accuracy and robustness on the ModelNet40 dataset. The results verify the effectiveness of the two added modules in point cloud classification and brings new ideas and methods to the field.