3D point clouds are becoming increasingly popular in various 3D object representations, where point-based methods have shown good performance on a variety of datasets. In response to PointNet++ only focuses on the information of the points themselves and not the information of the neighboring points, while it uses max-pooling to aggregate local information, resulting in the loss of sub-maximal information. As a result, the Con-PointNet++ network is proposed to make full use of the enhanced Local Information Module to focus on the information of neighboring points and thus enhance the local information feature extraction. Then, the fusion pooling module under the local attention mechanism is used to fuse the max-pooling and attention pooling feature information to obtain richer local feature information. The proposed method evaluates the model semantic segmentation ability on Area_5 of indoor dataset S3DIS, with mIoU of 55.2%; and the model classification effect on dataset ModelNet40, with OA of 91.2%. Compared with other methods, the performance of the proposed model is improved, which further demonstrates the effectiveness of the proposed method.