WGNet: Wider graph convolution networks for 3D point cloud classification with local dilated connecting and context-aware

•A local dilated connecting (LDC) module is presented to generate the adjacency matrix for a graph, which expand the receptive field of graph convolution to encode more information.•To extract the node features as the initial input of GCNs, a context information aware (CIA) module is designed to embed the distribution characteristics of its neighborhood points and its local dimension features, resulting in rich distribution pattern awareness.•A wider and efficient skip-connection-based GCNs combined with LDC and CIA modules is proposed to mine richer features to compensate the insufficiency on the depth of GCNs.