Compared with convolutional neural networks, graph convolution network is more suitable for processing irregular point cloud data. However, it has the problem that the number of network layers is limited and the fixed and standardized aggregation method affects the result of point cloud semantic segmentation. To solve these problems, a depth graph attention convolutional network for point cloud semantic segmentation is proposed herein. The network uses residual connections to deepen the number of layers of the graph convolutional network, which can effectively solve the problems of gradient disappearance and network degradation caused by the network being too deep. The attention mechanism is used to make the network selectively focus on the most relevant neighborhood points, and it assigns different attention weights to it. Simultaneously, the graph is reconstructed after each layer of graph convolution to better characterize the graph structure. Experimental results show that the average intersection ratio of the network on Stanford large-scale three-dimensional indoor spatial dataset reaches 64.5%.