Scene Classification Method based on CNN

To address the challenge faced by indoor robots with two-dimensional (2D) LiDAR in accurately recognizing scenes, a scene classification method based on one-dimensional convolutional neural network is proposed. Firstly, the original LiDAR data are transformed in polar coordinate form, and the distance is utilized to construct 1D convolutional neural network (1D-CNN). Secondly, five laboratories data were collected, and the proposed network was trained with different kernels. The classification accuracy of optimal network can exceed 98%. Finally, it is verified that the proposed algorithm can also recognize the scene for dynamic objects. Thus, the CNN can effectively classify indoor scenes.