Unsupervised Representation Learning Method for UAV's Scene Perception

In view of the application requirements for UAV in forest rescue and border patrol, this paper takes a single small UAV equipped with visual sensors as the research object. Combined the encoder and decoder structure in the field of deep learning, a deep coding algorithm for scene perception is put forward. Based on the autoencoder, the algorithm achieves the dimension reduction and scene reconstruction by increasing the number of layers in the coding network. Based on the ROS environment, this paper sets up a simulation experiment data acquisition system and verifies the effectiveness of the dimensionality reduction algorithm.