Multi-AUV Target Recognition Method Based on GAN-meta Learning
Qiankun Sun, Lei Cai · 2020
The presence of adverse factors such as turbid water quality and target occlusion prevented the acquisition of valid target characterization data. Due to the variable shape of the target, the accuracy of recognition for untrained new targets is low. In view of the above problems, this paper proposes a Multi-AUV target recognition method based on GAN-meta learning. VGG-19 network is used for feature extraction of target images, and the WGAN network is used to make up for missing target information. Based on the meta-learning theory, the parameters of the feature extraction process are trained by using stochastic gradient descent to improve the algorithm's ability to recognize new targets. Ensure that the GAN-meta learning model has a strong generalization ability. Simulation experiments are conducted on four underwater targets (underwater reconnaissance equipment, submarine, frogman, and torpedo) in the SUN dataset, and the results demonstrate that the proposed model achieves better performance.