Signal Classification Using Joint Multi-order Domain Distribution and Deep Learning Method

To solve the problem of small sample data transfer learning in aircraft environment experiments, we propose a joint multi-order domain distributed difference method (JMDD), which uses Gaussian kernel function to combine and unify the distribution difference information of multi-order saturated domain, so as to learn more domain invariant features end-to-end and solve the cross-domain alignment problem more significantly. The JMDD method can be applied to the actual migration task and is easy to train and implement. It is solved that the existing domain adaptive methods cannot use simple domain distribution matching information, so they cannot always compensate the performance degradation caused by domain migration. This method can be applied to the transfer learning process of small data samples. This kind of deep transfer learning is based on a deep adaptive network that can learn to migrate, which effectively reduces the distribution difference between domains and compensates for the performance degradation caused by domain migration. Experiments on common data sets show that our method has better classification accuracy than the existing adaptive methods.