Entropy Based Generative Adversarial Network for PolSAR Image Classification

In order to solve the problem that the effect of generator feature learning is not paid enough attention in generative adversarial network(GAN) for Polarimetric synthetic aperture radar(PolSAR) data, a new GAN called Entropy-based Auxiliary Classifier Generative Adversarial Networks (E-ACGAN) was proposed in this paper. The decomposition discrepancy was obtained by calculating the entropy decomposition of the real data and the generated data. Then the decomposition discrepancy is used to measure the similarity between the generated data and the real data. This discrepancy will be introduced into the model as an additional optimization goal of the generator. Therefore, the generator can learn more characteristics of PolSAR data to generate more realistic data. In the step of adversarial learning, the discrimination and classification capabilities of the discriminator are also improved with the generator. The experimental results on the Flevoland2 data set show that the classification accuracy of E-ACGAN is 2.36% higher than that of the original ACGAN and it is also improved to different degrees than other traditional classification methods.