Different PolSAR Input Schemes and Its Influence for Costal WetLand Area Classification: A Comparison

To explore the impact of data input schemes for deep learning on the final classification results, this paper studies different combinations of polarization feature input schemes using the polarization decomposition method.The classification results obtained through convolutional neural networks indicate that some current polarization input schemes fail to effectively utilize polarization information. The proposed schemes in this paper are superior to these schemes. Under limited resources, information from the polarization coherence matrix can be utilized. When the performance of the machine configuration is sufficiently good, using polarization features from the polarization decomposition scheme yields the best classification results.