Hybrid Conditional Random Fields Based on Complex-valued 3D CNN for PolSAR Image Classification

In this paper, we propose a hybrid conditional random fields based on complex-valued 3-dimensional convolutional neural network (CV3DCNN), named as HCRF-CV3DCNN, for polarimetric synthetic aperture radar (PolSAR) image classification. HCRF-CV3DCNN extracts deep features of PolSAR image by CV3DCNN, and thus exploits amplitude and phase information of PolSAR data to measure the class probability in the framework of random fields. Additionally, based on the CV3DCNN-based class probabilities, the relative entropy of class distributions is derived and utilized to regulate the label interactions to enhance the accuracy of edge location in the classification. At last, to capture PolSAR image information in a more complete manner, the deep features and PolSAR scattering statistics are integrated into HCRF-CV3DCNN based on Bayesian fusion. In this way, HCRF-CV3DCNN effectively combines the representation-learning ability of deep learning model with the modeling power of random fields including spatial correlation and data statistics. The experimental results demonstrate the superiority of HCRF-CV3DCNN over the recent deep learning models for PolSAR image classification.