Multi-Temporal PolSAR Image Classification Based on Polarimetric Scattering Tensor Eigenvalue Decomposition and Deep CNN Model

Multi-temporal polarimetric synthetic aperture radar (PolSAR) image is an important tool to monitor crops growth and evaluate disaster damage. The multi-temporal PolSAR data has the high dimensional representation. Benefited from the tensor analysis, a three dimensional polarimetric scattering tensor is established. The polarimetric scattering tensor eigenvalue decomposition is proposed to derive the polarimetric features, which are polarimetric tensor entropy, polarimetric tensor alpha angle and polarimetric tensor anisotropy, respectively. Multi-temporal PolSAR image classification is applied to validate the effectiveness of the proposed features. To further improve the classification accuracy, the 1 × 1 convolutional kernel is introduced to learn the inter-temporal information. For the multi-temporal UAVSAR datasets, the proposed method achieves the excellent classification accuracy in the multi-temporal PolSAR image classification.