研究目的
To reconstruct full polarimetric (full-pol) information from single polarimetric (single-pol) SAR data using deep neural networks.
研究成果
The proposed deep neural network method effectively reconstructs full-pol SAR data from single-pol data, showing good agreement with true data and robustness across different areas. Applications like target decomposition and classification work well on reconstructed data, though some discrepancies exist in complex terrains.
研究不足
Reconstruction errors are higher in low-intensity areas and for specific targets like ships due to lack of training samples; generalization to different terrains may require additional constraints or data.
1：Experimental Design and Method Selection:
The method uses a two-stage deep neural network framework: a feature extractor network to extract multi-scale spatial features from single-pol SAR images, and a feature translator network to map these features to polarimetric feature space for reconstructing full-pol data.
2：Sample Selection and Data Sources:
L-band full-pol UAVSAR images from NASA/JPL over San Diego, CA, and New Orleans, LA, USA, acquired in
3：Training data is from San Diego, and testing data is from New Orleans. List of Experimental Equipment and Materials:
20 UAVSAR system for data acquisition, deep neural networks implemented with convolutional and fully connected layers, VGG16 pre-trained model for feature extraction.
4：Experimental Procedures and Operational Workflow:
Preprocess VV-pol intensity images by logarithm and normalization, extract features using CNN layers, interpolate and concatenate features into hyper-column descriptors, normalize features, feed into fully connected network for translation, reconstruct full-pol data, and evaluate using error metrics and applications.
5：Data Analysis Methods:
Quantitative evaluation using Mean Absolute Error (MAE) and Bartlett Distance; qualitative evaluation through visual comparison and applications like Freeman-Durden decomposition and unsupervised classification.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
