Abstract:This article proposes a novel autoencoder (AE) network based on the distribution of polarimetric synthetic aperture radar (POL-SAR) data matrix, called a mixture autoencoder (MAE). Through a detailed analysis of the data distribution POL-SAR data matrix, a normalization method is also presented in succession. The proposed MAE defines the data error term in the loss function according to the data distribution. It can be regarded as a process of unsupervised feature extraction designed specifically for POL-SAR data matrix. Then, a softmax classifier is trained with the help of data features and the corresponding label information. Next, a stacked MAE (SMAE) network is reasonably constructed by considering the data distribution among different layers. Finally, this article also presents a classification network through discarding the decoder process of the proposed SMAE and connecting with a softmax classifier. The SMAE is trained layer by layer using the unlabeled data. The softmax classifier is also trained with a small number of labeled pixels. With parameters obtained from the above-mentioned procedures as the initial parameters, the whole classification network is trained by the labeled pixels to get a well-trained model, which is used for predicting the corresponding label of the pixel in the data set. Three real POL-SAR data sets, including the AIR-SAR L-band data of Flevoland, The Netherlands, are used in the experiments. Compared with one classical algorithm and two related models with the similar structure, both the proposed methods show improvements in overall accuracy and efficiency as well as possess better adaptability of the parameter and preferable consistency with the classification performance.

Complex-Valued Wishart Stacked Auto-Encoder Network for Polsar Image Classification

POLSAR Image Classification Via Wishart-AE Model or Wishart-CAE Model

Pixel-Wise PolSAR Image Classification via a Novel Complex-Valued Deep Fully Convolutional Network

Complex-Valued Multi-Scale Fully Convolutional Network with Stacked-Dilated Convolution for PolSAR Image Classification

PolSAR Image Classification Via a Novel Semi-Supervised Recurrent Complex-Valued Convolution Neural Network

Polsar Image Classification Via Complex-Valued Multi-Scale Convolutional Neural Network

PolSAR Image Classification Based on Multi-Scale Stacked Sparse Autoencoder.

Multilayer Feature Learning for Polarimetric Synthetic Radar Data Classification

A Complex-Valued PolSAR Image Segmentation Network With Lovász-Softmax Loss Optimization

POLSAR Image Classification Via Clustering-WAE Classification Model.

Complex-Valued Self-Supervised PolSAR Image Classification Integrating Attention Mechanism

POL-SAR Image Classification Based on Modified Stacked Autoencoder Network and Data Distribution

PolSAR image classification based on complex-valued convolutional neural network and Markov random field

Superpixel-Oriented Classification of PolSAR Images Using Complex-Valued Convolutional Neural Network Driven by Hybrid Data

Efficiently Utilizing Complex-Valued PolSAR Image Data Via a Multi-Task Deep Learning Framework

PolSAR Image Classification with Active Complex-Valued Convolutional- Wavelet Neural Network and Markov Random Fields

Auto Encoder Feature Learning with Utilization of Local Spatial Information and Data Distribution for Classification of PolSAR Image

Complex Variational Inference Network for PolSAR Classification

Attention-Based Polarimetric Feature Selection Convolutional Network for PolSAR Image Classification

A Lightweight Complex-Valued DeepLabv3+ for Semantic Segmentation of PolSAR Image

Semi-Supervised Complex-Valued GAN for Polarimetric SAR Image Classification