doublew5/cv-cnn

Complex-Valued Convolutional Neural Network and Its Application in PoSAR Image Classification

Requirements:
Matlab

Experimental steps:
The test demo is on Flevoland dataset acquired by the AIRSAR sensor with L band.
run ./Test Demo/TestDemo.m to evaluate the performance of the CV-CNN in PolSAR image classification

The CV-CNN folder stores the implementation process of CV-CNN.

Reference: This CV-CNN code is created based on the DeepLearnToolbox, which is a real CNN open source toolbox.
https://github.com/rasmusbergpalm/DeepLearnToolbox

Datatsets: The datasets adopted in our paper "Complex-Valued Convolutional Neural Network and Its Application in PoSAR Image Classification" can be downloaded by the website "https://earth.esa.int/web/polsarpro/data-sources/sample-datasets". Meanwhile, the corresponding ground truth has been uploaded in this project.

Remark: The ground truth of Flevoland-1989 containing 14 classes is Label_Flevoland_14cls.mat
      The ground truth of Flevoland-1991 containing 15 classes is Label_Flevoland_15cls.mat
    The ground truth of Oberpfaffenhofen, Germany is Label_Germany.mat    

Updata: The input T matrix of the three datasets has been upload in the following Baiduyun link.

      Sampling data can be obtained at a certain sampling rate and divided into training set and validation set

    链接: https://pan.baidu.com/s/1GpA8f37hochOuPMWmZOkmQ 密码: 6qhk