Semi-supervised Classification of PolSAR Image Based on Self-training Convolutional Neural Network
- 18 Downloads
Convolutional neural networks (CNNs) have been successfully used in the field of polarimetric synthetic aperture radar (PolSAR) image classification. However, it is still a big challenge to perform them with limited labeled training samples. To address this problem, a semi-supervised scheme of PolSAR image classification based on self-training CNN is proposed in this letter. Our basic idea is that the PolSAR image to be classified is actually an important source of training samples, from which we iteratively extract reliable samples to enhance the CNN. In our scheme, a CNN is initially trained by the given training samples. Then, the PolSAR image is classified by the trained CNN, of which the pixels with high predicted probability, along with their predicted labels, are selected as new training samples. The operations of retraining CNN, reclassifying PolSAR image and reselecting training samples, are iteratively performed until a stop condition is met. Two actual PolSAR images acquired by AIRSAR and Gaofen-3 systems are employed to verify the effectiveness of the proposed algorithm. Experiment results demonstrate the superiority of our method to the general CNN-based classifiers.
KeywordsPolarimetric synthetic aperture radar Semi-supervised classification Self-training convolutional neural network
This work is supported by National Science Foundation of China under Grant 41601436, 61403414 and 61773396 and in part by the Natural Science Basic Research Plan in Shaanxi Province of China under Grant 2018JM4029 and 2019JM-554.
- 5.Liu B, Hu H, Wang H, Wang K, Liu X, Yu W (2013) Superpixel-based classification with an adaptive number of classes for polarimetric SAR images. IEEE Trans Geosci Remote Sens 51(2):907–924Google Scholar
- 6.Hou B, Kou H, Jiao L (2016) Classification of polarimetric SAR images using multilayer autoencoders and superpixels. IEEE J Sel Topics Appl Earth Observ Remote Sens 9(7):3072–3081Google Scholar
- 12.Goodfellow I, Bengio Y, Courville A (2016) Deep learning. MIT PressGoogle Scholar
- 14.Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional neural networks. Proc Adv Neural Inform Process Syst 1097–1105Google Scholar