Semi-supervised polarimetric SAR images classification based on FixMatch

Polarimetric synthetic aperture radar (PolSAR) images are widely used in the fields of disaster detection and military reconnaissance. PolSAR image classification is one of the most important applications in PolSAR image interpretation. Existing image classification methods based on deep learning usually rely on a large number of labeled training data. Labeling large-scale data sets is expensive and time-consuming Therefore, semi-supervised learning is an important research direction in the field of image classification. Fixmath has achieved excellent semi-supervised classification performance for nature image, so it is introduced into the field of PolSAR image interpretation, and to improve the performance of semi-supervised PolSAR image classification. Fixmatch has the following advantages: (1) It extends the training data, using generated, labeled and unlabeled data to train the network. (2) By utilizing the information of unlabeled data, FixMatch can improve the generalization ability of the classification model, thus allowing the model to better predict on unseen data. Because of these two advantages, FixMatch has shown excellent results in semi-supervised PolSAR image classification. Experiments on two real PolSAR datasets show that the overall accuracy reaches 89.9 % when the number of labeled samples is set to 110, and reaches 98.6 % when the number of labeled samples is set to 1000.