Multi-Domain Fusion Graph Network for Semi-Supervised PolSAR Image Classification

Round 1

Reviewer 1 Report

Please check the comments in the attachment.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper proposes Multi-Domain Fusion Graph Network for semi-supervised PolSAR image classification. In this paper, the authors propose a semi-supervised learning framework called multidomain fusion graph network (MDFGN) to explore fusing the features from multiple domains like spatial and feature domains. To expand the training set with unlabeled data, the authors propose a novel graph-based selection criterion. Then, the triplet encoder is proposed to extract the feature in a better way. Besides, a multi-leveled fusion strategy is also introduced to make the framework adaptive on image patches with different sizes. The given experiments show the strong performance of the proposed MDFGN. And this method achieves good performance on three real PolSAR datasets. The proposed innovations are novel and practicable, then corresponding experimental results are sufficient.

 

Here are some specific questions.

1.       In the patch size module, is the patch size calculated adaptively and how much does the patch size module affect the results.

2.       For multi-model triplet encoder, do all the trained models participate in the inference of the final classification result? How are the results of multiple models combined?

3.       For multi-model triplet encoder, what is the metric distance used to measure and compare the feature vectors of different patches?

4.       Is the combination of feature domain and spatial domain adaptive in sample selection? How much influence does the spatial domain have on feature extraction?

5.       The authors mention that the MDC of labeled samples is 1, but is the MDC of unlabeled samples less than 1, and if so, how is the MDC limited to less than 1?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

No more comments.

Author Response

Thanks very much for your professional comments. We have carefully spell checked the English language