H-RNet: Hybrid Relation Network for Few-Shot Learning-Based Hyperspectral Image Classification
, Jinchang Ren 3,5, Huimin Zhao 3, Hao Li 3, Weiqi Chen 3 and Zhanhao Xiao 3
Round 1
Reviewer 1 Report
Review for H-RNet: Hybrid Relation Network for Few-shot Learning-2 Based Hyperspectral Image Classification
This paper proposes a new few-shot learning based deep learning network, H-385 RNet, for effective feature extraction and data classification of HSI. By combining the hy-386 brid 3-D/2-D CNNs with relation networks, HR-Net can help to extract spectral and spa-387 tial information of HSI efficiently.
The paper is generally well presented and structured and the idea seems fine to me however, it suffers some minor issues as follow:
1. There is not enough explanation about the proposed technique. Please present the objectives clearly in the introduction section.
2. Related work of the paper is well written and provides a good literature survey of the field. However, more updated references are missing.
3. The quality of figures is not good. So, authors are requested to include a good quality figure.
4. Experiments should be extended. There are other indexes that can be compared.
5. The proposed approach is mixing several works. Please explain where is the novelty?
6. The proposed method overhead is high. It is recommended to compare proposed method with previous methods.
7. The conclusions section should conclude that you have achieved from the study, contributions of the study to academics and practices, and recommendations of future works.
Overall, I feel the manuscript may be accepted for publication if the above comments/questions are Addressed.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
A hybrid relation network for hyperspectral image classification, H-RNet, is proposed, which combines the three-dimensional (3-D) convolution neural networks (CNN) and two-dimensional (2-D) CNN to extract the spectral-spatial features. But the idea is similar to the reference [40]. Therefore, this paper is not innovative enough in my opinion. In addition, there are some defects that need further improvement.
(1) This paper takes a C-way K-shot N-Query episode to train the model that C is set to the class number of the dataset. But C is usually set to be less than the class number of the dataset in common few-shot methods. Parameter C is an important parameter and its influence on the accuracy of the model should be analyzed.
(2) The relation learning module is not presented in detail. What is the output of the relation learning module? How to get the output in Figure.1?
(3) Is the parameter number in the last layer of the relation learning module 3137? More details are needed.
(4) Recent methods are needed for comparison to demonstrate the effectiveness of the proposed model.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
no comment.
Reviewer 3 Report
The authors have addressed all my questions, and I have no other extended questions.
