Hyperspectral Image Classification Based on Adaptive Global–Local Feature Fusion

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The author proposed an adaptive global-local feature fusion (AGLFF) method To solve the problem that the semi-supervised method based on graph can not fully utilize the spatial-spectral information,which leads to low classification performance. My detailed comments are listed below, which hopefully can help the authors improve the quality of their work.

1.In Figure1,Framework of AGLFF model，the way of presentation is not concise enough, and the text description in the picture is too much. It is suggested that the author redesign the frame diagram to simplify the content and highlight the key points in order to communicate the core ideas and key steps of the research more clearly.

2.The purpose of this paper is to solve the shortcomings of the graph-based semi-supervised method, and it is suggested to add semi-supervised algorithms when doing comparison experiments, especially the comparison experiments of graph-based semi-supervised algorithms.

3. In the comparison experiment, it is not clear whether the number of training samples of this algorithm is consistent with that of the comparison algorithm.

4.It is suggested to add the chapter of ablation experiment in the experimental section to better demonstrate the effectiveness of the innovation in this paper.

5.The year of the comparative experiment selected in this paper is relatively old, so it is suggested that the author consider adding several representative algorithms in recent years to enhance the timeliness and credibility of the study.

Comments on the Quality of English Language

Minor editing of English language required.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

An AGLFF classification method is proposed. Adaptive fusion of global high-order and local data can realize feature smoothing of intra-class data and increase the discriminability of inter-class data. The adaptive method automatically learns the weight parameters of the global high-order and local data, which can reduce the number of parameters calculated.  But the novelty of this method is not well described, and the experimental results are not clearly displayed.

Comments on the Quality of English Language

Moderate editing of English language required。

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

 

The main objective of the paper is a semi-supervised learning (SSL) method for hyperspectral image classification that incorporates feature fusion, graph processing, and regularization. Those methods are state of the art in machine learning. The contribution of the paper is good as it contains theoretical developments and extensive experiments. Thus, it might be interesting from a practical standpoint. In general, literal presentation of the paper is good, but there is still room for improvement in this regard. Some related works should be discussed. The evaluation of the results should be improved including statistical significance analysis and computational cost estimation. In summary, I consider the contents of the paper are potentially publishable, but the following issues should be addressed in a revised version of the paper.

- The literal presentation has room for improvement. For instance, (i) to improve readability, acronym definition of “n.l.s” and “n.u.s” could be include at header of Table 1. (ii) Title of Section 3.3.1 it could be added or changed to “Semi-supervision ratio 1 – 100%”. This latter term is used more often than “Impact of Different Sample Numbers”. It should be changed in figures, e.g., Figure 5. (iii) Please consider to change the “probability class structure” to “class probability structure”. According to the equations, I think, this latter term seems more adequate.

Therefore, an English proofreading of the paper is recommended.

- References to agglomerative clustering based on probabilistic models for hyperspectral images should be included and discussed. I suggest the following reference: https://doi.org/10.3390/rs12213585.

- The paper lacks a comprehensive analysis of the statistical significance of the results. It is important to include measures of statistical significance, such as p-values or confidence intervals, to assess the reliability and significance of the reported findings. Incorporating a rigorous statistical analysis would enhance the scientific rigor and strengthen the conclusions drawn from the experiments. In addition, the variability of the classification results should be estimated and discussed. Thus, the mean and standard deviation of a set of Montecarlo experiments (randomly changing the training and testing datasets) should be estimated and discussed.

- Please extend the discussion on results of Figure 3.

- The proposed method implements a feature fusion step, which is usually called early fusion. In addition, regularization and graph processing are included. The fusion can be also implemented at result level using the posterior probability provided by multiple classifiers, which is called late fusion. Recently, graph regularization methods for soft late fusion have been proposed, improving results of single classifiers. Please discuss it as a possible line of research. I suggest the following reference: https://doi.org/10.1109/ACCESS.2023.3344776. 

- The computational burden of the implemented methods should be theoretically and/or experimentally analyzed. Some particular execution times of the methods could be included.

 

 

Comments on the Quality of English Language

In general, English is good, some comments are in "Comments and Suggestions for Authors" section.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed all my comments. 

Reviewer 3 Report

Comments and Suggestions for Authors

The quality of the paper has been improved significantly. All my concerns have been adequately addressed, including the following: improvement of the literal presentation of the paper; improvement of related bibliographic references; addition of a statistical significance analysis of the results; extension of the result discussion; and addition of a computational burden analysis of the implemented methods. Therefore, the paper should be ready for publication.