RSB: Robust Successive Binarization for Change Detection in Bitemporal Hyperspectral Images

Round 1

Reviewer 1 Report

Review of 'RSB: Robust Successive Binarization for Change Detection in
bitemporal Hyperspectral Images'


The authors present a methods for performing change detection in hyperspectral imagery that uses morphological operators on top of standard similarity measures. The authors evaluate their method using a Hyperion dataset from the mid 2000s.

Overall the paper is well presented, but the material is somewhat basic. While the individual methods used in the paper are not novel, the way they are combined seems to be a novel method. However, this is a case where it is hard to judge whether this approach will be useful in other datasets, mainly because there are few high quality change detection Hyperspectral datasets. The fact that the two datasets are registered in some way, makes pixel-pixel analysis possible and useful and is probably key to the success of the algorithms in the paper.

The authors have limited the scope of this paper to a manageable size and have included the main similarity metrics as well as a survey of thresholding methods.

There are a few small grammar and spelling errors, but otherwise the paper is in good shape.

Author Response

The author thanks the reviewer for the comments. A detailed answer is uploaded in the attachment.

 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

As a general comment, the proposed method is interesting but the whole manuscript is organized unclearly. The reviewers suggest to the major revisions. Several issues are addressed unclearly, which are listed in the following:

Comment 1: The proposed method proposed a new change detection (CD) method for Hyperspectral Image (HSI), however the author neglect the noise case that the various noise exist in the original HSI. Thus, the author should discuss the denoising work such as [1-2], and analyze the anti-noise performance of the proposed method.

[1] Hyperspectral Image Restoration via Global L1-2 Spatial-Spectral Total Variation Regularized Local Low-Rank Tensor Recovery. IEEE Trans. Geosci. Remote. Sens. 59(4): 3309-3325 (2021)

[2] Nonlocal low-rank regularized tensor decomposition for hyperspectral image denoising. IEEE Transactions on Geoscience and Remote Sensing 57.7 (2019): 5174-5189.

Comment 2: The HSI is a cube, a third-order tensor as shown in [3-4], jointly considering the spatial and spectral property under tensor framework is helpful to improve the CD accuracy [5], thus it is necessary to analyze these works.

[3] Multilayer Sparsity-Based Tensor Decomposition for Low-Rank Tensor Completion. IEEE Transactions on Neural Networks and Learning Systems (2021).

[4] When Laplacian Scale Mixture Meets Three-Layer Transform: A Parametric Tensor Sparsity for Tensor Completion. IEEE Transactions on Cybernetics (2022).

[5] Huang F, Yu Y, Feng T. Hyperspectral remote sensing image change detection based on tensor and deep learning[J]. Journal of Visual Communication and Image Representation, 2019, 58: 233-244.

Comment 3: There are many parameters in the proposed methods, please provide the sensitivity analysis.

Comment 4: The manuscript misses the computational complexity analysis of the proposed algorithm.

Author Response

The author thanks the reviewer for the comments. Please, see the attachment for a detailed answer. 

Author Response File: Author Response.pdf

Reviewer 3 Report

The reviewers believe that this work is not innovative enough and does not meet the standards of remote sensing. The following points are also made:

 

1.      There is an error in the description of line 22. Generally, a 3-mode tensor or 3-dimensional tensor is used.

2.      Authors need to divide more subsections to make the article structure clearer.

3.      Authors need to add a point-by-point discussion at the end of the introduction to explain the author's contribution.

4.      The experimental results lack comparison with state-of-the-art methods and are not convincing. Why was the method chosen to make comparisons between 1995-2000 and does it make sense?

5.      The output of the continuous binarization method proposed by the author should complement the output image after each binarization.

6.      The output of the method proposed by the author is a matrixed image, and does not mention how to restore the three-dimensional structure of the hyperspectral image itself.

7.      The article format should be modified according to the journal.

 

8.      Is there something wrong with the result of the f subgraph in Figure 6.

Author Response

The author thanks the reviewer for the comments. A detailed answer has been provided in the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have addressed my comments，I suggest to accept the manuscript for publication.

Author Response

The author thanks the reviewer for the positive feedback and for the useful comments.

Reviewer 3 Report

1. The contributions of the manuscript are suggested to be summarised in three items.

2. In table 4, the proposed method does not achieve the optimal results, the reason should be addressed, and an in-depth analysis is also necessary.

3. More experiments on other data are suggested to show the effectiveness of the proposed method.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 3 Report

all my previous concerns have been modified. There's no more comments.