Hyperspectral Image Classification Based on a Least Square Bias Constraint Additional Empirical Risk Minimization Nonparallel Support Vector Machine

Round 1

Reviewer 1 Report

In this manuscript, the authors propose an SVM model for the hyperspectral data classification task. The proposed model is expected to solve the problem of the long computation time of some similar algorithms when the number of hyperspectral training datasets is large.

Although I find the topic interesting, in my opinion, this manuscript lacks originality, which is highly required by the Remote Sensing Journal. Indeed, the authors only combine known techniques to achieve the considered task. Moreover, the conducted experiments and obtained results are not convincing: further experiments are required to validate the proposed model. Furthermore, the manuscript is difficult to read and follow.

Author Response

Please see the attachnment.

Author Response File: Author Response.docx

Reviewer 2 Report

I have found this manuscript clear and well written.
IIts content suggests interesting and original ideas to improve the execution speed of LS-BC-AERM-NSVM compared to the basic one of BC-AERM-NSVM while obtaining a competitive (similar) classification accuracy.
This is achieved by the formalisation of the dual problem of LS-BC- AERM-NSVM as an unconstrained convex quadratic programming problem.
The underlying ideas are original and quite relevant in this context and for this specific objective of reducing execution time.
The Authors have provided a well-structured exposition of their material with a gradual description of the underlying ideas fully appreciable.
The content is self-explanatory and described with a sufficient level of detail to understand the topic, techniques and results.
The experimental part is well described and offers a clear (but still limited) illustration of the contribution of the proposed method.
The analysis provided is well conducted and corresponding results are fully appropriate to the text and its content.
The list of references to the literature related to the field is also quite appropriate.

Nevertheless, I have some questions of detail that need to be developed for a better clarity.

1) At top of page 3, BC-AERM-NSVM is supposed to be based on the non parallel plane SVM

2) Recall input parameters to Algorithm 1

3) Give general recommendations (if possible) to set the hyperparameters c1, c2, c3, c4 of the LSBAENSVM model

4) Specify how the solution depends on this setting (mentioned in the previous question) and indicate trends when the setting is not sufficiently optimised. Should this optimization be done manually?

5) Why consider only the AO as an evaluation criterion in subsection 2.6? Please justify your choice further or introduce Kappa coefficients as well while explaining what one crtiterion contributes in relation to the other one (as OA and Kappa are used in a basic way in 3.1).

6) In table 1, specify the percentage of training data sets (10%, 20%, 30% and 40%) in each column heading.

7) In table 2, if the results reported are those obtained after 10 runs, this must be specified and the standard deviation must also be given

8) Other hyperspectral data sets (among those usually employed in the literature) could be used to complete the experimental part

9) How to make full use of the spatial and spectral information in the hyperspectral data by combining them within the framework proposed here could be discussed more deeply in this manuscript before concluding

Typo(s):
to Formula (17) ...  from Formulas
FIG. 4

Author Response

Please see the attachnment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Although the authors have made some modifications to the initial manuscript, I remain on my first feeling: the article lacks the necessary originality.

Reviewer 2 Report

The authors corrected the content of their manuscript according to my expectations. They introduced the itemss that were necessary for a better presentation of their work. The revised content has been thus clarified and enriched. 

The current revision results in a better delineation of the strengths and weaknesses of the proposed method.

The revised manuscript can be accepted for publication.