An Aeromagnetic Compensation Algorithm Based on a Residual Neural Network

Round 1

Reviewer 1 Report

Dear Editor,

The manuscript topic "A neural network aeomagnetic compensation algorithm based on residual connection" is interesting, but some correction needs to be made.

The abstract should be more comprehensive, and the key words should not be suitable. Use other keywords.

Is the line 2 introduction referenced in 12-2?

The introduction is too weak. rewrite, and what is the novelty of this manuscript?

The quality of Figure 4 is weak.

 conclusion should be improved.

Use more and new references.

Mainly, the quality of figures is rather weak.

Author Response

Please see the attachment

Author Response File: Author Response.doc

Reviewer 2 Report

The authors presented a very interesting approach for aeromagnetic compensation based on a Residual Back propagation neural network (Res-bp). Res-bp is introduced to address the problem of gradient disappearance in neural networks. Data collected by UAV is considered to highlight the effectiveness of the proposed methodology. The paper is well structure, the contributions are clear, the experiment is well presented and conducted, and the limitations and future work introduced by the authors are well specified. The following points are just minor comments that the authors should consider for the revision of this manuscript:

1)    Please provide the full description of the acronym before utilising them. For instance, UAV is introduced without previous indication that this acronym refers to Unmanned Aerial Vehicles.

2)    There seems to be a mistake on line 2 when introducing the reference (12-2]). Analogously, please revise the last paragraph of the introduction section ([12]12).

3)    Please provide references or other means of evidence that sustain the advantages presented at the end of the introduction section with regards to the application of residual connection.

4)    Please include recent advances related to deep learning and aeromagnetic compensation. These studies should be considered in the form of a critical review. Which are the advantages and disadvantages of these studies? How these studies compare with the proposed method? Why is the contribution of this manuscript needed if these studies are considered? Please see an example of relevant paper hereunder.

 

Zhang et al. (2021). Analysis of Aeromagnetic Swing Noise and Corresponding Compensation Method. IEEE Transactions on Geoscience and Remote Sensing, doi: 10.1109/TGRS.2021.3095564.

Zhang et al. (2021). Analysis of Aeromagnetic Swing Noise and Corresponding Compensation Method. IEEE Transactions on Geoscience and Remote Sensing, doi: 10.1109/TGRS.2021.3095564.

 

5)    The reviewer considers that it would also be interesting to analyse how analogous sectors deal with the problem of gradient disappearance in neural networks and how they compare with the proposed approach. For instance, Wang (2022) presented a fault diagnosis method of wind turbine generator based on residual autoencoder network (RAE). Shortcut connections were considered in both encoder and decoder to avert the problem of gradient disappearance.

 

Wang Z., 2022. Fault Early Warning of Wind Turbine Generator based on Residual Autoencoder Network. Proceedings of the 6th International Conference on High Performance Compilation, Computing and Communications, url: https://dl.acm.org/doi/proceedings/10.1145/3546000.

 

6)    The neural network structure is widely known and utilised in multiple studies. Thus, the reviewer considers that the Figure 1 can be skipped unless a novel contribution is specified in form of a graphical representation.

7)    The authors compared the methods BP and Res-BP by utilising STD and IR in order to evaluate the magnetic interference compensation algorithm. Why are these metrics specifically considered? What about the computational time?

8)    Could the authors provide further information with regards to the implementation of the Res-Bp model in case the readers are interested in replicating the proposed methodology?

 

Overall, a very interesting article. The reviewer is willing to see further advancements in this topic based on the future work specified by the authors.

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Round 2

Reviewer 1 Report

The quality of the manuscript has been improved. I suggest minor English checking.