Multi-Scale Convolution-Capsule Network for Crop Insect Pest Recognition

Round 1

Reviewer 1 Report

I would remove the third contribution of the work in the introduction. The proposed method is supposed to be tested on different samples to show how your method is better than the others.

Check the last line of equation (1). Also later in the text W, V and U look like small letters, related to the rest of the text.

In line 205, the mathematical expression for the sj looks shifted in relation to the rest of the text. And check similar situations  in the rest of the text.

Correct the text in line 268.

Text of Table 1.?

You have written number (6) of the equation two times.

Maybe you should rename chapter 2. somehow different (methods or something else).

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Starting from the first line of abstract, it is a very poor writing. For instance, you said something and then used “and the” which is irrelevant to the previous line. You are narrating a story!

Reaching the end of your abstract, you need to rewrite and reformat it. What we learnt from the abstract?! You need to give some background about topic, what has been done and what makes your research unique!

Again, starting from your introduction, you have “seriously affect”, what is this? Is this a story?

“using a natural language” what do you mean?Is this NLP?

Long paragraphs throughout the paper!

“Timor-Leste, consisting of 28.526 root words, 1.309 stop words and 180 words” where are you going? I thought you want to talk about CNN as you presented some related images! Or is it NLP? Going over your title and few paragraphs of introduction, it is not clear what you are trying to say, the paper is not organized and it is clear that the authors have no clear path of direction!

“is very better” again, this is an academic paper not a story! You need to proof edit your paper in terms of structure before submitting it!

Pseudocode should be added, also original codes should be shared in the GitHub

I am not going to go further over your paper as it is very poorly written and i cannot follow.

You need to proof edit and reorganize your paper before re-submission.

Answer to my comments one by one and above are just some examples. If i see any error i have to reject your paper if the editor has not rejected that already.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper presents the approach to Crop Insect Pest recognition using Multi-Scale Convolution-Capsule Network. The scope of this paper is actually. The article is very interesting.   

Comments and suggestions are the following:

• I think that it was stated too early in section 3 that MSCCN was selected for crop insect pest recognition. It should be better motivated. Only in the experimental part, experiments were carried out on other architectures and then this choice should be emphasized.
• Figure 1 – the source is missing.
• Why did you selected the following deep learning methods: “ including CNNs [12], CapsNet [15], 238 MS-CapsNet [17] and DCNN+transfer learning (DCNNTL)”. In my opinion the statement: “mainstream” is not sufficient. Please add the exploitation, why do you not  applied ML techniques architectures - e.g. ResNet or NASNet or the models which employ: Softmax, SVMand moreover also AlexNet, GoogLeNet, Cifar10Net?
• Please define more precisely and detailed the additional contribution of the research to the recent state of the research field. The discussion (now this section is missing) must include the results obtained in comparative analysis.
• The conclusion should be more informative. How the findings can be exploited by future similar works?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

This contribution presents original ideas in the study and advances the previous research in this area. The paper deserves publication. The level of the originality of contribution to the existing knowledge with an emphasis on the paper’s innovativeness in both theory development and methodology used in the study is very high.

This work makes a significant practical contribution and it makes impact on the research work on the research community.

The quality of arguments, the critical analysis of concepts, theories and findings, and consistency and coherency of debate are well addressed in this paper. 

The paper has a good writing style in term of accuracy, clarity, readability, organization, and formatting. 

Nevertheless, in Figure 2. the structure of Inception V1 is reported. Please discuss more in depth the "stability" of the proposed procedure. In fact, the structure of the algorithm presents Feedback and the convergnece of the algorithm must be discussed. In particular thestop criteria.

The objective function defined in (4) states an interesting optimization problem. Please discuss more in depth the convexity of the otpimization problem.  

Concerning the cited literature you can consider the following papers to improve the tutorial aspects of the paper.  Some of These papers are also already published in MDPI Journals.

Schimmack, M. et al. An Adaptive Derivative Estimator for Fault-Detection Using a Dynamic System with a Suboptimal Parameter. Algorithms 2019, 12, 101. https://doi.org/10.3390/a12050101

 

Mercorelli, P. A Fault Detection and Data Reconciliation Algorithm in Technical Processes with the Help of Haar Wavelets Packets. Algorithms 2017, 10, 13. https://doi.org/10.3390/a10010013

Costa, E.D.; Tjandrasa, H.; Djanali, S. Text Mining for Pest and Disease Identification on Rice Farming with Interactive Text  Messaging. International Journal of Electrical and Computer Engineering. 2018, 8, 1671-1683.

Denoising and harmonic detection using nonorthogonal wavelet packets in industrial applications P Mercorelli, Journal of Systems Science and Complexity 20 (3), 325-343, 2007.

Ai, Y.; Sun, C.; Tie, J.; Cai, X. Research on Recognition Model of Crop Diseases and Insect Pests Based on Deep Learning in Harsh Environments. IEEE Access. 2020, 8, 171686-171693. 

Biorthogonal wavelet trees in the classification of embedded signal classes for intelligent sensors using machine learning applications, P Mercorelli, Journal of the Franklin Institute 344 (6), 813-829, 2007.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

addressed