Channel–Spatial Segmentation Network for Classifying Leaf Diseases

Round 1

Reviewer 1 Report (New Reviewer)

In the proposed study, the channel based Spatial segmentation network is used that confirms that the sequential arrangement of the attention based mechanism suits for the leaf disease detection task and disease degrees and the model successfully predicts the four classes of diseased leaves with the highest accuracy of 99.76%..

The grammar and style of English language needs to be improved, such as on line 70 (The above study 69 focuses to detect the different in health conditions..) should be difference in health conditions. 

On line 71 it is written that” Simonyan et al. used deep convolutional neural network (DCNN) approach achieves the disease detection” . It should be, approach that achieves the disease detection.

There are seven contributions of the proposed work but in fact they are overlapped such as (i) and (ii) can be combined. Similarly, (iv) and (v) should be combined. There is no need of points (vi) and (vi) and classes types can be mentioned in point (v).

As segmentation mechanism is appended and described in the proposed work. There should be some type of segmentation results that should be presented to show the affect of segmentation for enhancing results. One such factor is the DICE ratio of the predicted and ground truth areas of the diseased regions.

Apart from the affected size of the disease, are there any other features, such as the texture of leaves, that can identify the disease. Please provide mathematical equations for features extraction from the leaves images. 

In Figure 6, the validation curve shows noticeable dips. Is it the sign of under fitting as the model is complex and thee are only a few features.

The confusion matrix for 13 types of leaves should be explained and the number of  False positives and False negatives should be mentioned. 

In the comparison work, the reference 50 is very close to your work and results are also alike. Can you implement the same data of 10 tomato classes on your model and then compare the both models. In Figure 11, multiple diseases are mentioned.. Please provide the details for the multiple diseases. 

Author Response

Dear Reviewer,

We sincerely thank you for your humble reply to our manuscript

As per your comment, we answer all the questions in both the document's responses to the reviewer and the manuscript. 

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

Authors presented the work "Channel-Spatial Segmentation Network for Classifying Leaf Diseases" which seems good but some points of suggestions are as under:

1. T prove the novelty authors should describe that how your work is different than "Early detection and classification of tomato leaf disease using high-performance deep neural network" and "Enhanced convolutional neural network model for cassava leaf disease identification and classification". Elaborate in introduction.

2. Authors claim much of the work significance. Out of all these some are just the working instead of main contributions. So, it is suggested to have a close look and narrow down the significance points to the fruitful/contribution points only.

3. Result graphs are not explained fully. Authors are suggested to explain them in detail.

Author Response

Dear Reviewer,

We sincerely thank you for your humble reply to our manuscript

As per your comment, we answer all the questions in both the document's responses to the reviewer and the manuscript. 

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Overall, the research work is interesting and well written. However, you may consider the following for betterment:

Abstract

1.     Line 20- Use appropriate terminologies; crop field and experiments in place of crop industries.

2.     Line-21: effective for plant disease prediction. Strike off “in the field of”

3.     Line – 21: less precise, strike off less potent.

Introduction

1.     Line 38, 42-45: Incorporate the appropriate terminologies highlighted with comments

2.     Line 38, 42-45: Add more global references on yield loss due to diseases. Citing only one refence mentioning Australia seems paper is intended for Australia. Give global perspective of the work, journal or crop wise loss.

3.     Line-54-55 : recommends ‘pesticide doses to cure plants’ (instead of just ‘to cure diseases of plants’).

4.     Line -73: limitations or barrier in place of “considered as a disadvantage”

5.     Line-118-119: Suggest to write a brief or reference about Top-13, 8, 5 leaf group

Materials and Methods

1.     Lines 156-162: : Why bold in some notations and why not others? Keep uniformity in the mathematical notions, Line-160 and 161-162 for activation function, at one place sigma is bold and other place not bold letter.

Results and Discussion

2.     Lines 198-211: Remove 3.1 datasets from Results and Discussion part and keep it here in Materials and Methods Section

3.     Line 200: Why not ‘classes’ instead of types

4.     line 210: Explain/elaborate about the basis of research of Top-5, 8, 13 leaf groups

5.     Line 201: Give Reference not link (as per MDPI format)

6.     Add more references on attention modelling

7.     Provide well annotated codes, or GitHub (private) repository for verification purpose

Results and discussion

1.     Line 287: Figure 5 - follow uniform pattern of MDPI. Why bold ‘Sample images of the dataset’?

1.     In Figure 6, increase legend font size for visibility, Line-297-312 and Figure-7, 8, 9; Line 313-340.

Conclusion

1.     Line-385, to identify the healthiness and unhealthy (diseased) leaf types, in place of “to identify the healthiness and illness of few leaf types)

Author Response

We were heartily Thanking you for pointing out the essential flaws in our research paper. We changed all the mentioned corrections from the reviewers.

Author Response File: Author Response.docx

Reviewer 2 Report

·       Kindly proofread the article. Many grammatical errors and incomplete sentences were detected.

·       The literature review on plant diseases detection using deep learning and attention based model is insufficient. Moreover, the limitation of previous studies should be highlighted, how the limitations direct to this study, and how the proposed method can overcome the aforesaid limitations should be pointed out.

·       Why a segmentation strategy is integrated the channel attention portion is vaguely explained in the introduction.

·       The explanation of the proposed method needs significant improvement.

·       The results and discussion are poorly presented. Authors do not critically discuss the obtained findings.

·       Comparison with other models using the same dataset should be included.

·       Statistical test to validate the obtained results should be included.

·       Figure 4 can be omitted or summarized in table form.

Author Response

We were heartily Thanking you for pointing out the essential flaws in our research paper. We changed all the mentioned corrections from the reviewers.

Author Response File: Author Response.docx

Reviewer 3 Report

Literature review/ related works section can be added as a separate section.

Conclusion section should be improved summarizing the results of the work with directions related to future work.

Deatails related to hyperparameters selection should be added.

Comparsion study is fine but need more explation in results discussion.

Advantages of CBAM should be summarized.

Explain how the proposed work is novel.

Author Response

We were heartily Thanking you for pointing out the essential flaws in our research paper. We changed all the mentioned corrections from the reviewers.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Please follow the font size as per MDPI guidelines for figure titles, subtitles, legend, and tables.

 Example: Figures 7, 8 and 9 

Author Response

Response to Reviewer 1 Comments

1. Please follow the font size as per MDPI guidelines for figure titles, subtitles, legends, and tables.

 Example: Figures 7, 8 and 9 

Response: Thanks for the denoting format mistake. To avoid this conflict, we changed the confusion matrix Figure to a table. Tables 5, 6, and 7 now represent the confusion matrix

Author Response File: Author Response.docx