Overview of Pest Detection and Recognition Algorithms

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this manuscript, the primary neural network architectures and evaluation metrics in the field of pest detection and pest recognition are introduced. Public datasets widely used for pest detection and recognition are summarized. The algorithm and their respective performance metrics are also introduced. The challenges and future research directions are provided. This work is very good, but there is one suggestion.

Point 1: The manuscript lists the work and results of many references, and it is recommended to add essential comparisons of the differences between these results, otherwise the paper will be like a journal entry.

 

Author Response

Comments 1: The manuscript lists the work and results of many references, and it is recommended to add essential comparisons of the differences between these results, otherwise the paper will be like a journal entry.

Response 1: Thank you for pointing this out. I agree with this comment. Therefore, I have added essential comparisons of the differences among various algorithms based on the different datasets used. These comparisons include statistical charts comparing the main metrics of pest detection tasks and pest identification tasks, as well as the pros and cons of each algorithm. These changes can be found - page 12, Figure 12 and Table 3; page 13, Table 4; page 14, Figure 13; page 15, Table 5; page 17, Figure 14; page 19 Table 7; and page 20 Figure 15 and Table 8.

The marked revised manuscript is attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In the manuscript titled "Overview of Pest Detection and Recognition Algorithms", the authors present a comprehensive review of pest identification algorithms, focusing on target identification and target detection purposes in the field of pest detection and identification. The references are within the last five years, and the summarized studies are highly relevant and informative for further research in this field.

However, the lack of analysis is a major shortcoming of this study. Therefore, I believe it is suitable for publication after major revisions.

Main comments:

1. The number of references is limited and should be increased.

2. The paper has less analysis of each network model, and a section should be added to analyze and compare the pros and cons of different networks and summarize them in a table.

3. Although the second section introduces CNN and Transformer as deep learning frameworks, respectively, there is limited analysis specifically for Transformer.

4. The title of Figure 4 is incorrect; it should be changed to "Forward propagation architecture of Transformer model".

5. Section 3 introduces more challenging tasks compared to the D0 dataset using IP102 and Pest24 datasets, but does not explain the differences between these datasets and the D0 dataset.

6. In Section 4, the literature of the past five years is summarized by two purposes: object recognition and object detection. However, the subsequent analysis does not distinguish the effects achieved by these two purposes.

7. In Sections 4.1 and 4.2, when summarizing each literature, the focus should be mainly on describing the improvements achieved in the research and comparing the results achieved.

8. There is a lack of strong logical coherence between the challenges discussed in Section 5 and future research directions.

9. In Section 5, the diversity of pest appearances is mentioned, and the use of a hybrid network structure can make up for the shortcomings of a single network. However, earlier in the paper, only the effect of combining the two types of networks is described without directly comparing them with a single network.

Comments on the Quality of English Language

The quality of English in this article is good, the statements are concise enough, and more professional vocabulary is used.

Author Response

Comments 1: The number of references is limited and should be increased.

Response 1: Thank you for pointing this out. I agree with this comment. Therefore, I have increased the number of references for algorithms based on CNN and Transformer structures. These changes can be found – page 2, line 77 and line 78; page 5, line160 and line 161.

Comments 2: The paper has less analysis of each network model, and a section should be added to analyze and compare the pros and cons of different networks and summarize them in a table.

Response 2: Thank you for pointing this out. I agree with this comment. Therefore, I have summarized the advantages and disadvantages of CNN and Transformer, as well as the pros and cons of each network model, and compared them in different tables based on the dataset. These changes can be found – page 7, Table 1; page 12 Table 3; page 13, Table 4; page 15 Table 5; page 19, Table 7; and page 20 Table 8.

Comments 3: Although the second section introduces CNN and Transformer as deep learning frameworks, respectively, there is limited analysis specifically for Transformer.

Response 3: Thank you for pointing this out. I agree with this comment. Therefore, I have added detailed descriptions of the Transformer architecture, the Vision Transformer for image recognition, and the Detection Transformer for object detection. These changes can be found – page 5, line 167 and page 6, line 182.

Comments 4: The title of Figure 4 is incorrect; it should be changed to "Forward propagation architecture of Transformer model".

Response 4: Thank you for pointing this out. I agree with this comment. Therefore, I have corrected this error. This change can be found – page 5, Figure 4.

Comments 5:  Section 3 introduces more challenging tasks compared to the D0 dataset using IP102 and Pest24 datasets, but does not explain the differences between these datasets and the D0 dataset.

Response 5: Thank you for pointing this out. I agree with this comment. Therefore, I have added a brief description of the D0 dataset. The D0 dataset is a small-scale dataset, and the following part of Section 3 discusses the challenges that the IP102 and Pest24 datasets provided, which the D0 dataset could not. This change can be found – page 5, line 244.

Comments 6: In Section 4, the literature of the past five years is summarized by two purposes: object recognition and object detection. However, the subsequent analysis does not distinguish the effects achieved by these two purposes.

Response 6: Thank you for pointing this out. I agree with this comment. Therefore, I have distinguished between pest detection and pest recognition in the future development directions section. Since in practical applications the challenges faced by pest detection algorithms and pest recognition algorithms are essentially the same, and different methods can be used to improve accuracy depending on the task, pest detection and pest recognition were not distinguished in the section on algorithm challenges. This change can be found – page 21, line 573 and page 22, line 597.

Comments 7:  In Sections 4.1 and 4.2, when summarizing each literature, the focus should be mainly on describing the improvements achieved in the research and comparing the results achieved.

Response 7: Thank you for pointing this out. I agree with this comment. Therefore, I have revised the descriptions of each algorithm, focusing on the methods and improvements adopted by each algorithm, and compared the performance of different algorithms based on the different datasets. These changes can be found – page 10, line 298; page 12, Figure 12 and line 315; page 13, line 336; page 14; page 15, line 411; page 17, Figure 14 and line 446; page 17; page 19, line 517; and page 20, Figure 15.

Comments 8: There is a lack of strong logical coherence between the challenges discussed in Section 5 and future research directions.

Response 8: Thank you for pointing this out. I agree with this comment. Therefore, I have revised the future research directions for pest detection and recognition algorithms based on the current challenges they face and the comparative analysis of their performance presented earlier. This change can be found – page 21, line 574 and page 22, line 598.

Comments 9: In Section 5, the diversity of pest appearances is mentioned, and the use of a hybrid network structure can make up for the shortcomings of a single network. However, earlier in the paper, only the effect of combining the two types of networks is described without directly comparing them with a single network.

Response 9: Thank you for pointing this out. I agree with this comment. Therefore, I have added the data comparing the improved hybrid network proposed in paper [108] with various single network. This change can be found – page 21, line 593.

The marked revised manuscript is attached.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors reviewed the latest advancements in pest detection and recognition algorithms, providing detailed descriptions of various algorithms proposed in recent years and their performance metrics. Deep learning algorithms are gradually replacing traditional machine learning-based pest detection methods, particularly Convolutional Neural Networks (CNN) and Transformer models. These deep learning algorithms can automatically learn complex feature representations, improving detection accuracy. The reviewers have the following suggestions and concerns regarding the study:

1. The study employs traditional methods without any original theoretical contributions.
2. It is recommended that the authors include visual results in the manuscript to compare the performance of different models across various datasets.
3. The manuscript's writing and structure need to be revised to meet the standards of journal article writing.

Author Response

Comments 1: The study employs traditional methods without any original theoretical contributions.

Response 1: The references cited as number 95 and number 115 in this paper are our previous works. This paper is a review article focusing on recent advances in pest detection and pest recognition algorithms over the past five years. It summarizes the current challenges faced by these algorithms and proposes future research directions for pest detection and recognition. This paper does not involve any original theoretical contributions.

Comments 2: It is recommended that the authors include visual results in the manuscript to compare the performance of different models across various datasets.

Response 2: Thank you for pointing this out. I agree with this comment. Therefore, I have added statistical graphs comparing the metrics of algorithms based on different datasets. These changes can be found – page 12, Figure 12; page 14, Figure 13; page 17, Figure 14; and page 20, Figure 15.

Comments 3: The manuscript's writing and structure need to be revised to meet the standards of journal article writing.

Response 3: Thank you for pointing this out. I agree with this comment. Therefore, I have reorganized the structure of my paper, adding several subsections and sub-subsections to meet the standards of journal article writing. These changes can be found – page 5, line 167; page 6, line 182 and line 203; page 10, line 298; page 12, line 315; page 13, line 336; page 17, line 445; page 18, line 497; page 21, line 573; and page 22, line 597.

The revised manuscript with marked annotations is attached.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

As a literature review, the methods and strategies used by the authors to collect the literature is a must, how many pieces of literature the authors have collected, the characteristics and sources of these pieces of literature need to be given in a graphical form.

Figures 12 through 15 need further optimization, graphic fonts and formatting.

The formatting of the references is misplaced and needs to be standardized across the board.

What specific kinds of pests are described by the authors as pest recognition algorithms, and what scenarios are these pests used in? These need to be described clearly, too large a range is not appropriate.

Comments on the Quality of English Language

English needs to be modified appropriately

Author Response

Comments 1: As a literature review, the methods and strategies used by the authors to collect the literature is a must, how many pieces of literature the authors have collected, the characteristics and sources of these pieces of literature need to be given in a graphical form.

Response 1: Thank you for pointing this out. I agree with this comment. Therefore, I have added the "Literature Collection Methods" Methods. This change can be found - page 2, line 57; and page 3, line 72.

Comments 2: Figures 12 through 15 need further optimization, graphic fonts and formatting.

Response 2: Thank you for pointing this out. I agree with this comment. Therefore, I have optimized these figures. I have optimized these figures, changed the fonts in these figures to Palatino Linotype, which the article uses, and adjusted the text size. These changes can be found – page 13, Figure 13; page 15, Figure 14; page 18, Figure 15; and page 21, Figure 16.

Comments 3: The formatting of the references is misplaced and needs to be standardized across the board.

Response 3: Thank you for pointing this out. I agree with this comment. Therefore, I have checked the reference format in this article and corrected the previous errors. This change can be found - page 23, line 634.

Comments 4: What specific kinds of pests are described by the authors as pest recognition algorithms, and what scenarios are these pests used in? These need to be described clearly, too large a range is not appropriate.

Response 4: Thank you for pointing this out. I agree with this comment. Therefore, I have added the types of pests that each algorithm can recognize in the description of every pest identification algorithm model tested on the self-built dataset. These changes can be found - page 19, lne 521; page 19, line 530; page 20, line538; page 19, line 543; page 20 line 553; page 21, line 565; and page 21, line 580.

The marked revised manuscript is attached.

Author Response File: Author Response.pdf