Image Recognition and Classification of Farmland Pests Based on Improved Yolox-Tiny Algorithm
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe main objective of the paper is a method to detect up to 7 farmland pests. The proposed method is based on an improved Yolox-tiny-based detection method that incorporates a Convolution Block Attention Module (CBAM) step and powerful data preprocessing. Although the methods are known, the paper is interesting from a practical standpoint. Thus, the contribution of the paper is fair. However, there are issues that need to be solved. In general, the literal presentation of the paper is good. The evaluation of results has room for improvement, including: adding a variability statistical significance analysis, computational burden analysis, and improving the analysis of the implemented detectors. In addition, the discussion on feature fusion should be improved. In summary, I consider the contents of the paper are potentially publishable, but the following issues should be addressed in a revised version of the paper.
- Please state the list of contributions of the paper at the end of the Introduction section.
- The dataset is from the 10th "Teddy Cup" Data Mining Challenge in 2022. Please extend the discussion on previous results using this database. In addition, please consider to add another dataset to the experiments.
- The paper lacks a comprehensive analysis of the statistical significance of the results. It is important to include measures of statistical significance, such as p-values or confidence intervals, to assess the reliability and significance of the reported findings. Incorporating a rigorous statistical analysis would enhance the scientific rigor and strengthen the conclusions drawn from the experiments. In addition, the variability of the classification results should be estimated and discussed, i.e., the mean and standard deviation of a set of Montecarlo experiments (randomly changing the training and testing datasets).
- It is discussed in Section 2.2 about the step of feature fusion. The fusion at different levels of the data processing allows the performance in terms of accuracy and variability of the results to be improved. Recently, a theoretical study provided a comparative analysis of early (feature) and late (scores) fusion. In addition to the feature fusion, a late fusion step might be considered by combining, under some criterion of optimality, the results from the several implemented methods (e.g., Yolox-tiny, improved Yolox-tiny, YoloV4-tiny). Please discuss this. I suggest the following references: https://doi.org/10.1109/ACCESS.2023.3296098, https://doi.org/10.1109/ACCESS.2023.3344776.
- The comparisons of the implemented methods should include a theoretical and/or experimental computational burden analysis (estimation of the computational order, e.g., using big O notation) of the implemented methods. For experimental analysis, some particular running times of the implemented methods could be compared.
- Receiver operating characteristic (ROC) curve and area under the ROC curve (AUC) are good indicators to evaluate the performance of detectors at different regimes. AUC estimated in the full range of false positive rate (FPR) may not allow you to see the detector's capabilities. In practice, AUC is evaluated in the low or very low FPR range (e.g., 0 to 0.1), which is the one that is really of interest in most applications. Please add a ROC curve analysis showing some of the estimated ROC curves zooming in the low or very low FPR values and discuss this. In addition, the Precision-Recall curve and area under the Precision-Recall curve (AUPR) also should be estimated.
Comments on the Quality of English Language
In general, English is fine, but an English proofreading is recommended.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThere are several issues with the paper. First of all, the related work section is not robust enough to be considered into a research paper. The purpose of this section (i.e., related research) is to clearly highlight the deficiencies of existing research work in this area and thereby pointing out the research gap or research problem…Thus showing clear justification and rationale of the proposed work. In this version of the paper, the research gap / problem, justification of this research etc. is not evident.
To effectively solve this issue, the authors need to conduct a more robust literature survey (only 20 references are extremely poor, especially in this well-established research domain of image recognition / classification). Then, the authors need to present the findings in a TABLE in section 2, highlighting the algorithms, technology used, performance, and disadvantages of all the surveyed existing approaches referencing to each of the reviewed literatures. Talking about reference, what are these???
· 15. Sozzi, M.; Cantalamessa, S.; Cogato, A.; Kayad, A.; Marinello, F. Automatic Bunch Detection in White Grape Varieties Using 367 YOLOv3, YOLOv4, and YOLOv5 Deep Learning Algorithms. Agronomy. [CrossRef]
· 16. Kulkarni, S.; Mittal, N.; Gupta, R.; N, P. Investigation of YOLO models in the detection and classification of multiple negative road anomalies. 2023 14th International Conference on Computing Communication and Networking Technologies 370 (ICCCNT), pp.1-7.
· 17. Peng, M.; Zhang, W.; Li, F.; Xue, Q.; Yuan, J.; An, P. Weed detection with Improved Yolov 7. EAI Endorsed Transactions on Internet of Things. [CrossRef]
· 18. Ren, R.; Sun, H.; Zhang, S.; Wang, N.; Lu, X.; Jing, J.; Xin, M.; Cui, T. Intelligent Detection of Lightweight “Yuluxiang” Pear in Non-Structural Environment Based on YOLO-GEW. Agronomy. [CrossRef]
What sort of reference style is this? Where are the Publication year, Volume, issue etc. for these reference? For example, reference 17 says “Peng, M.; Zhang, W.; Li, F.; Xue, Q.; Yuan, J.; An, P. Weed detection with Improved Yolov 7. EAI Endorsed Transactions on Internet of Things.”. Which year was it published? What was the volume / issue / page number for this?
This is extremely unprofessional from the authors’ part and frustrating and irritating towards the readers.
The methods section is vague. Each of the components depicted within Figure 3 needs to be properly described in text. I urge the authors to add a Pseudocode style algorithm to demonstrate the methodology.
Section 4.2 should reference the location of the dataset. Line 254 says, “After image preprocessing, 340 representative images were selected for model development, divided into 238 images for the training set and 68 for the testing set.” These sorts of statements are vague. For research reproducibility, the authors need to be clear and precise. What are the image preprocessing steps performed? Be specific, so that other researchers can follow these steps.
Conclusion section should address the limitation and future avenues of research endeavor in this area.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe authors present the article entitled “Image recognition and classification of farmland pests based on improved Yolox-tiny algorithm”
The article presents the following concerns:
-
I suggest listing the contributions of the manuscript after line 59.
-
It is suggested to present figures 8 and 9 as subplots in a single figure for a more efficient and comparative representation.
-
I suggest adding the precision and recall curves to get a complete view of the model performance.
-
Section 4: I suggest adding a table that compares the main contributions of the work vs the already reported in the state of the art in order to highlight the contributions of the work.
-
Please add a confusion matrix to better understand how the model behaves in different classes.
-
It would be advisable to include an image that illustrates how the algorithm was able to properly classify a specific type of pest.
-
Please mention possible future work in the conclusion section.
-
Please vectorize Figures in order to see details.
-
This kind of text must be written in the third person or the passive voice.
-
Add hyperlinks to tables, and figures.
-
I suggest reviewing the format of references, as some do not present important information, such as the year of publication.
The following misspellings should be checked:
-
line 208: The word “clear” is often overused. Consider using a more specific synonym to improve the sharpness of your writing: “more precise”.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe quality of the paper has been improved. Several of my concerns have been adequately addressed such as emphasizing the contributions of the paper and analysis of the precision-recall curve. Regarding feature fusion, a theoretical discussion of this subject considered in the proposed method is required. I understand, implementation and comparison of late fusion methods could be challenging. However, as commented in my previous review, at least a general theoretical discussion, including adequate bibliographic references should be added to the paper, e.g., in Section 2.2. I think this should be more than feasible considering that some points required in my review were not addressed (statistical significance analysis, adding another case of study, and adding a computational burden analysis). The reasoning for not addressing the latter issues was straightforward: “due to time constraints”. In addition, I agree implementation of other fusion capabilities might be taken into account for future research work. Please state that in the paper.
In summary, the issues considered above should be addressed in a revised version of the paper.
Author Response
Comments 1: Regarding feature fusion, a theoretical discussion of this subject considered in the proposed method is required. I understand, implementation and comparison of late fusion methods could be challenging. However, as commented in my previous review, at least a general theoretical discussion, including adequate bibliographic references should be added to the paper, e.g., in Section 2.2.
Response 1: Agree. We have added a theoretical discussion on feature fusion in Section 2.2 and included appropriate references. (page number:4; line:133.)
Comments 2: I think this should be more than feasible considering that some points required in my review were not addressed (statistical significance analysis, adding another case of study, and adding a computational burden analysis). The reasoning for not addressing the latter issues was straightforward: “due to time constraints”. In addition, I agree implementation of other fusion capabilities might be taken into account for future research work. Please state that in the paper.
Response 2: Agree. We have addressed these points in the paper. (page number:14; line:372.)
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors have addressed most of my previous concerns. I don't have any further comments to add.
Author Response
Comments 1: The authors have addressed most of my previous concerns. I don't have any further comments to add.
Response 1: We are pleased to hear that we have addressed most of your previous concerns. We would like to express our sincere gratitude for your thorough review and constructive feedback on our manuscript.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe authors addressed my concerns. However, I recommend to vectorize Figures 1,2,3,9 and 10 before publication. After this, the manuscript will be ready por publication.
Author Response
Comments 1: The authors addressed my concerns. However, I recommend to vectorize Figures 1,2,3,9 and 10 before publication. After this, the manuscript will be ready por publication.
Response 1: Agree. We apologize for the oversight in the first review, where we did not fully implement your suggestion to vectorize the figures. We have now vectorized Figures 1, 2, 3, 9, and 10 as per your recommendation.
