Automated Broiler Mobility Evaluation Through DL and ML Models: An Alternative Approach to Manual Gait Assessment
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper proposes an automated system to evaluate broiler chicken mobility and gait as an alternative to manual scoring methods. The system uses a computer vision technique called YOLOv8 to detect and track individual color-coded chickens in videos. Several mobility indicators are extracted, such as distance traveled, speed, idleness ratio, and time spent at feeder/drinker. Machine learning models are then developed to correlate these mobility indicators with expert-assessed gait scores.
Some recommendations to improve the paper's quality:
- In the introduction, the authors need to expand a bit more on the limitations of current manual methods for gait scoring to highlight the need for an automated approach.
- Provide more details on the camera specifications, video resolution, frame rate, etc.
- Explain how the sample size of 10 broilers was determined.
- Explain why only 4 times of day were used for training data and how this dataset was validated.
- Provide more rationale for the specific YOLOv8 model selected out of available options.
- Explain how optimal model hyperparameters were selected during training.
- Provide details on how manual gait scoring rubric was standardized between assessments.
- Report quantitative evaluation metrics for the trained YOLOv8 model like precision, recall, F1-score, etc.
- Perform some statistical significance testing between gait scores and mobility indicators.
- Explain the rationale for selecting final variables used in the machine learning models.
- Compare performance of ML models using metrics like RMSE, R-squared, etc.
- Assess the generalization error of the models with cross-validation.
The English could be improved to more clearly express the research.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors- In line 115, the author mentions that "the scale was determined to be 1.7 mm per pixel", but this approximation ignores the distortion of the image, and pixels farther away from the camera should have a larger distance.
- The authors need to explain the advantages of using image recognition to extract features rather than fitting a pedometer to each chicken.
- Why did the author only select three classic machine learning models, RF, SVM and OL, in Section 3.5? Would lightGBM or deep learning methods produce better results?
- In addition, the author's method of marking chickens by painting them does not seem to be applicable to chicken farms, as such a large number of color combinations are needed to distinguish thousands of chickens in a chicken farm.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis article combines the YOLOv8 deep learning model with various machine learning methods to achieve automated evaluation of broiler gait score (Gait Score), and explores its correlation with motion indicators, which has strong innovation and application prospects. The article has a complete structure, clear research ideas, sufficient data support, and persuasive results, especially in terms of experimental design and model validation, demonstrating strong rigor. However, there is still room for improvement in certain aspects of the paper. Here are my opinions and suggestions for improvement:
1. Some of the cited literature is outdated or incomplete, failing to cover the latest research findings in the field. Suggest the author to update the literature review and cite research results from the past three years to ensure the forward-looking and academic value of the paper.
2. The experimental scale of this study is relatively small and lacks discussion. The author only selected 10 broiler chickens as samples and conducted the experiment in a small laboratory environment (1.1m×1.5m). Suggest adding a discussion and outlook on the challenges posed by large-scale applications and changes in population density in the discussion section.
3. As mentioned in section 2.2, the study hired experts to evaluate the GS scores of broiler chickens, which mainly depends on the subjective experience of the experts and may result in false positives. Is there a standardized evaluation method?
4. The author mentioned in lines 157-158 that, “In addition to the color codes mentioned, the broilers from surrounding pens were labelled as ‘unknowns’ to lessen confusion by the model.”, Can the camera's perspective be adjusted so that only the research object appears in the video, without the need to label individuals in the 'unknown' category, reducing the additional burden on the model?
5. How to obtain "the (x, y) coordinates of the brothers at each second 176 were obtained? How to define the position of a broiler?
6. In 3.5, the R2 value of the RF ML model is 0.62, but its misclassification rate is 0.35. This data does not demonstrate the superiority of the proposed solution.
7. Suggest setting up a separate discussion section to analyze research methods and results, and summarize the shortcomings of the study.
8. Other:
(1) Simplified formulas (2)~(8).
(2) The analysis of YOLOv8 detection results is not thorough enough, and there are not enough visualization results.
(3) Figure 4 is not clear enough.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsI have reviewed the revised manuscript and the authors' responses to the comments from the initial review. I am pleased to see that all the questions and concerns raised have been thoroughly addressed in their revisions. The authors have made substantial improvements to the manuscript.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe author provided a reasonable explanation of the proposed opinions and there are no further issues that need to be addressed.
Reviewer 3 Report
Comments and Suggestions for AuthorsThe manuscript has beens sufficiently improved.
With my best wishes.