Next Article in Journal
Enhancing Farmers’ Capacity for Sustainable Management of Cassava Mosaic Disease in Côte d’Ivoire
Next Article in Special Issue
Improved YOLO-Goose-Based Method for Individual Identification of Lion-Head Geese and Egg Matching: Methods and Experimental Study
Previous Article in Journal
Digitalization’s Role in Shaping Sustainable Agriculture—Evidence from Chinese Provincial Panel Data Using the Baidu Index
Previous Article in Special Issue
Research on Video Behavior Detection and Analysis Model for Sow Estrus Cycle Based on Deep Learning
 
 
Article
Peer-Review Record

Detecting Lameness in Dairy Cows Based on Gait Feature Mapping and Attention Mechanisms

Agriculture 2025, 15(12), 1276; https://doi.org/10.3390/agriculture15121276
by Xi Kang 1,*, Junjie Liang 1, Qian Li 2 and Gang Liu 2
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Agriculture 2025, 15(12), 1276; https://doi.org/10.3390/agriculture15121276
Submission received: 6 May 2025 / Revised: 5 June 2025 / Accepted: 11 June 2025 / Published: 13 June 2025
(This article belongs to the Special Issue Computer Vision Analysis Applied to Farm Animals)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is an interesting approach to deal with lameness detection that could be further used for detecting important welfare and health issues on farm. The authors compare their model to existing detection methods, making the study rather complete and exhaustive. 

The manuscript is generally well written and explained. However, the jargon used makes it difficult and challenging to follow the manuscript throughout, in particular for readers not accustomed to advanced image analyses. I would invite the authors to rephrase parts of the manuscript at their discretion to facilitate the reading. 

One big limitation of this study is the limited dataset consisting of cows from one single research facility, one single breed and short collection period. I understand that this is a feasibility study. Nonetheless, please refer to these potential limitations and the need to validate the model using data from a different dataset and the need to retrain the model on a more varied dataset. 

As the setup was done such to ensure consistent video quality and minimize disturbance to normal cow movement, would the proposed method be suitable for a production environment? Are the algorithms capable of extracting features in an environment prone to dust and clustered animals?

Line 64: Please explain “history energy images”. Do you refer to making use of historic images with known features 

Author Response

Response to the Reviewers

Dear Reviewer and Editor,

Thank you very much for your comments on our manuscript, entitled “Lameness detection of dairy cows based on gait feature map and attention mechanism”. Those comments were all valuable and very helpful for revising and improving our paper, as well as for guiding our future research. We have carefully studied all of the comments and have made revisions, which we hope meet with your approval. The following are our point-by-point responses to the reviewers’ comments, and the revised portions of this paper are also indicated by yellow highlights in the article.

 

Reviewer: The manuscript is generally well written and explained. However, the jargon used makes it difficult and challenging to follow the manuscript throughout, in particular for readers not accustomed to advanced image analyses. I would invite the authors to rephrase parts of the manuscript at their discretion to facilitate the reading.

Response: We sincerely appreciate the reviewer's constructive feedback regarding the technical terminology in our manuscript. We have carefully addressed this concern through the following improvements:1) Added brief explanations for specialized terms upon their first appearance(e.g., "Segment Anything Model (SAM), a state-of-the-art image segmentation algorithm"). 2) Replaced overly technical phrases with more accessible language where possible, while preserving scientific accuracy(e.g., "Combining multiple movement indicators helps overcome these challenges”)

Reviewer: One big limitation of this study is the limited dataset consisting of cows from one single research facility, one single breed and short collection period. I understand that this is a feasibility study. Nonetheless, please refer to these potential limitations and the need to validate the model using data from a different dataset and the need to retrain the model on a more varied dataset. 

Response: Thank you for your noting the problems. Indeed, different datasets and backgrounds may have an impact on the algorithm. Since our algorithm first localizes the cows, and the features of cows in the images are quite distinct, I believe that under normal circumstances, the object detection for cows should perform well. If we develop a system in the future, we will also consider building it in less complex backgrounds.

Regarding the issue of limited data diversity, we have supplemented the discussion section of the paper with validation tests using cow data collected from another farm, and the results were satisfactory. Unfortunately, due to experimental constraints, our dataset did not include cows of other breeds. This limitation has also been noted in the section of the discussion. Page 12.

Reviewer: As the setup was done such to ensure consistent video quality and minimize disturbance to normal cow movement, would the proposed method be suitable for a production environment? Are the algorithms capable of extracting features in an environment prone to dust and clustered animals?

Response: Regarding the suitability for production environments, we acknowledge that our current approach requires cows to walk sequentially through the passageway, which may limit its direct application in free-moving herds. However, as the reviewer astutely noted, modern dairy farms increasingly incorporate controlled passageways in their infrastructure (for milking, weighing, and health checks), making our method highly suitable for these standardized production settings. This design choice was intentional to ensure consistent data quality while minimizing stress on the animals.

Concerning environmental robustness, we fully agree that dust and animal clustering pose significant challenges. Therefore, when constructing the detection system in the follow-up, we will give priority consideration to this issue. Thank you for your reminder as well.

Reviewer: Line 64: Please explain “history energy images”. Do you refer to making use of historic images with known features 

Response: We sincerely appreciate the reviewer's request for clarification regarding the term "history energy images.", we have added the explain “Superimposition of multi-frame binarized images of a walking cow”, page 2

Reviewer 2 Report

Comments and Suggestions for Authors

Congratulations on your study! This manuscript presents a lameness detection algorithm based on gait feature map and attention mechanism. The manuscript is well-written, demonstrating significant innovation and research value. The content is comprehensive, the structure is well-organized, and the presentation is clear and easy to follow. Below are some minor suggestions for improvement.

 

  1. The paper is mostly written in the present tense: it must be written in the past tense as far as possible
  2. The current Introduction is somewhat lengthy, and some background descriptions are overly detailed
  3. The "Specificity" column in Table 4 shows misaligned data, check formatting consistency across all tables. Ensure headers and data cells are properly aligned both vertically and horizontally

Author Response

Response to the Reviewers

Dear Reviewer and Editor,

Thank you very much for your comments on our manuscript, entitled “Lameness detection of dairy cows based on gait feature map and attention mechanism”. Those comments were all valuable and very helpful for revising and improving our paper, as well as for guiding our future research. We have carefully studied all of the comments and have made revisions, which we hope meet with your approval. The following are our point-by-point responses to the reviewers’ comments, and the revised portions of this paper are also indicated by yellow highlights in the article.

 

Reviewer: The paper is mostly written in the present tense: it must be written in the past tense as far as possible

Response: Thank you for your recognition of our work and noting the problems. We have carefully reviewed the manuscript and revised the entire text to ensure consistency in using the past tense, as suggested by the reviewer. All relevant sections have been updated accordingly.

Reviewer: The current Introduction is somewhat lengthy, and some background descriptions are overly detailed

Response: Thank you for your noting the problems. We have streamlined the Introduction section. Page 2.

Reviewer: The "Specificity" column in Table 4 shows misaligned data, check formatting consistency across all tables. Ensure headers and data cells are properly aligned both vertically and horizontally

Response: We have carefully reviewed and corrected the formatting of Table 4 to ensure proper alignment of the "Specificity" column and all other data cells. Consistency in headers and layout has been verified across all tables in the manuscript. Page 11.

Reviewer 3 Report

Comments and Suggestions for Authors

Please mark what has been adjusted in the text so that the reviewer can observe the changes.

 

 

Lameness detection of dairy cows based on gait feature map 2 and attention mechanism

Keywords: dairy cattle; lameness detection; computer vision; deep learning; precision 28 livestock farming

There are keywords in the title. These should appear in only one of these places.

The work does not have a defined objective. At the end of the introduction, the objective of this work must be stated.....

 

One question. When the animals passed through the observatory, was the camera on the whole time? Or was there a device that would turn the camera on when the animal entered the enclosure?

In “Figure. 5. Dairy cows hoof track processing.” a broader and more explanatory title is needed. It is not clear from the title of the figure what it represents.

For figures 5 and 8 you need a legend informing what each line of different colors means.

The conclusion seems more like a discussion. It should be redone, and the authors should make a robust and direct conclusion showing, based on what was done, the contributions of the work. Do not present results in the conclusion, nor say that more work needs to be done. That is not a conclusion!

 

 

Author Response

Response to the Reviewers

Dear Reviewer and Editor,

Thank you very much for your comments on our manuscript, entitled “Lameness detection of dairy cows based on gait feature map and attention mechanism”. Those comments were all valuable and very helpful for revising and improving our paper, as well as for guiding our future research. We have carefully studied all of the comments and have made revisions, which we hope meet with your approval. The following are our point-by-point responses to the reviewers’ comments, and the revised portions of this paper are also indicated by yellow highlights in the article.

 

Reviewer: There are keywords in the title. These should appear in only one of these places.

Response: Thank you for your noting the problems. Based on your suggestion, we have removed the keywords "dairy cattle; lameness detection" to avoid redundancy with the paper's title. Page 1.

Reviewer: The work does not have a defined objective. At the end of the introduction, the objective of this work must be stated.....

Response: Thank you for your valuable suggestion. As recommended, we have added a clear statement of the research objectives at the end of the Introduction section. Page 3.

Reviewer: One question. When the animals passed through the observatory, was the camera on the whole time? Or was there a device that would turn the camera on when the animal entered the enclosure?

Response: Thank you very much for your insightful questions and close attention. In our study, cows are milked two times a daytime. During data collection, we turn on the camera when the cows are being milked and ensure that personnel stay away to prevent the cows from experiencing stress due to the presence of humans when they exit the milking parlor. It should be clarified that when the algorithm proposed in this paper is applied to practical engineering scenarios, it is necessary to comprehensively consider the system's energy consumption and the specific requirements of the farm, so as to determine whether to keep the camera continuously on or activate it manually and remotely, ensuring that the algorithm is applied efficiently and meets the actual farming conditions.

Reviewer: In “Figure. 5. Dairy cows hoof track processing.” a broader and more explanatory title is needed. It is not clear from the title of the figure what it represents.

Response: Thank you for your valuable suggestion. As recommended, we have changed the title to “Processing pipeline for dairy cow hoof trajectory extraction and smoothing”. Page 6.

Reviewer:For figures 5 and 8 you need a legend informing what each line of different colors means.

Response: Thank you for your valuable suggestion regarding the figure legends. We have carefully revised Figures 5 and 8 by adding clear legends to distinguish the different colored lines. Page 6 and 8.

Reviewer:The conclusion seems more like a discussion. It should be redone, and the authors should make a robust and direct conclusion showing, based on what was done, the contributions of the work. Do not present results in the conclusion, nor say that more work needs to be done. That is not a conclusion!

Response: We sincerely appreciate your constructive feedback regarding the Conclusion section. As suggested, we have completely rewritten the Conclusion, avoiding discussion of results or future work. The revised version concisely states the key findings and their significance based on what was accomplished in this research. Page 13.

Back to TopTop