Research on Automatic Recognition of Dairy Cow Daily Behaviors Based on Deep Learning

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Yu et al.'s manuscript evaluates the application of computer vision in identifying different cow behaviors within a farm setting. The use of computer-based image processing represents an emerging field with promising future implications. This study holds significant relevance for researchers engaged in precision livestock farming. The manuscript was well-structured, offering sufficient information to implement the model used in the study. For improved clarity, consider using the past tense appropriately when describing the materials and methods, presenting results, and discussing the findings. Below are a few minor suggestions that could further enhance the content.

Line 20: Consider rephrase the sentence for better clarity.

The introduction was well-organized in establishing the study's necessity. To further enhance it, consider incorporating the study's objectives within the concluding paragraph of the introduction.

Please consider adding references of figures within the text when discussing the comparison of models in subsections 3.3.2 and 3.3.4.

In livestock farming, insights into individual animals hold greater importance. The manuscript should incorporate recommendations for potential solutions to enable individual identification of cows, not just the group behavior. Consult this article for reference please (Kawagoe et al. (2023); https://doi.org/10.3390/agriculture13051016). It's advisable to propose strategies for individual animal identification. 

 

Comments on the Quality of English Language

To enhance clarity, ensure the appropriate use of the past tense when describing the materials and methods, presenting results, and discussing the findings.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

In this article, the authors develop an improved model based on YOLOv5, a real-time object detection framework, incorporating the CoordAtt attention mechanism and the SIoU loss function, and named Res-DenseYOLO. Finally, it is concluded that Res-DenseYOLO outperforms the baseline YOLOv5 model in terms of precision, recall, and mAP by 0.7%, 4.2%, and 3.7%. It is possible to monitor the behavior of cows in real-time through video only. However, the readability of the manuscript can also be greatly improved. Through editing and some modifications, I think this manuscript will be more suitable for publication.

1. In Abstract section，

-the writing's logical order was not smooth, the author should first explain the model name and then introduce its function and characteristics.

- Lines 28-29, the results are described very subjectively without objective conclusions. Such as "effective" and "accurate". Suggest the author supplement objective conclusions in this section.

-Suggest the author to rewrite the Abstract. A logically clear abstract is essential for readers to better understand the entire research content.

2. Lines 28-29, "......outperforming the baseline YOLOv5 model by 0.7%, 4.2% and 3.7% respectively." From the data, the results of the improved model are not very significant. Is the model improvement still meaningful?

3. Line 342 and 392. It is suggested that the number of digits should be kept uniformly, 89 should be changed to 89.0, and other numbers should also be unified.

4. The format of the Reference section of this manuscript is very confusing. Please make revisions according to the journal's reference requirements and standardize the format.

 

Comments on the Quality of English Language

Minor editing of English language required

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript proposes an improved model, Res-DenseYOLO, based on YOLOv5 for the automatic recognition of dairy cow daily behaviors. The study aims to overcome inefficiencies in manual observation and potential stress caused by wearable sensors. The model is evaluated on a dataset collected from monitoring videos of a dairy farm. The paper addresses a relevant issue in agricultural technology, and the proposed model seems promising. However, some improvements and clarifications are suggested before it’s accepted for publication.

Provide a more comprehensive introduction, discussing the importance of monitoring dairy cow behavior, existing challenges, and the need for automated systems. Additionally, consider introducing more relevant on the application of lightweight CNN networks for detection tasks, such as: https://doi.org/10.1007/s00170-022-10335-8

Elaborate on the integration of the dense module into the YOLOv5 backbone network. Provide more details about the dataset, including the diversity of behaviors captured, duration of monitoring, and any challenges specific to the farm environment. Clarify whether any transfer learning was performed using pre-trained YOLOv5 weights and how the model was initialized.

Interpret the results in the context of the study objectives. Discuss the practical implications of achieving high precision and recall in dairy cow behavior recognition. Address potential limitations and challenges faced during the development and evaluation of Res-DenseYOLO.

Overall, the paper addresses an important problem in dairy farming, and the proposed model shows promise. Enhancements in detail, result presentation, and discussion will strengthen the manuscript.

Comments on the Quality of English Language

The language and style of the manuscript are generally clear and concise.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

A brief summary/ General concept comment:

 

The objective of the article is to obtain an object detection model for the precise identification of cow behavior in an intricate dairy farming environment. It is not clear enough what gap the authors want to fill with their deep learning model, and it should be better specified what is meant by “intricate” environment. The paper is scientifically sound and very detailed in the Introduction and Materials and Methods sections. It is a bit confusing in the Results and Discussion section, where the organization of the subheadings should be improved.

There have been other studies on cow behavior detection on farms (see some papers below) and the authors should take them into account when discussing and comparing their results. The results do not specify the size of the model and the speed of detection, although the conclusions state that the authors will investigate techniques to reduce them.

 

- Ma, S., Zhang, Q., Li, T., & Song, H. (2022). Basic motion behavior recognition of single dairy cow based on improved Rexnet 3D network. Computers and Electronics in Agriculture, 194, 106772.

 

- Tassinari, P., Bovo, M., Benni, S., Franzoni, S., Poggi, M., Mammi, L. M. E., ... & Torreggiani, D. (2021). A computer vision approach based on deep learning for the detection of dairy cows in free stall barn. Computers and Electronics in Agriculture, 182, 106030.

 

Specific comments:

Line 118: What do the authors mean by "high quality"? What is the parameter considered?

Line 119: More information on the 90 cows selected for the study should be added, e.g. age, production status, etc.

Line 123: What is the model of the cameras used. It is better to specify. Also, more details about the position of the camera should be provided, i.e. the height from the ground.

Line 124: Add the resolution unit.

Line 126: Are the infrared images taken by the same Hikvision cameras?

Lines 129-130: Figure 1 should be improved. The authors should add the measurement units in the dimension indicators, could highlight the areas covered by each camera also with different colors for each behavior areas. In addition, it is not very clear where the lying and eating areas are on the map.

Line 141: The authors want to analyze a complex environment, but they analyze a part of it due to the inclination of the cameras and therefore the images that would not allow them to identify the cows in the opposite part of the barn. In fact, with black masking, you cut off half of the barn.

Lines 252-253: This figure is not self-explanatory within its caption. The authors should specify, for example, what the yellow and blue boxes are.

Lines 295-322: Subheadings 3.1 and 3.2 are more appropriate for the Materials and Methods sections. Consider moving these subheadings to this section.

 Line 309: How many samples were used in the training phase, and what is the final learning rate?

Line 338: The unit of measurement of the performance parameters must be given in the table.

Line 342: The unit of measurement of the performance parameters must be specified in the table.

Line 358: The unit of measurement of the performance parameters must be specified in the table. 

Line 385: Table 6 is not very clear. What does the checkmark mean?

Line 515: The authors what to "achieve precise identification........ in intricate ..environment". However, if “intricate” refers to the number of animals present in the barn, this does not seem to be such an intricate and complex environment, at least considering Figures 12, 11, 13, and 14, where the maximum number of cows per image is 7.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx