Pig Face Recognition Based on Metric Learning by Combining a Residual Network and Attention Mechanism
Round 1
Reviewer 1 Report
The paper presents a pig face recognition system based on Metric Learning Combining Residual Network and Attention Mechanism. The paper is well-written and organized, deals with an interesting topic, and the experimental analysis is well-conducted. However, some points should be considered to improve its quality further.
- Acronyms should be defined in their first apparition, and some acronyms are not defined, e.g., RFID, FAST, SIFT, FLANN,….
- The paper’s organization should be added in the last part of the introduction.
- In the introduction section, the main contribution should be rewritten as points to facilitate its comprehension.
- It is recommended to add a “Related Work” section that includes the similar works discussed in the introduction and groups them into several classes.
- The manuscript requires proofreading and revision because it contains several mistakes.
- A comparison to some recently published papers under the same conditions (i.e., dataset + evaluation protocol) is necessary.
- Authors should improve the future work section, and it is recommended in future work to test the deep unsupervised active learning to improve the quality of learning and render it more semantic (10.1109/JSEN.2021.3100151).
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This study establishes an improved ResNAM network as a backbone network for pig face image feature extraction by combining the NAM (Normalization-based Attention Module) attention mechanism and the ResNet model to probe non-contact open-set pig face recognition. And the recognition accuracy reaches 95.28%, which is 2.61% higher than that of the human face recognition model. The paper presents a good topic, however, the paper can be improved further as shown below:
1. The expression of "normalized attention module (NAM)" in Line134 and "is made by combining normalized attention mechanism (NAM)" in Line126 are repeated.
2. Line134-135: " normalized attention module (NAM) to reduce the loss of pig face features is reduced by using jump joins. " does not sound right.
3. Line153: " the 224 pixel × 224 pixel "should be revised to " pixels ".
4. Line169:" Euclidean distance" should be revised to "euclidean distance".
5. Line190:" The model’s accuracy rate" should be revised to "The recognition accuracy of the model".
6. Line231-242: " Table3 demonstrates how several attention processes, ……, in the same backbone network to construct Models 1-3. ". These models are named No. 1-3 in Table 3 and Models 1-3 in the paper. Please express it uniformly.
7. Generally, avoid using abbreviations without defining them in the text. For example, BAM and CBAM is not defined in the text.
8. In Table5, it should be "model weight" instead of "parameters".
9. Line365: The abbreviation of HRNet is not used in the paper.
10. Line292: "pig No. 4" or "pig4"?
11. You should explain, in the text, the meaning of parameter Bin in Equation (1).
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Non additional comments. The paper can be accepted in the current form.
