Goat-Face Recognition in Natural Environments Using the Improved YOLOv4 Algorithm

Round 1

Reviewer 1 Report

The overall structure of the paper is good

The study of related methods should be increased. Authors have studied minimal papers. Create a separate section of related works after the introduction. Structure/organization of content should also be provided at the end of introduction section

What if there are multiple goats in the same frame? Is the model robust enough to handle that scenario?

Comparison with existing state-of-the-art methods should be shown on standard goat datasets.

Justification for choosing deep learning models over ML models should be mentioned. What are the limitations of ML models on goat recognition? How do Deep models overcome that? 

Author Response

Thank you for review,please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

1.     The first line in abstract "In view of the low accuracy and slow speed of goat face recognition in the real breeding 16 environment, the dairy goat was taken as the research object, and video frames were used as data 17 source" is broken into two lines. 

2.     The authors don’t explain the difference between Yolov4 and improved YOLOV4 for goat face recognition. Have the authors improve the YOLOV4 accuracy in improved YOLOV4.

3.     Why the authors uses CSPDarkNet53 as feature extraction in improved YOLOV4 module. There are lots of pretrained models such as resnet 50/100 for feature extraction in improved YOLOv4 model.

4.     The following papers could be included as supporting literature those are related to machine learning, ROI, and image processing.

·         Recognizing Wheat Aphid Disease Using a Novel Parallel Real-Time Technique Based on Mask Scoring RCNN

·         Image-Based Wheat Mosaic Virus Detection with Mask-RCNN Model

·         Quantifying the Severity of Loose Smut in Wheat Using MRCNN

·         Deep learning in wheat diseases classification: A systematic review

·         Automatic Classification of Wheat Rust Diseases Using Deep Convolutional Neural Networks

·         An Instance Segmentation Approach for Wheat Yellow Rust Disease Recognition

5.     The authors did not indicate the gap and their contribution compared to the existing literature. The main question here would be why the new method is needed.

6.     Conceptual illustration along with a statistical illustration also needs to add to the proposed objective those have provided in the initial section of the paper.

7.     All images used in the figures should be enhanced. All are too blurry.

8.     There is no comparison of proposed approach with previous methods in terms of their performance in results and discussion section.

Author Response

Thank you for your review,please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The paper introduces a new improved model from YOLOv4 for recognizing goat faces by establishing comparisons between these models.

The described model introduces important improvements and is of great interest in the areas of facial recognition.

The document is well organized and described.

Some suggestions to improve the document:

- Correct the hyphenation of words throughout the text.

- line 202 has an extra space ("Figure. 6")

- Equation 1 should be improved. It has mathematical parameters with different sizes.

- Mathematical equations entered in the text between lines 217 and 222 are out of line.

- The CIOU function appears in the text and is only described further on. It must be described before it is spoken.

- Equations (4) and (5) are the same.

- In line 228 an alpha parameter is mentioned that does not appear in the equations.

- In line 254 when AP is described, its equation must be placed in parentheses. In order to guide the reader. Same for mAP on line 256.

- The description of metrics should be reformulated. It's very confused. It should first be described how Accuracy and Recall are calculated, defining what is TP, FP, FN before talking about AP and mAP.

- In lines 264-266 the meaning of TN is described, but it does not appear in the equations!

- In table 1 the entries of each column must all have the same number of decimal places. For example in column 6 (Flops/G) all values ​​must have 2 decimal places.

- Line 295-296 refers to facial recognition of goats 9, 11, 13, 22, using different methods. But it is not described that goat is on line 1, goat 11 on line 2, etc.

- In figure 7 it is not necessary to always place the captions under the images. All that was needed was a top caption and all the images aligned, like a table.

- The Model on the last line of table 2 can be written all in a single line. The table looks better.

- In the text before Figure.9 it is not clear which goats are represented in the Figure. It appears to be goat13 on line 1, on line 2 it is unknown and on line 3 goat9. But here in the text we talk about goat17, goat21 and goat20. Where are they represented? and goat9 is not described in the text.

- Check the reference 2 pages, it seems to me that they are not well.

 

Author Response

Thank you for your review,please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Grammatical corrections are applied

There should be an independent Literature review section with more in-depth study. Th study seems very shallow

Structure/organization of content should also be provided at the end of the introduction section

 

Author Response

Thank you for your review, please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The number of goat masks images along with their IOU is missing. Number of articles that are related to your work is missing. Please follow the following papers and cite that references in your paper:

 1. Quantifying the Severity of Loose Smut in Wheat Using MRCNN

2. Recognizing Wheat Aphid Disease Using a Novel Parallel Real-Time Technique Based on Mask Scoring RCNN

3. Detection of DoS attacks using machine learning techniques

4. N-CNN Based Transfer Learning Method for Classification of Powdery Mildew Wheat Disease

Along with your paper, future work is missing in paper. The limitations of your proposed work is missing. How many images were labeled manually?

What do the numbers mean in Figure 6? The description must be in legend.

Author Response

Thank you for your review, please see the attachment.

Author Response File: Author Response.docx