Molecular-Clump Detection Based on an Improved YOLOv5 Joint Density Peak Clustering
, Xiang-Yun Zeng 1,3, Xiao-Yu Luo 1,3 and Chen Long 1,3Round 1
Reviewer 1 Report
Comments and Suggestions for Authors1． The description of the data is unclear to me: what is a data sample in the paper? It’s like Fig.2(a) or Fig.3(a)? How many dimensions does it have? What is the meaning for each dimension? In 2.2 it says ‘Three typical regions’, so what does ‘typical’ mean here?
2． How big is the dataset? The image number and the clump number?
3． One advantage that MCD-YOLOv5 joint DPC has is a two-parameter configuration, which is less than FellWalker and Clumpfind, as shown in Table 5-7. However, how should the parameters in Table 7 be selected in practical applications? The authors need to provide additional instructions on how to set the parameters.
4． Fig. 12/17 shows the output of the detection results by MCD-YOLOv5 joint DPC on a velocity-integrated intensity map. It does show that some targets have been detected. It is suggested that the authors add evidence that these targets are indeed clumps.
5． Fig. 2 and 3 give the selection ranges and integral maps for the background region. Why were such 3 regions chosen for the experiment and what are their peculiarities in distribution?
6． Fig. 13 shows the PSNR and flux versus recall rate. It is recommended that the authors add a description of how these indicators were quantified from the detection results.
7． The method has a low percentage of missed clumps, but there are still some missed ones in Fig. 10 and 12. The possible factors of missed detections can be analyzed additionally to give the user a reference.
8． Fig.14 gave some overlapped samples, how does the method handle those samples?
9． Sometimes in one image, the sizes of clumps may vary a lot. What is the effect for the method in this situation?
10. Fig.4 needs more detailed explaination.
Comments on the Quality of English LanguageToo much 'we' is used in the paper.
Writing skills should be enhanced: sentences are not so fluent in most of the paragraphs. Try to keep the same subjects for the sentences, or the object of the 1st sentence should be the subject of the 2nd sentence.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors present a method of a clump-detection based on an improved YOLOv5 joint Density Peak Clustering applied for the automatic searches of the molecular clouds. After the localizing the clumps in the Galactic coordinates with the YOLOv5, these clumps are clustered in the 3D space applying the Density Peak Clustering to combine the spatial information with velocity data. The proposed method uses the deep learning for the detection of the clumps by marking of the features for the searching. It was shown the ability of the YOLOv5 method enhanced with Coordinate Attention module and Normalized Wasserstein Distance function to localize the small clumps.
Authors report high detection efficiency of the presented method using fewer initial parameters been applied both to simulated data and to experimental observation data from Milky Way Imaging Scroll Painting project.
The precise recognition of the structures of the molecular clouds for the its further analysis are of high importance for the deeper understanding of star formation in the Milky Way, thus the presented manuscript deserves publication.
However, the following points require clarification before the publication.
It would be helpful if the authors could provide in the manuscript separately the percentage of false recognition of clumps caused by both the set of limitations of the proposed target-detection network method and limitations due to the fluctuations in the experimental.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe font size in Fig.2, Fig.3, Fig.13, Fig.16, and Fig.19 should be larger.
In Fig.6, why do the 3 'Head' have different font sizes? And why does it has 4 branches? What are the relationships among them?
Eq.(11), what does the last ',' (after P(r)) mean?
Line 444, 'with the traditional parameter-tuning algorithm', it should be some ... algorithms. Otherwise please show out all the relevant algorithms' names here.
Line 445, 'find the region', should be find the 'locations' or 'positions'.
Line 447, 'optimizing the quality and quantity', 'quantity' cannot be optimized. This word could be deleted.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors revised the manuscript by adding the additional information about the detection result of the second-Galactic-quadrant synthesized data, providing some details about the errors of the method.
I think the revised manuscript can be recommended for publication
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
