Round 1

Reviewer 1 Report

Referee report on the manuscript "Identification of shark species based on their dried dorsal fins through image processing" submitted to applied sciences

This is in my opinion a well-executed study on image processing, including neural networks, to recognize shark species based on dry samples from fins. This possiblility can allow better tracking of shark trade for the benefit of saving endangered species. The manuscript is well written and easy and interesting to read, the neural networks are adequately described, and the conclusions are supported by the contents. In my opinion the manuscript can be pulished in Applied Sciences.

Author Response

  We are so grateful for your comments and words. Thank you very much.

Author Response File: Author Response.docx

Reviewer 2 Report

The file with the comments to the authors is attached.

Comments for author File: Comments.pdf

Author Response

Reviewer 2

Comments to the authors

Overview comment:

The authors describe a study devoted to test algorithms to identify dry fins of several shark species participating in shark fin trade from 14 countries, investigating sensitivity and specificity of image processing. They compare a methodology based on a non-linear composite filter using Fourier transform with a new methodology based on a neural network.

Neural network appears to be the best methodology because it supports lower-quality images, very low processing time of the order of seconds and independence on different rotation, illumination, sizes of the images.

The state of art is well described, with several references that allow to enter more in detail of the different mechanisms. I agree with the general logic of the paper, but in my opinion a several revisions addressing the following remarks would be required to accept the manuscript.

 

I think the manuscript would benefit of further elaboration of the description of the methodology in section 2, to better highlight the original contribution. My main concern is that in the present form, the paper appears to be more qualitative than quantitative in description of methods. The equations must be better introduced and commented in detail, and each step of the algorithm should be described to its scope.

More detailed motivations are given in the following, where possible major revisions are listed. In my opinion, a revision addressing the following remarks would be required to accept the manuscript after a further review.

We are so grateful for your comments to improve the article. In the next sections, we attend to all your comments. In red color is our answer to each comment

 

 

Detailed comments:

• I would encourage the authors to better describe the methodology of the two investigated methos and algorithms For example, what software was used? Was it properly developed or is it commercially available? Answer: See Page two, lines 92-94.

It is challenging to understand the logic of the algorithms because the description is too qualitative and general. Please add details both for subsections 2.2 and 2.3.

Each term of the equation and each different step must be described and explained in the text for both approaches.

Answer: We added information to describe step by step. See Section 2.2, page 4 and 5, lines 154-175 and Section 2.3, Page 5 and 6, lines 180-216.

 

• Page 1, line 23: after the term “taking” the sentence “processing times of the order of few”

must be added.

Answer: We added that line, thank you very much.

• Page 1, line 37: replace 1,199 with 199.

Answer: We did.

• Page 2, line 91: the authors write “we provide computer techniques and digital correlation systems”. Specify whether this code has been self-developed, has been properly improved for specific purposes or has been used only and was commercially available. In the latter case, the references must be

Answer: See Page 2, lines 92-94.

• Page 3, line 109-110: “from 14 countries” is repeated 2 times. Please remove

Answer: The second one was removed.

 

• Page 3, line 113-114: delete the sentence “Most filters do not function efficiently when the problem image has small distortions, different sizes, rotations, or illumination.” Because it is the same to the one at the lines 94-95.

Answer: We removed it.

• Page 3, line 115: in this section, it would be interesting to show a typical image that authors analyze, a set of input images,

Answer: See figure 1, 124-149

• Page 3, line 128: insert “on the right” after “step (A)”.

Answer: See Page 4, lines 155-156.

• Page 3, equation (2): in the first term, “dx” is missing in the

Answer: We changed the equation 2 for a better explanation. Please see Page 4 and 5, lines 162-169.

• Page 3, equation (2): if possible, insert the equation in a single

Answer: Done

• Page 3, line 137: please introduce in the text each term of the equation: “where c1, c2 are

respectively…, f(x) g(x) and S are... etc.”

Answer: We changed the equation 2 for a better explanation. Please see Page 4 and 5, lines 162-169.

 

• Page 4, Figure 1: It is not clear to find the correspondence between the sketch shown in the image and the description in the text. Specify the procedure better, describing each term and step and their effects on the analysis. In the figure it should be added each term described in the text, for example in the Input image, author can indicate “In”.

Answer: We added information to describe step by step. See Section 2.2, page 4 and 5, lines 154-175

• Page 4, line154: specify if the code was self-written or is commercially

Answer: See Page 2, lines 92-93

• Page 4, equation 3: “x” and “y” must be defined in the

Answer: See Page 5, line 194-195

• Page 5, equation 4 and 5: describe terms of equation: “where VP and FN are,

respectively…”.

Answer: See Page 6 and 7, lines 226-231.

• Page 5, line 222: what is “k”?

Answer: See Page 5, lines 167-169

• Describe what are the vantages/disadvantages in using different number of neurons in the hidden layer, species, and noise in the images as done in the 4 type of experiments authors Results are clear, but it is missing their interpretation to better understand limits and possible improvements that can be achieved in perspective.

Answer: See Page 16, lines 352-355.

• Page 7, Table 2 and 3: there is a typo: “efficiency”  “efficiency”.

Answer: Done.

• Page 9, lines 256-266: remove the statement “Table 4 shows the sensitivity and specificity of each shark species in the neural net-265 work with 90% efficiency.” because it is a repetition of the previous one at lines 255-256.

Answer: Done.

• Page 11, Discussion section: Are the obtained sensitivity values sufficient for the future objectives? Or are larger samples mandatory to make them reliable? Please add comments in the

Answer: See Page 16, lines 354-355

 

Author Response File: Author Response.docx

Reviewer 3 Report

The methodology proposed in the manuscript might have a great potentiality in supporting custom officers in the identification of shark dried fins, especially in the immediate future, due to upcoming CITES Conference of the Parties and the probable decision to include many more shark species in the CITES Appendices.

The current draft of the manuscript presents some major issues, related to both technical and language. Technical issues: i) in the introductory and result sessions some concepts are anticipated while they would be more appropriate in the discussion section; ii) lack of explanations in the methodology with the consequent difficulty of comprehension of some results and points of discussion. Language issues are related to both grammar and style, resulting in English language often not precise or incorrect, therefore a technical revision is clearly advisable. Moreover, many references are lacking and in one case not related to the description provided.

Some details in the preparation of the manuscript need to be revised:

-          The keywords seem not properly selected, e.g. the first is “filters” having a too generic meaning to be possible linked to the subject. A recall to the subject, “shark fins” and “CITES” is advisable.

-          References do not follow the guidelines, and in particular the use of the author’s names/surnames.

Hereunder the details of the improvements required:

Referring to technical issues:

·       In the methodology the authors do not describe how the samples of dried shark fins, or the photos of the dried fins, have been originally identified: by means of visual identification or by DNA analysis, and eventually the related references. This is fundamental for the reliability of the final identification, when applying the methodology proposed.

·       In the methodology, the authors do not describe the second samples of photos mentioned in the “final experiment (see line 280 referring to 4 438 images of dried dorsal shark fins).

·       In the methodology, the authors do not describe the background characteristics of the photos, white or with noise, mentioned in the results and discussion.

·       179‒183: it is not clear to the reader how the 10% of images were used to validating data and the difference with the 10% used for testing.

·       111‒114: the authors declare that the “digital correlation systems are the best solution for image processing”. This sentence is not substantiated, it anticipates concepts related to the discussion, it is not referenced and finally not demonstrated in this paper.

·       219‒222: (Results) the authors did not describe the two groups of CITES-listed and not listed species and the sentence results not clear. In addition, the methodology seems to applies at species level and not at group level (CITES-listed versus not CITES-listed).

·       290: Tables 4 and 7 do not include the number of samplings by species as it is done in Table 1, reducing the understanding of the results.

·       332‒336: referring to “This indicated that 66% of the images were correctly identified as belonging to Sphyrna zygaena” it is not clear 66% of how many images? Same for “Sphyrna mokarran had sensitivities of 78% and 82% respectively”, how many images and respectively to what?

·       343: (Discussion section) referring to “local binary pattern function”, as well as the rotation of the photos, are not mentioned before. The explanation on this methodology should be added to the methodology section.

·       When talking about the “best methodology” the author should refer to those applied in the study described not in general.

Issues related to references and concepts:

·       65‒66: the software iSharkFin has been developed to detect wet fins, and by definition it should not be used for the identification of dried fins. The sentence should be rephrased.

·       74: citation to reference number 20 seems not appropriate. The cites reference is a field guide not a paper demonstrating that the users have low capability to discriminate between similar species of hammerheads.

·       75‒77: sentence not referenced or eventually it anticipate the discussion, should be modified or moved.

·       209, 214 and 222: definitions missing for VP, FN, VN, FP and k.

·       Tables 2, 3, 5 and 6: missing definition in the methods/text of “Epochs” and “% efficiency”.

·       288‒289: the sentence “This result was better than expected” should be rephrased as the expectation is subject to the experience of the authors and it is not explained further; moreover these type of comments should be moved to the discussion section.

·       311‒325: This paragraph result not clear as the methodology is not explained before. The abbreviations CMF and POF filters should be spelled out as they were not mentioned before.

English language issues have been found at the following points:

·       31: referring to the first sentence saying “The increased human exploitation and habitat deterioration over the last half-century 30 has exposed shark populations worldwide”, the sentence seems not complete (exposed to what?) or the use of the verb “exposed” not proper.

·       44: “At present, CITES has listed forty-six shark and ray species in the Appendices, including 184 countries worldwide”: did he authors intend that 184 countries are CITES Parties?

·       119 and 376: “Shark Conservation Funding” in the text does refer to “Shark Conservation Fund”? In case amend.

·       Overall, the final sentences of the section of Discussion and the whole section of Conclusions should be revised: (351‒357) avoid use of language like “must be provided”; “we would be finished with all shark species”; (351‒357) the concepts expressed, after a revision of the language, would be more appropriate at the end of the Conclusions.

Other minor issues:

The addition of images representing the photos of some shark dorsal fins used in the experiments, with the different background descripted, is advisable.

·       Tables 1, 4 and 7: The order of the species in the list is not clear, it is neither alphabetical nor taxonomical. 

I

 

Author Response

Reviewer 3:

Comments and Suggestions for Authors

 

The methodology proposed in the manuscript might have a great potentiality in supporting custom officers in the identification of shark dried fins, especially in the immediate future, due to upcoming CITES Conference of the Parties and the probable decision to include many more shark species in the CITES Appendices.

 

We are so grateful for your comments to improve the article. In the next sections, We will attend to all your comments. In red is our answer to each comment

 

The current draft of the manuscript presents some major issues, related to both technical and language. Technical issues: i) in the introductory and result sessions some concepts are anticipated while they would be more appropriate in the discussion section; ii) lack of explanations in the methodology with the consequent difficulty of comprehension of some results and points of discussion. Language issues are related to both grammar and style, resulting in English language often not precise or incorrect, therefore a technical revision is clearly advisable. Moreover, many references are lacking and in one case not related to the description provided.

 

Some details in the preparation of the manuscript need to be revised:

 

• The keywords seem not properly selected, e.g. the first is “filters” having a too generic meaning to be possible linked to the A recall to the subject, “shark fins” and “CITES” is advisable.
• Answer: We removed filters and added CITES and shark fins see27

 

• References do not follow the guidelines, and in particular the use of the author’s names/surnames.

 

Hereunder the details of the improvements required:

 

Referring to technical issues:

 

• In the methodology the authors do not describe how the samples of dried shark fins, or the photos of the dried fins, have been originally identified: by means of visual identification or by DNA analysis, and eventually the related references. This is fundamental for the reliability of the final identification, when applying the methodology

Answer: See Page 3, Section 2.1 General information about the image database, lines 129-131

 

• In the methodology, the authors do not describe the second samples of photos mentioned in the “final experiment (see line 280 referring to 4 438 images of dried dorsal shark fins).
• Answer: See Page 3, Section 2.1, lines 120-140

 

• In the methodology, the authors do not describe the background characteristics of the photos, white or with noise, mentioned in the results and

 

• Answer: See Page 3, Section 2.1, lines 136-140

 

 

• 179‒183: it is not clear to the reader how the 10% of images were used to validating data and the difference with the 10% used for
• Answer: See Page 6, lines 212-216

 

• 111‒114: the authors declare that the “digital correlation systems are the best solution for image processing”. This sentence is not substantiated, it anticipates concepts related to the discussion, it is not referenced and finally not demonstrated in this

 

• Answer: We erased this paragraph.

 

 

• 219‒222: (Results) the authors did not describe the two groups of CITES-listed and not listed species and the sentence results not clear. In addition, the methodology seems to applies at species level and not at group level (CITES-listed versus not CITES-listed).
• Answer: See Page 3, lines 120-126

 

• The database includes two groups (CITES-listed and not CITES-listed). The CITES-listed is very important because most of the shark populations are in critical danger, however, there are shark species that are not CITES-listed but they are important as the ones who are CITES-listed. That is why we decided to merge the two groups

 

• 290: Tables 4 and 7 do not include the number of samplings by species as it is done in Table 1, reducing the understanding of the result  

Answer: It was corrected.

 332‒336: referring to “This indicated that 66% of the images were correctly identified as belonging to Sphyrna zygaena” it is not clear 66% of how many images? Same for “Sphyrna mokarran had sensitivities of 78% and 82% respectively”, how many images and respectively to what?

 Answer: See page 16, lines 334-338.

• 343: (Discussion section) referring to “local binary pattern function”, as well as the rotation of the photos, are not mentioned The explanation on this methodology should be added to the methodology section.

 Answer: See Page 3, Section 2.1, Lines 134-135. See Section 2.3

• When talking about the “best methodology” the author should refer to those applied in the study described not in

Answer: See Page 16, line 374.

Issues related to references and concepts:

 

• 65‒66: the software iSharkFin has been developed to detect wet fins, and by definition it should not be used for the identification of dried The sentence should be rephrased.
• Answer: See page 2, lines 69-70

 

• 74: citation to reference number 20 seems not appropriate. The cites reference is a field guide not a paper demonstrating that the users have low capability to discriminate between similar species of

 

• Answer: We don’t understand this comment. We are only saying that it is very difficult for a human to identify two species that are very similar, so before technology, the only way to identify them was through the guides.

 

• 75‒77: sentence not referenced or eventually it anticipate the discussion, should be modified or

 

• Answer: It was removed

 

• 209, 214 and 222: definitions missing for VP, FN, VN, FP and

 

• Answer: See page 6 and 7, lines 226-231, See Page 5, lines 167-171.

 

• Tables 2, 3, 5 and 6: missing definition in the methods/text of “Epochs” and “% efficiency”.

 

• Answer: See page 9, lines 254-257.

 

• 288‒289: the sentence “This result was better than expected” should be rephrased as the expectation is subject to the experience of the authors and it is not explained further; moreover these type of comments should be moved to the discussion section.

 

• Answer: It was removed

 

• 311‒325: This paragraph result not clear as the methodology is not explained before. The abbreviations CMF and POF filters should be spelled out as they were not mentioned

 

• Answer: See Page 15, lines 314-318

 

English language issues have been found at the following points:

 

• 31: referring to the first sentence saying “The increased human exploitation and habitat deterioration over the last half-century 30 has exposed shark populations worldwide”, the sentence seems not complete (exposed to what?) or the use of the verb “exposed” not
• Answer: We changed it to decreased

 

• 44: “At present, CITES has listed forty-six shark and ray species in the Appendices, including 184 countries worldwide”: did he authors intend that 184 countries are CITES Parties?

 

• Answer: The best world to describe that these countries are part of CITES is by participating

 

• 119 and 376: “Shark Conservation Funding” in the text does refer to “Shark Conservation Fund”? In case

 

• Answer: Shark Conservation Fund

 

• Overall, the final sentences of the section of Discussion and the whole section of Conclusions should be revised: (351‒357) avoid use of language like “must be provided”; “we would be finished with all shark species”; (351‒357) the concepts expressed, after a revision of the language, would be more appropriate at the end of the

 

• Answer: We changed it to Conclusions

 

Other minor issues:

The addition of images representing the photos of some shark dorsal fins used in the experiments, with the different background descripted, is advisable.

Answer: See Figures 1 and 2.

 

• Tables 1, 4 and 7: The order of the species in the list is not clear, it is neither alphabetical nor

Answer: We put the species from highest to lowest risk of extinction

 

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have adequately taken into account all the points made in the previous report and implemented all the comments and corrections. The quality of the paper has significantly improved, by implementing a better quantitative description of methods and results. Therefore, I believe it is currently ready for publication.

Author Response

 We are so grateful for your comments and words. Thank you very much. 

Author Response File: Author Response.docx

Reviewer 3 Report

Dear Authors,

Many thanks for the amendments you did to the manuscript. From a second round of revision, I have still some technical points that remain not well addressed to facilitate the comprehension of the methodology and the possibility to replicate the experiments.

1)    (Lines 181‒192) You say that each image has 59 features and that the input layer represents the texture. If the texture is described by 59 features, it would be useful to describe better this relation and in case if/how/why you selected these 59 features.

2)    (136‒139) Relevant for the point below: still not clear whether the dataset of 4 438 was obtained from the original dataset of photos including the background noise, scaling and rotating. Or from a dataset of 4 438 you edited 1029 photos removing the background.

3)    (neural network methodology) You are testing the capacity of the algorithm to distinguish fins independently from the background and the input is the texture of the fin. My concern is that the same photo of fins with different backgrounds can be selected for the training and the testing dataset (from the dataset of 4 438?), therefore a percentage of the efficiency might be due to the matching of the same photos of fin (even if they have different background) and not to the efficiency in the identification of the species. Please clarify this in the methodology and the results.

4)    (neural network experiments) Still it is not clear the background associated with each experiment of the neural network. Please add a clear description in the methodology of the testing you are conducting.

5)    (Results) The Table of the "neural network runs" might be converted in graph facilitating the comprehension? (this is just a suggestion).

6)    (Discussion) When it says: "Sphyrna zygaena 66% of 92 images....This indicates that 66% of the images were correctly identified as belonging to Sphyrna zygaena", from the methodology described, 92 images are the total available so about 74+9 were used for training (80%) +validation (10%) and 9 for testing (10%), therefore the result might be 66% of 9 images (or 5 out of 9) identified correctly.

Other minor points to be considered:

7)    The input layer was the texture vector of the image, the texture is the way something feels when touched, or how it looks caused by the way in which it is woven...is this the proper language to describe the photo of a fin?

8)    109‒110 “to identify the species of origin of 37 dry dorsal fins”: did you mean the 37 species of origin of a testing set of dry dorsal fins?

9)    123-124. "The dry fin sharks’ database was identified": The meaning of this sentence is not clear, did you mean "the dried fins were identified at species level"? How? (I read this is explained below, I suggest to be more consistent and avoid repetition)

10)     127‒129: This explication is generic and not related to the objective of the study. The species level is needed to respect the shape diversity and the possibility to identify a fin at species level depends in this case on the confidence of the expert.

11)     130‒132: “To validate the use of the algorithms, all the shark fin photos were previously identified visually by shark fin identification experts [18, 20], based on their knowledge and published fin field guides”: This is a very sensitive point in the discussion about the capability of the algorithm you are testing to correctly identify the species to which the fins belong. I suggest discussing about the potential bias in the original identification, eventually introduced in the training tests, to be transparent and make the reader aware when evaluating the results.

Language:

12)     Check the use of “dry/dried”

Author Response

 Comments and Suggestions for Authors

Dear Authors,

Many thanks for the amendments you did to the manuscript. From a second round of revision, I have still some technical points that remain not well addressed to facilitate the comprehension of the methodology and the possibility to replicate the experiments.

1)    (Lines 181‒192) You say that each image has 59 features and that the input layer represents the texture. If the texture is described by 59 features, it would be useful to describe better this relation and in case if/how/why you selected these 59 features.

Answer: We are so sorry. Each image has not 59 features. We use the local binary pattern function to obtain a vector, for each image, of 59 elements. Line 185, blue color.

2)    (136‒139) Relevant for the point below: still not clear whether the dataset of 4 438 was obtained from the original dataset of photos including the background noise, scaling and rotating. Or from a dataset of 4 438 you edited 1029 photos removing the background.

Answer: From the second dataset of 4,438 we edited the first dataset of 1029 photos removing the background. Lines 141-142, blue color.

3)    (neural network methodology) You are testing the capacity of the algorithm to distinguish fins independently from the background and the input is the texture of the fin. My concern is that the same photo of fins with different backgrounds can be selected for the training and the testing dataset (from the dataset of 4 438?), therefore a percentage of the efficiency might be due to the matching of the same photos of fin (even if they have different background) and not to the efficiency in the identification of the species. Please clarify this in the methodology and the results.

Answer: The photos of fins with different backgrounds used in the training of the neural network are not the same as those used to perform the validation and testing of the network. Lines 214-216, blue color.

We must to remember, we have two databases for the neural network. Several experiments were performed: a) Table 2: 1029 photos of fins with clean background were used in a network of 10 neurons.

1. b) Table 3: 1029 photos of fins with clean background were used in a network of 20 neurons.
2. c) Table 4: to calculate the sensitivity and specificity we used the 1029 photos.
3. d) Table 5: we used 27 species only with 20 neurons in the network.
4. e) Table 6: Table 6 shows the fourth experiment with 37 species and one control group. Here, we have two neural networks with an 89% efficiency. In this experiment, there was a good percentage because the number of dried shark fins increased for each species.

The final experiment (Table 6) was performed using a database of dried shark-fin images with and without background noise to increase the number of dry fins in each species. From this database, 4,438 images of the dried dorsal shark fins were obtained. Lines 289-294, blue color.

4)    (neural network experiments) Still it is not clear the background associated with each experiment of the neural network. Please add a clear description in the methodology of the testing you are conducting.

Answer: See lines 257-258, 268, 281, 289-294. Blue color.

5)    (Results) The Table of the "neural network runs" might be converted in graph facilitating the comprehension? (this is just a suggestion).

Answer: Clear description has been done in each Table.

6)    (Discussion) When it says: "Sphyrna zygaena 66% of 92 images....This indicates that 66% of the images were correctly identified as belonging to Sphyrna zygaena", from the methodology described, 92 images are the total available so about 74+9 were used for training (80%) +validation (10%) and 9 for testing (10%), therefore the result might be 66% of 9 images (or 5 out of 9) identified correctly.

Answer: See lines 340-342.

 

Other minor points to be considered:

7)    The input layer was the texture vector of the image, the texture is the way something feels when touched, or how it looks caused by the way in which it is woven...is this the proper language to describe the photo of a fin?

Answer: Generally, the texture is the way something feels when touched. It can be represented mathematically of different ways.  One manner is using the Local Binary Patterns.

8)    109‒110 “to identify the species of origin of 37 dry dorsal fins”: did you mean the 37 species of origin of a testing set of dry dorsal fins?

Answer: We had information of 37 species only. It is not origin of a testing set, it is all the information we had.

9)    123-124. "The dry fin sharks’ database was identified": The meaning of this sentence is not clear, did you mean "the dried fins were identified at species level"? How? (I read this is explained below, I suggest to be more consistent and avoid repetition)

Answer: See line 131, we mentioned three references to identify at species level.

10)     127‒129: This explication is generic and not related to the objective of the study. The species level is needed to respect the shape diversity and the possibility to identify a fin at species level depends in this case on the confidence of the expert.

Answer: This explication is not related to the objective of this study, but we wanted to enhancement the importance or the application of this study.

11)     130‒132: “To validate the use of the algorithms, all the shark fin photos were previously identified visually by shark fin identification experts [18, 20], based on their knowledge and published fin field guides”: This is a very sensitive point in the discussion about the capability of the algorithm you are testing to correctly identify the species to which the fins belong. I suggest discussing about the potential bias in the original identification, eventually introduced in the training tests, to be transparent and make the reader aware when evaluating the results.

Language:

Answer: See lines 132-133.

12)     Check the use of “dry/dried”

Answer: Done. Thank you very much.

Author Response File: Author Response.docx