Gender Identification of Chinese Mitten Crab Juveniles Based on Improved Faster R-CNN

Round 1

Reviewer 1 Report

In this investigation, the researchers strategically integrated deep learning technology with the distinctive criteria relevant to Chinese mitten crab juveniles, with the primary objective of proficiently discerning between male and female specimens. This innovative approach marks a significant stride in the evolution of contemporary fishery sorting apparatus, transitioning from semi-mechanized systems to intelligent ones. The research paper introduces an enhanced Faster R-CNN algorithm, and through the deliberate manipulation of image attributes employing techniques such as brightness adjustment, noise addition, rotation, cropping, flipping, and other forms of data augmentation, the authors successfully simulated a diverse array of real-world conditions pertinent to practical applications.

The presented research is notably intriguing, and concurrently, the manuscript exhibits commendable clarity and organizational coherence. Nevertheless, there exist certain constructive comments that merit consideration to augment the overall quality of the manuscript.

1.       The delineation of the paper's structure should be incorporated into the concluding segment of the introduction section.

2.       The authors ought to explicitly elucidate the methodology employed in defining the ground truth images utilized for training.

3.       It is advisable to include a comparative analysis with recently published papers operating under analogous conditions, encompassing both data and the evaluation protocol.

4.       Some perspectives should be added in the last part of the conclusion.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The presented manuscript is well written and can be further improved after the inclusion of following comments

1. Author has to include more latest articles related to study for better understanding  of the problem statement.

2. Author has to highlight the reason of selection of R-CNN model for the work.

3. Is author checked the performance of other model for the same problem? If yes then provide the obtained results.

4. Author includes literature work in the introduction section. Is it feasible for author to make an seperate section for related work?

5. Results are well described but not compared with previous work. So kindly include a comparison table for previous work and your work.

6. Minor language proof reading required.

7. Author has to also highlight the experimental work flow diagram for the better understanding of experiments.

8. Author has to include more parameters for performance evaluation.

Minor language proof reading required.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Apparently, there is no need to talk about traditional methods (lines 54-69), since they are not considered in the work. The introduction does not cite the key methods used to improve Faster R-CNN. They are well detailed on subsubsections 2.2.1 to  2.2.5, but no motivation for their use is contextualized in the introduction. So the authors are required to redistribute the introduction paragraphs from 2.2.1-2.2.5 to section 1. Even when describing PAFPN, the authors did not define the advantages of PAN before proposing to aggregate it to FPN  (line 207).

It is not clear whether data augmentation was performed separately in the training and test sets. Apparently, data augmentation was performed over the whole database before it was split into training and test sets. This affects the classification results and must be specified.

Also, the authors used only one-fold analysis. They are required to perform a k-fold analysis on the dataset to avoid any bias introduced by a single choice of training and test sets. This could be accomplished by partitioning the dataset in multiple combinations of mutually exclusive training and test sets.

In section 4.1, the authors claim they compare their proposition with two-stage algorithms, but YOLO and SSD are comprised in the benchmarking. So the authors are required to include these methods in the text.

Figure 5 noes not sufficiently illustrates the effects of filtering candidate frames by NMS and Soft-NMS and must be improved to evidence the improvements. What is x_e? Are AbsVal and Box_std applied to the BOX procedure output?

At line 279, it should be y’ instead of y^’.

At line 319, the self attention mechanism is related to [30] instead of ROI as indicated at line 320. 

The authors are required to perform an extensive review on English writing to improve readability. First of all, they are required to use pronouns since there is a lot of repeated names in each sentence. Some sentences miss the verb and show no logic consequence. Some sentences are written twice but with different approaches, possibly revealing that the authors were in doubt of how to say something ad forgot to remove one of them. (e.g.: “whereas for male crabs, it extends from November to December for male crabs is slightly later than that of female crabs” (lines 45-46); “In addition, due to the different reproductive molt times of male and female crabs, the molt time is different. ” (lines 46-47)). Some expressions like “my country” should be replaced by the name of the country (line 39), or even “detection head” (line 158) that should be described more technically. The authors are also required to replace “unisex culture of Chinese mitten crabs” (line 100) for “Chinese mitten crab single-gender culture” in order to avoid misunderstandings. 

The authors are required to perform an extensive review on English writing to improve readability. First of all, they are required to use pronouns since there is a lot of repeated names in each sentence. Some sentences miss the verb and show no logic consequence. Some sentences are written twice but with different approaches, possibly revealing that the authors were in doubt of how to say something ad forgot to remove one of them. (e.g.: “whereas for male crabs, it extends from November to December for male crabs is slightly later than that of female crabs” (lines 45-46); “In addition, due to the different reproductive molt times of male and female crabs, the molt time is different. ” (lines 46-47)). Some expressions like “my country” should be replaced by the name of the country (line 39), or even “detection head” (line 158) that should be described more technically. The authors are also required to replace “unisex culture of Chinese mitten crabs” (line 100) for “Chinese mitten crab single-gender culture” in order to avoid misunderstandings. 

There are some unspecified concepts, values or variables all along the text. The authors are required to specify them. For instance, IOU is undefined before line 246. TP, FP, FN are undefined before used in equations (4) and (5) (lines 350-356).

Something is missing or wrong in the sentence “When (1 − 𝑦′) becomes 296 smaller after being raised to the power of β, that is, (1 − 𝑦′)𝛽 tends to 1, which does not 297 affect the loss weight, resulting in a smaller value of the loss function.” (lines 296-298). Also some notation mus t be revised since the authors use a1, a2, a3, a4 to both the attention weights and images on Figure 10.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The manuscript was improved and can be accepted in the current form.