A Method for Estimating the Injection Position of Turbot (Scophthalmus maximus) Using Semantic Segmentation
, Kang Wu 1,2, Songming Zhu 1,2,3, Zhangying Ye 1,2,3,* and Jianping Li 1,2,3,*
Round 1
Reviewer 1 Report
129 line: What is the sample size after data enhancement, please specify.
130 line: Is it appropriate to consider random cropping? Is it possible to ensure that random cropping does not destroy the integrity of the sample? For example, a sample with only half of the fish's body?
280 line: Please add the loss curves of the training set and validation set during training.
296 line: Please highlight in the diagram what the advantages of the Atten-Deeplabv3+ model are compared to the Deeplabv3+ model.
271 line: Please provide the Error for the training set, validation set, and test set under different models (calculated as Equation 10).
348 line: How is the injection position derived? The model doesn't seem to output position information.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In the experiment part of this paper, IOUs are compared. It is felt that the accuracy of DeepLabV3+is basically improved by about 1%, which belongs to the error range. This is not enough to show that the method in this paper is more effective. On the other hand, can you compare other indicators, such as F-measure, Precision, Recall, etc.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This is an interesting study about the application of Deep Learning (DL) in estimating the injection position of a turbot. The authors have collected a unique dataset using a cutting-edge methodology (semantic segmentation by attention mechanism). The paper is generally well written and structured. However, in my opinion, the paper has some shortcomings in regard to data analyses and text. Below I have provided some remarks on the text that should be considered and addressed before it is accepted for publishing.
1- The advantage of the DL method with respect to the traditional machine learning methods should be stated in the introduction. In this way, the author uses a recent paper (following paper) that refer to this matter.
https://doi.org/10.1007/s12145-022-00885-6
2- The best way to highlight the novelty in your study is by comparing it with the work that was done by others and pointing out the things that your study does which was never done before. In this way, please explain more about the objective and novelties of the presented work. For example why the author used an attention mechanism in the presented work. I suggested adding a list of the contributions in the last paragraph of the introduction.
3- Please add the correct training time of each considered method for comparison. The reader wants to know the efficiency of the presented method in terms of accuracy and execution time. please add to table 1. We expected the attention method to increase the computation time when it increases the accuracy by focusing on the, considered object.
4- Please add the segmentation results of the Unet, and PSPnet in figures 4 and 5; and show the improvement in a zoomed area.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors done a good job addressing the really key issues and concerns.
Author Response
Dear respectful reviewer,
Thanks for your reviews on our manuscript. We are also much grateful to you for putting forward the valuable suggestions in our manuscript, which are very helpful for improving the quality of our manuscript.
Best wishes,
Yours sincerely,
Wei Luo
On behalf of all the authors
