Improved Mask R-CNN Combined with Otsu Preprocessing for Rice Panicle Detection and Segmentation

Round 1

Reviewer 1 Report

The author has done a lot of work, and his experiments are relatively rich. However, there are still some problems to be improved.

1. The introduction should highlight the differences between the contributions of this paper and previous studies. It is suggested to refer to the similar literature framework of applied science. At the same time, the chapter framework should be added in the last paragraph, such as Section2, we ...; The next section, ....

2. Lack of literature review chapters. Although some literatures are listed in the introduction, the existing research should be reviewed and discussed.

3.Figure 2 is recommended in the appendix.

4.Figure 3 should be described in detail, not simply put up.

5. mean average precision should be MAP

6.In Table 4, the Evaluation indicators of comparison between different methods are inconsistent, so it is suggested to keep them uniform. Otherwise, it is difficult to make horizontal comparison.

Author Response

First of all, I would like to express my heartfelt thanks to the reviewers, deputy editors and chief editors for their constructive suggestions. I would like to thank all experts for taking time out of their busy schedule to put forward valuable suggestions for my article. Secondly, we have modified the paper according to the comments of the reviewers. We have revised the manuscript carefully according to the template provided by the journal and read it several times to make sure its correctness. We have submitted picture files for figures, keeping in mind to meet the requirements mentioned in template. Finally, thanks again to the editors and reviewers for their valuable comments on the paper.

Author Response File: Author Response.pdf

Reviewer 2 Report

Paper presents an elegant solution for improving R-Mask CNN based segmentation by applying well known Otsu method for thresholding. An experimental investigation performed on real collected dataset showed better performance to alternative solutions, which are also provided in the paper.

Few remarks:

1. "greyscale" -> "grayscale"

2. Font size in Figure 5 and Figure 6 should be increased. Vector graphics prefered.

 

Author Response

First of all, I would like to express my heartfelt thanks to the reviewers, deputy editors and chief editors for their constructive suggestions. I would like to thank all experts for taking time out of their busy schedule to put forward valuable suggestions for my article. Secondly, we have modified the paper according to the comments of the reviewers. We have revised the manuscript carefully according to the template provided by the journal and read it several times to make sure its correctness. We have submitted picture files for figures, keeping in mind to meet the requirements mentioned in template. Finally, thanks again to the editors and reviewers for their valuable comments on the paper.

Author Response File: Author Response.pdf

Reviewer 3 Report

1.    In Figure 1, Area 6 and Area 7 are outside the range of cameras, so how to detect the rice in these two areas?

2.    Is the adjustment of the RPN anchor box and Bounding box adaptive according to the model or manually adjusted by yourself?

3.    Is there an improvement in the OTSU method? If so, please write more clearly.

4.    Line363: Fig.5 should be changed to figure7?

5.    Line 387: How to see that the results obtained by the Otsu algorithm were both on the high side and those obtained by the original Mask R-CNN algorithm were both on the low side in Figure 6?

6.    How is the Reference values in the Figure10 and Figure 11 obtained? and is there a specific explanation in the article?

7.    The Evaluation indicators in Table 4 should be explained specifically, what are the meanings of P, F, R, RMSE, and accuracy? Why the methods of different indicators can be compared together? The method proposed in this article seems to have worse results than the method inDuan Lingfeng et al. 2018[32]?

8.    The article only shows the detection effect of 20 pictures, is each picture tested separately, and is there an average result on the entire dataset?

9.    This article is only tested on the AE dataset, is the proposed model applicable on other datasets as well?

Author Response

First of all, I would like to express my heartfelt thanks to the reviewers, deputy editors and chief editors for their constructive suggestions. I would like to thank all experts for taking time out of their busy schedule to put forward valuable suggestions for my article. Secondly, we have modified the paper according to the comments of the reviewers. We have revised the manuscript carefully according to the template provided by the journal and read it several times to make sure its correctness. We have submitted picture files for figures, keeping in mind to meet the requirements mentioned in template. Finally, thanks again to the editors and reviewers for their valuable comments on the paper.

Author Response File: Author Response.pdf

Reviewer 4 Report

The work presents a hybrid solution, which combines positive features of existing algorithms (Otsu and Mask R-CNN) to increase the accuracy in yield estimation for rice. The materials and methods show the care that the authors took with the data set generated, from the setup of the experiments, arrangement of cameras, to the treatment via software. The results show significant gains from the proposed approach, in relation to the approaches known in the literature and, therefore, I consider that the work is ready to be published. In addition, the authors propose to improve the model in future works, making it clear the continuity of the work.  I only suggest two modifications to the original text. Figure 3 shows the Mask R-CNN model, but the text must contain an explanation, even if simplified, of the steps that are performed in this model, as was done for the approach proposed in Figure 4.  Equation 4, in my understanding, has an error. The second line should be I = 0, I < threshold and not as is.

Author Response

First of all, I would like to express my heartfelt thanks to the reviewers, deputy editors and chief editors for their constructive suggestions. I would like to thank all experts for taking time out of their busy schedule to put forward valuable suggestions for my article. Secondly, we have modified the paper according to the comments of the reviewers. We have revised the manuscript carefully according to the template provided by the journal and read it several times to make sure its correctness. We have submitted picture files for figures, keeping in mind to meet the requirements mentioned in template. Finally, thanks again to the editors and reviewers for their valuable comments on the paper.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The author has carefully revised it, and his work deserves recognition.