Optimization of VGG16 Algorithm Pattern Recognition for Signals of Michelson–Sagnac Interference Vibration Sensing System
Round 1
Reviewer 1 Report
This study used optimized VGG16 network to recognize big data signals generated by a Michelson-Sagnac interferometric vibration sensor system. It showed a high accuracy and fast speed. The statement is clear. However, there are some suggestions and questions:
(1) Fig.1 and Fig.2 are structures from others. The authors may draw their own optimized VGG16 network structure model instead of giving others’ figures and a long statement of improving.
(2) The authors can also draw a figure for Michelson interferometer system to clearly demonstrate the process in line 112-123 in this article.
(3) For VGG16, it contains numbers of layers. In this work, only 900 data was used for training and 180 data was used for test. The data used for training and testing may be too little.
(4) The signal is a 1D data and VGG16 is a network that dealing with image. This data processing can be more detailed.
(5) Is there a detail loss function? If have, the authors may provide the loss values and training epochs.
Author Response
Thank you for your comments concerning our manuscript. Those comments are valuable and very helpful. We have read through comments carefully and have made corrections. Based on the instructions provided in your letter, we uploaded the file of the revised manuscript. Revisions in the text are shown using yellow highlight for additions. The responses to the comments are presented following.
(1) Fig.1 and Fig.2 are structures from others. The authors may draw their own optimized VGG16 network structure model instead of giving others’ figures and a long statement of improving.
---Response: We are grateful for the suggestion. We deleted Fig.1 and Fig.2. We drawed our own optimized VGG16 network structure model.
(2) The authors can also draw a figure for Michelson interferometer system to clearly demonstrate the process in line 112-123 in this article.
---Response: We are grateful for the suggestion. The process in line 112-123 is illustrated in our early work(Opt. Express 2020, 28, 7207-7220). In this version, we cited this paper and deleted the previous relevant expressions to increase readability.
(3) For VGG16, it contains numbers of layers. In this work, only 900 data was used for training and 180 data was used for test. The data used for training and testing may be too little.
---Response: This is a very good question. The VGG16 model consists of small convolution kernels and a special model structure. Even in the case of few samples, it has strong learning ability. We added this explanation to make the article easier to understand.
(4) The signal is a 1D data and VGG16 is a network that dealing with image. This data processing can be more detailed.
---Response: We are grateful for the suggestion. After the one-dimensional vibration signal is normalized, it is converted from one-dimensional to a two-dimensional matrix. This matrix is the corresponding grayscale image after signal conversion. We added this explanation to make the article easier to understand.
(5) Is there a detail loss function? If have, the authors may provide the loss values and training epochs.
---Response: This is a very good question. The loss function used by VGG1D is the cross entropy loss function. We added this explanation to make the article easier to understand.
Author Response File: Author Response.docx
Reviewer 2 Report
In this manuscript, since the more accurate and faster pattern recognition methods are key problems which limit the wider applications of optical fiber vibration sensor systems, this work is very important, original and novel. Using the VGG16-1D method, high accuracy of pattern recognition is obtained. Using the VGG16 method, fast pattern recognition speed is obtained. I think this manuscript can be accepted with minor modifications:
(1) Change "VGg" to "VGG".
(2) Revise this sentence so that it will be easier to understand: VGG16-1D has 1/85th of the parameters than VGG16-2D.
Author Response
Thank you for your comments concerning our manuscript. Those comments are valuable and very helpful. We have read through comments carefully and have made corrections. Based on the instructions provided in your letter, we uploaded the file of the revised manuscript. Revisions in the text are shown using yellow highlight for additions. The responses to the comments are presented following.
(1) Change "VGg" to "VGG".
---Response: We are grateful for the suggestion. We have changed "VGg" to "VGG".
(2) Revise this sentence so that it will be easier to understand: VGG16-1D has 1/85th of the parameters than VGG16-2D.
---Response: We are grateful for the suggestion. We have changed this sentence to "The number of parameters of VGG16-1D is 85th that of parameters of VGG16-2D"
Author Response File: Author Response.docx
Reviewer 3 Report
The application of optical fiber vibration sensor system becomes a hot topic recently. How to get a more accurate and faster identification method is a very important problem, which usually attract readers' interest in discussion. In this paper, the optimization of VGG16 algorithm pattern recognition for signals of Michelson-Sagnac interference vibration sensing system is proposed. A higher accuracy of 98.44% and a shorter recognition time of 0.03 seconds are obtained by the authors. The proposed methods are novel and the performance is good. The logic structure is well arranged with a good presentation. Minor revisions as required for being published by this journal.
(1) Please briefly explain the essential reason in the manuscript why this proposed method can obtain such a good performance comparing to other methods.
(2) In 4.2, VGG16-1D has 1/85th of the parameters than VGG16-2D. Please revise the related description for clearer expression.
(3)There are many typos in the manuscript, please check and revise.
Author Response
Thank you for your comments concerning our manuscript. Those comments are valuable and very helpful. We have read through comments carefully and have made corrections. Based on the instructions provided in your letter, we uploaded the file of the revised manuscript. Revisions in the text are shown using yellow highlight for additions. The responses to the comments are presented following.
(1) Please briefly explain the essential reason in the manuscript why this proposed method can obtain such a good performance comparing to other methods.
---Response: This is a very good question. The VGG model uses a small (3×3) convolution kernel to construct a deeper convolutional neural network structure. In this paper, we optimize the VGG model by adding a maximum pooling layer to the first layer of the network. In order to prevent the model from over-fitting, the original full connection layer was removed. Therefore, it is more suitable for processing the optical fiber sensing data in this paper to get good performance comparing to other methods.
(2) In 4.2, VGG16-1D has 1/85th of the parameters than VGG16-2D. Please revise the related description for clearer expression.
---Response: We are grateful for the suggestion. We have changed this sentence to "The number of parameters of VGG16-1D is 85th that of parameters of VGG16-2D"
(3)There are many typos in the manuscript, please check and revise.
---Response: We are grateful for the suggestion. We have checked and revised the manuscript. Revisions in the text are shown using yellow highlight for additions.
Author Response File: Author Response.docx