Lightweight YOLOv7 Algorithm for Multi-Object Recognition on Contrabands in Terahertz Images

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1- define the abbreviations before using it.

2- correct the caption of figure 5 (Sandglass structure)

3- Equations (1) and (2) @ lines 255 and 256 are the same, please correct.

4- Regarding the LSK, what is the effect of the kernel size on the performance of the model? and what are the criteria of selecting them?

5- Combining preprocessing methods (e.g. non-local mean filtering and histogram equalisation) may improve the results, why did you not try it?

6- you should measure the effect of non-local mean filtering objectively since the images may not need this type of preprocessing or its effect may be low.

Author Response

Dear reviewer,

The authors would like to thank editor and reviewers for their constructive comments and suggestions on our manuscript entitled “Lightweight YOLOv7 Algorithm for Multi-object Recognition on Contrabands in Terahertz Images” (Manuscript Number: applsci-2814768). Those comments and suggestions are very helpful for revising and improving our paper. We have revised the manuscript based on the reviewers' comments and submitted a revised version.

 

Sincerely yours,

Zihao Ge

 

 

The point to point response to the two reviewers’ comments is in the word file.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors, thank you for an interesting article which, thanks to your descriptions, in an interesting way expands the knowledge in the area of contraband identification using image recognition algorithms (using a modified approach to the YOLO algorithm).

Before publishing, please refer to the following issues in the text:
- substantive issues:
-- please consider shortening the abstract and removing short names that may be incomprehensible to the reader without reading the entire text, e.g. SMLSK-YOLOv7,
-- in line 87 - this sentence needs more clarification, "better balance between" what?
-- the authors omitted the fact that the test installation provides an image of 512 x 256 pixels (in grayscale, upscaled to pseudocolor), and the algorithm presented in Figures 4 and 9 analyzes an image of 640 x 640 pixels (RGB) - please comment on the coloring method and image scaling,
-- the authors only mention that they conducted ablation experiments, but do not indicate which ones, which may make table No. 2 difficult to read - please comment and include them in the text,
-- table no. 1 indicated on line 380, does not represent the experimental environment configuration, please complete and renumber subsequent tables if a new one is added,
- editorial issues:
-- missing or additional space in line: 42, 47, 49, 89, 95, 101, 108,
-- "(1)" text in line 229,
-- "DIoU" text in line 420,
-- the abbreviation DIOU appears in the text on line 136, the formula describing this issue is on line 323, and there is no expansion of this abbreviation - even if it is obvious, the reader should learn how the authors understand this abbreviation.

Author Response

Dear reviewer,

The authors would like to thank editor and reviewers for their constructive comments and suggestions on our manuscript entitled “Lightweight YOLOv7 Algorithm for Multi-object Recognition on Contrabands in Terahertz Images” (Manuscript Number: applsci-2814768). Those comments and suggestions are very helpful for revising and improving our paper. We have revised the manuscript based on the reviewers' comments and submitted a revised version.

Sincerely yours,

Zihao Ge

 

 

The point to point response to the two reviewers’ comments is in the word file.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper is interesting, and addresses an important problem, in a field that is not much explored. The authors have started from an existing network and made multiple optimizations, which are all explained in terms of their working mechanism and the reasons for their use. The new architecture is tested against the baseline starting architecture and the results seem to be better.

However, the dataset is quite small, so you should be aware of the possibility of overfitting. What happens when the problem object is placed in a bundle with other objects, like in a real bag? Can you detect other types of knives or guns besides the ones that you train with? Maybe a solution for enhancing the dataset would be to use visual spectrum images and generate a pseudo-terraherz image from them through image processing, so you can have more data for training and testing.

Some details about the camera and the output image are missing: what is the pixel depth (how many bits per pixel) produced by the camera? How is the pseudo-colorization achieved? I suspect that the camera produces a grayscale image which can be processed directly, so why use color images?

Some text seems to be left over from editing: ".The main role of the MP module is to downsampling, which is achieved by splic- 212 ing the maxpool downsampling branch and the convolutional The main function of MP 213 module is downsampling, by splicing maxpool" - some text is repeated.

Also, equations (1) and (2) seem to be identical - maybe equation 2 was meant to be something else?

 

Comments on the Quality of English Language

The language quality is ok for me.

Author Response

Dear reviewer,

 

The authors would like to thank editor and reviewers for their constructive comments and suggestions on our manuscript entitled “Lightweight YOLOv7 Algorithm for Multi-object Recognition on Contrabands in Terahertz Images” (Manuscript Number: applsci-2814768). Those comments and suggestions are very helpful for revising and improving our paper. We have revised the manuscript based on the reviewers' comments and submitted a revised version.

 

Sincerely yours,

Zihao Ge

 

 

The point to point response to the two reviewers’ comments is in the word file.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

I saw no major improvements in the revised paper. None of my concerns have been properly addressed.

Author Response

Dear reviewer,

 

The authors would like to thank editor and reviewers for their constructive comments and suggestions on our manuscript entitled “Lightweight YOLOv7 Algorithm for Multi-object Recognition on Contrabands in Terahertz Images” (Manuscript Number: applsci-2814768). Those comments and suggestions are very helpful for revising and improving our paper. We have revised the manuscript based on the reviewers' comments and submitted a revised version. All modifications are marked yellow in the text.

 

Sincerely yours,

Zihao Ge

Author Response File: Author Response.pdf