A Practical Hybrid IoT Architecture with Deep Learning Technique for Healthcare and Security Applications

Round 1

Reviewer 1 Report

1. On May 5, 2023, the World Health Organization declared that COVID-19 no longer constitutes a "public health emergency of international concern. What other applications does this article have besides preventing the spread of COVID-19 infection？

2. The quality of Figure 6 is too poor to see some data clearly.

3. The conclusion paragraph can be appropriate to talk more about future work.

4. It is suggested to modify the keyword. I think the pandemic and COVID-19 are repeated.

5. Lack of description of programming language, framework, etc.

Moderate editing of English language required.

Author Response

 Reviewer 1

(1) On May 5, 2023, the World Health Organization declared that COVID-19 no longer constitutes a "public health emergency of international concern. What other applications does this article have besides preventing the spread of COVID-19 infection？

Response:

Thanks for the comment. This study will not be only valuable for any upcoming outbreaks, but also for many other applications such as containing other healthcare-associated infections (HAIs), ensuring personal protective equipment (PPE) compliance, protecting individuals from air pollution, enhancing security protocols, and reducing criminal activities, etc. We have significantly extended the Introduction section to include those applications by adding pages 47-166 to the revised manuscript.

2. The quality of Figure 6 is too poor to see some data clearly.

Response:

Thanks for the comment. Figure 6 has been revised in the revised manuscript to enhance the quality.

3. The conclusion paragraph can be appropriate to talk more about future work.

Response:

Thanks for the comment. The conclusions have been modified as suggested.

The present smart mask detection system showed outstanding accuracy, less processing time, and the lowest model size with about 98%, 8.95 s, and 33 MB, respectively compared to other used models.

The next study will include state-of-the-art detection (YOLOv5, motion detection) in the future to enhance the accuracy for individuals at a distance and work well on edge devices. A potential benefit of this work is that it can be applied to other applications by creating more accurate models, which will result in better accuracy and higher quality results. This includes self-driving cars, traffic signals, parking, facial recognition, robotics, and the medical industries that use these technologies to detect objects in a variety of different scenarios.

4. It is suggested to modify the keyword. I think the pandemic and COVID-19 are repeated.

Response:

Thanks for the comment. The keywords have been modified as suggested.

5. Lack of description of programming language, framework, etc.

Thanks for the comment. We have added the description of programming language and the frame work to implement the proposed architecture in the revised manuscript

This section presents the implementation of the IoT architecture with deep learning technique which leverages a face mask detection model YOLOv4-tiny network, which is implemented in TensorflowRT and optimized by DeepStream. The model uses a combination of four public dataset: Kaggle Medical Masks, MAFA, WiderFace, and WIDER FACE dataset, with approximately 6000 labels for each object classes. The dataset includes 4 object classes: Face with mask; Face without mask; Face not visible; and Misplaced mask. With the integrated C270 HD Webcam, the quality of video can reach at 30 FPS with 1780×720 of resolution. Various object detection models were compared including MobileNetv2, YOLOv4 with a full version, and a tiny variant of YOLOv4.

---------------------------------------------------------------

Author Response File: Author Response.pdf

Reviewer 2 Report

in discussion refer in which other face recognition features can your system be applied and also where is this big data source accessible and to whom. Also refer to the investment costs of your project;

in conclusions, include limitations of the study and future research. Also so the implications for health crisis management for other diseases 

Minor editing of English language required

Author Response

 Reviewer 2

1. In discussion refer in which other face recognition features can your system be applied and also where is this big data source accessible and to whom.

Response:

Thanks for the comment. This study will not be only valuable for any upcoming outbreaks, but also for many other applications such as containing other healthcare-associated infections (HAIs), ensuring personal protective equipment (PPE) compliance, protecting individuals from air pollution, enhancing security protocols, and reducing criminal activities, etc. We have added a rich-information paragraph, presenting details in those applications to the beginning of Introduction section, pages 47-166 in the revised manuscript. In addition, this study will enhance the edge devices in terms of accuracy of the data flowing at the boundary between two networks. Other applications include self-driving cars, facial recognition, and robotics. We have added more information to the conclusions section.

2. Also refer to the investment costs of your project;

Response:

Thanks for the comment. In this study, NVIDIA Jetson Nano, Logitech 270 HD Webcam, and Intel Dual Band Wireless AC 8265 are used in the hardware setup of the intelligent camera device. The total cost for those devices is about 350 USD. In addition, in order to implement the IoT architecture online we need to install a local server and an IoT dashboard. We have added more descriptions of the proposed architecture in the revised manuscript.

3. In conclusions, including limitations of the study and future research. Also so the implications for health crisis management for other diseases.

Response:

Thanks for the comment. This study will not be only valuable for any upcoming outbreaks, but also for many other applications such as motion detection either for humans or vehicles. In addition, it will enhance the edge devices in terms of accuracy of the data flowing at the boundary between two networks. Other applications include self-driving cars, facial recognition, and robotics. We have added more information to the conclusions section.

The next study will include state-of-the-art detection (YOLOv5, motion detection) in the future to enhance the accuracy for individuals at a distance and work well on edge devices. A potential benefit of this work is that it can be applied to other applications by creating more accurate models, which will result in better accuracy and higher quality results. This includes self-driving cars, traffic signals, parking, facial recognition, robotics, and the medical industries that use these technologies to detect objects in a variety of different scenarios.

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper introduces a mask detection camera (MaskCam) system that leverages the computing power of NVIDIA's Jetson Nano edge nanodevices and has been built with a smart camera application for detecting a mask on the face of an individual, using a developed deep learning neural networks algorithm named You Look Only  Once (YOLO) as a real-time object detection method.

More information is required about training and testing datasets for deep learning algorithms. More information is required about the deep learning algorithms used and how they are evaluated. Modify the paper title to reflect the paper’s aim and purpose. More information about the methods and results should be included in the abstract. Figure 10 displays the accuracy, precision, recall, and F1 of the suggested model, although further information is needed on how these metrics were obtained. The related works are not provided. The paper’s contribution should be presented clearly in the Introduction section. In conclusion, it is said that the suggested method is cost-effective, although no evidence is shown in the paper to back this up. The conclusion should also incorporate the findings. 

Author Response

Reviewer 3

The article would be appropriate to explicitly indicate the scientific benefits of the present article explicitly.

The paper introduces a mask detection camera (MaskCam) system that leverages the computing power of NVIDIA's Jetson Nano edge nanodevices and has been built with a smart camera application for detecting a mask on the face of an individual, using a developed deep learning neural networks algorithm named You Look Only-Once (YOLO) as a real-time object detection method.

1. More information is required about training and testing datasets for deep learning algorithms. More information is required about the deep learning algorithms used and how they are evaluated.

Response:

Thanks for the comment. The authors have been adding more information about the dataset, and training dataset for deep learning algorithms in the revised manuscript.

This section presents the implementation of the IoT architecture with deep learning technique which leverages a face mask detection model YOLOv4-tiny network, which is implemented in TensorflowRT and optimized by DeepStream. The model uses a combination of four public dataset with approximately 6000 labels for each object classes. The dataset includes 4 object classes: Face with mask; Face without mask; Face not visible; and Misplaced mask. With the integrated C270 HD Webcam, the quality of video can reach at 30 FPS with 1780×720 of resolution. Various object detection models were compared including MobileNetv2, YOLOv4 with a full version, and a tiny variant of YOLOv4.

2. Modify the paper title to reflect the paper’s aim and purpose. More information about the methods and results should be included in the abstract.

Response:

Thanks for the comment. The title of the paper has been modified as suggested. In addition, The abstract has been revised to include more information about the method and results.

3. Figure 10 displays the accuracy, precision, recall, and F1 of the suggested model, although further information is needed on how these metrics were obtained.

Response:

Thanks for the comment. We added the model’s metrics which are defined in Table 2, which we have revised in the revised manuscript.

Table 2. Model evaluation metrics

                              * TP is true positive; TN is true negative; FP is false positive; FN is a false negative.

4. The related works are not provided. The paper’s contribution should be presented clearly in the Introduction section.

Response:

Thanks for the comment. The related works are added in the Introduction section.

• The YOLO tiny V4 was used to measure the accuracy of the masked face using a custom-built data set of a blend of several data sets.
• Analyze mask detection algorithms that provide high accuracy and frame rates, so that the system can operate in real-time.
• The use of MQTT communication for IoT applications allows the connection of devices to servers and data storage to become simpler and more convenient.

5. In conclusion, it is said that the suggested method is cost-effective, although no evidence is shown in the paper to back this up. The conclusion should also incorporate the findings.

Response:

Thanks for the comment. The conclusions have been modified as suggested.

NVIDIA Jetson Nano, Logitech 270 HD Webcam, and Intel Dual Band Wireless AC 8265 are used in the hardware setup of the intelligent camera device. The total cost for those devices is about 350 USD. In addition, in order to implement the IoT architecture online we need to install a local server and an IoT dashboard. We have added more descriptions of the proposed architecture in the revised manuscript.

The findings of proposed architecture have added in Conclusion.

In this study, Jetson Nano was successfully used in practice to deploy deep learning models using the proposed method. Yet, it is also possible to improve a neural network-based product in terms of accuracy and the implementation of deep-learning algorithms. Lastly, a web application was developed for data visualization and analysis. The present smart mask detection system showed outstanding accuracy, less processing time, and the lowest model size with about 98%, 8.95 s, and 33 MB, respectively compared to other used models.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

I think the work lacks a phrase or paragraph clarifying how the device can choose from a situation of wearing a mask or not wearing it.

For example, in a bank the person is using it because of a pandemic (obliged in that social context) but can be a thief . So, what to decide, what to do?

and so on... in other diverse situations mentioned for real application

If there are specific instructions, you should show them through a more necessary picture

Author Response

Reviewer 2

1. Comments and Suggestions for Authors

I think the work lacks a phrase or paragraph clarifying how the device can choose from a situation of wearing a mask or not wearing it. For example, in a bank the person is using it because of a pandemic (obliged in that social context) but can be a thief. So, what to decide, what to do? and so on... in other diverse situations mentioned for real application If there are specific instructions, you should show them through a more necessary picture.

Response:

Thank you for your comments and suggestions. Indeed, the facial mask detection technology in this study can be employed in many situations. At some places like hospitals and transportation hubs or during a pandemic where there is a high risk of spreading contagious diseases and wearing masks is mandatory, this technology helps identify individuals who are not following the mask-wearing protocol and then advise them to wear masks. However, in some restricted accessing and sensitive places like banks, military bases, power plants, etc., where people are asked to not wear masks while entering, this technology helps identify those who are wearing masks and inform them to take off their masks so that their identity will be checked and therefore enhances security. Generally, there are instructions for individuals to follow whether they need to/must wear or not wear masks in particular situations, but in case they do not follow, this facial mask detection technology in our study will help inform them to comply with those instructions.

We have added a paragraph to include this point as suggested by the reviewer in section Results and Discussions, lines 330-344 (highlighted in yellow) in the newly revised manuscript.

We hope that with this improvement, the manuscript is now satisfactory for publication. We would like to thank all the honored reviewers for your comments and suggestions in improving the quality of the paper.

Author Response File: Author Response.docx