A Thermal Imaging Flame-Detection Model for Firefighting Robot Based on YOLOv4-F Model
, Wei Huang 3 and Long Shi 4,*
Round 1
Reviewer 1 Report
- If the authors make the dataset publicly available, it would be a great contribution to the research community.
- Consider modifying the title. Instead of "An advanced", use wording to highlight the improvement of the model.
- The authors should explain the visible lights, near infrared, far infrared, and thermal infrared regions. Also, explain to which wavelength region the proposed method applies.
- Discuss the state-of-the-art methods. The literature is short. I suggest including a related work section. The proposed work is not robotics, but it is mainly computer vision. Therefore, the broad literature on camera-based fire/flame/heat detection should be discussed.
- There are many undefined acronyms. Examples are CSP and CBM.
- Provide a summary of the specifications of the thermal camera (max temperature, resolution, range etc.).
- The experiment section is weak. Authors should use compare their YOLO model with state-of-the-art detectors. Comparison with YOLO-tiny is not enough.
- Overall, the contribution of the paper is limited. To improve the paper, I suggest you publish the dataset and trained models, conduct more experiments and compare it with the state-of-the-art.
Author Response
Point 1: If the authors make the dataset publicly available, it would be a great contribution to the research community.
Response 1: We greatly agree with your idea on opening the dataset. We have applied for a GitHub account: wangyeheng123, and we will upload the dataset to the GitHub account when the work on organizing is completed.
Point 2: Consider modifying the title. Instead of An advanced, use wording to highlight the improvement of the model.
Response 2: Thank you very much for your suggestion, the title of the manuscript has been revised as An thermal imaging flame detection model for fire-fighting robot based on YOLOv4-F model.
Point 3: The authors should explain the visible lights, near infrared, far infrared, and thermal infrared regions. Also, explain to which wavelength region the proposed method applies.
Response 3: The visible wavelength is 0.4-0.75 um, the near infrared wavelength is 0.75-1.4 um, the far infrared wavelength is 50-1000 um, and the thermal infrared region wavelength is 8-15 um. The model mentioned in this manuscript is applied to the infrared camera, using a wavelength range of 7-14um, which belongs to the thermal imaging region. The wavelength regions of the specific different images with the wavelength regions studied are detailed in the manuscript.
Point 4: Discuss the state-of-the-art methods. The literature is short. I suggest including a related work section. The proposed work is not robotics, but it is mainly computer vision. Therefore, the broad literature on camera-based fire/flame/heat detection should be discussed.
Response 4: In the previous version of the manuscript, we wanted to introduce the reader more to the current state of research on firefighting robots, and neglected to introduce the flame detection algorithm. In this version of the manuscript, we have added a new chapter "2 Related work", as you suggested, to provide a detailed description of the flame detection algorithm.
Point 5: There are many undefined acronyms. Examples are CSP and CBM.
Response 5: Both CSP (Cross-stage partial connections) and CBM (Conv-BN-Mish) are the base modules in the YOLO algorithm. CBM is composed of convolutional layer, normalization layer, and Mish activation function. CSP is composed of multiple CBM modules connected. We have checked all the abbreviations in the manuscript and have noted the full name in the manuscript.
Point 6: Provide a summary of the specifications of the thermal camera (max temperature, resolution, range etc.).
Response 6: The thermal infrared camera parameters used in the manuscript are temperature range -40°C to 550°C, resolution 1280×1024, wavelength range 7~14um, and thermal sensitivity <40mK@f/1.0. The above parameters have been described in detail in the manuscript.
Point 7: Provide a summary of the specifications of the thermal camera (max temperature, resolution, range etc.)..
Response 7: In the new version of the manuscript, we have added a comparison with the current state-of-the-art lightweight network YOLOv5-s, YOLOv7-tiny algorithms. The experimental results and data have been added to the "5. Results and discussion" section of the article.
Point 8: Overall, the contribution of the paper is limited. To improve the paper, I suggest you publish the dataset and trained models, conduct more experiments and compare it with the state-of-the-art.
Response 8: Thank you very much for your review of the manuscript and your valuable comments. We have extensively revised the manuscript in accordance with your suggestions. The authors agree that the revised manuscript is indeed a significant improvement over the previous one. I am looking forward your reply.
Reviewer 2 Report
Review: A flame detection model applicable to firefighting robots is proposed in this paper. Based on YOLOv4-tiny model, the work in the paper includes replacing the Mish activation function with a Leaky ReLU activation function, adding a pyramid pooling layer (SPP), and using a path aggregation network (PANet) instead of a feature extraction network (FPN). The infrared image flame dataset is constructed to verify the superiority of the model, and an improvement in flame detection accuracy is demonstrated by the experimental data. The paper is well written and organized. Please, find here below some suggestions for possible improvements:
1. The more details about the specifications of the thermal imaging camera should be provided (e.g., temperature range, wavelength, thermal sensitivity, etc.)
2. The 'confidence' was used as an evaluation result in the paper, please give a definition of 'confidence' in the paper.
3. The font in Figures 3, 5, and 6 should be larger.
4. For Table IV, why discuss the average mAP under five IOUs? Please provide a more detailed interpretation of this table.
Author Response
Point 1: The more details about the specifications of the thermal imaging camera should be provided (e.g., temperature range, wavelength, thermal sensitivity, etc.)
Response 1: The thermal infrared camera parameters used in the manuscript are temperature range -40°C to 550°C, resolution 1280×1024, wavelength range 7~14um, and thermal sensitivity <40mK@f/1.0. The above parameters have been described in detail in the revised manuscript.
Point 2: The 'confidence' was used as an evaluation result in the paper, please give a definition of 'confidence' in the paper.
Response 2: Confience means the accuracy of the prediction frame in predicting the class and location of an object. A specific explanation of Confience has been added to the revised manuscript.
Point 3: The font in Figures 3, 5, and 6 should be larger.
Response 3: We have modified the text of the images in the manuscript accordingly, and the images in the manuscript have been replaced.
Point 4: For Table IV, why discuss the average mAP under five IOUs? Please provide a more detailed interpretation of this table.
Response 4: IOU means the ratio of intersection and concatenation of target prediction frame and real frame. The larger itis seted, the lower the average accuracy of detection. By setting 5 different values of IOUs, the detection accuracy (mAP) of YOLOv4-F algorithm is compared with other algorithms in detecting the average of flames at different scales. The results demonstrate that the mAP values of the YOLOv4-F algorithm for different scales of flames are higher than the other algorithms for five different IOUs values. Corresponding explanations have been added to the manuscript.
Round 2
Reviewer 1 Report
The authors have addressed my comments and improved the manuscript. However, the writing is not improved.
The modified text has grammatical errors.
The newly added Related Work section has serious writing flaws. Please make sure all the statements are correct, and the writing is smooth.
Author Response
Point 1: The authors have addressed my comments and improved the manuscript. However, the writing is not improved.
Response 1: The relevant description of the improved article has been rewritten in the text.
Point 2: The modified text has grammatical errors.
Response 2: The article has been corrected for grammatical errors.
Point 3: The newly added Related Work section has serious writing flaws. Please make sure all the statements are correct, and the writing is smooth.
Response 3: The article has been carefully revised for the related work section of the writing.
Author Response File: 
Author Response.docx
Reviewer 2 Report
No more comments.
Author Response
Thank you for your review of the manuscript and your valuable suggestions. We have double-checked the grammar and spelling of the article, so don't hesitate to get in touch with me if there are any problems.
Author Response File: Author Response.docx
