Accuracy Assessment of Drone Real-Time Open Burning Imagery Detection for Early Wildfire Surveillance
Round 1
Reviewer 1 Report
See attached file.
Comments for author File: 
Comments.pdf
Author Response
The letter of response has been attached as below,
Author Response File: Author Response.pdf
Reviewer 2 Report
The manuscript titled "Accuracy Assessment of Drone Real-Time Open Burning Imagery Detection for Early Wildfire Surveillance" has been submitted to the esteemed journal Forests (MDPI) within the Section "Forest Inventory, Modeling and Remote Sensing," as part of the Special Issue "UAV Aided Forest Fire Risk Prediction Based on Remote Sensing, Machine Learning, and Cloud Computing." The content presented within the manuscript is highly appropriate for the target journal and special issue.
The manuscript introduces a novel approach known as the Dr-TOBID platform, which employs a combination of various techniques to facilitate early wildfire surveillance. A significant component of the study involves a comprehensive series of experiments. Notably, the study outlines the development process of a novel deep learning model, encompassing the design of YOLOv5. This model is subsequently fused with the LSTM algorithm, thereby establishing a pioneering remote sensing method for real-time detection of smoke and fire incidents. Make just clear that you focus on optical datasets and is not exploring thermal. Also, briefly explain the reasons.
Despite the impressive content, there exist certain structural aspects that necessitate revision and adaptation. For instance, the inclusion of statistical assessment details (Section 4.1) within the Results and Discussion section (Section 4) appears somewhat incongruous. In general, the content, tables, figures could better presented. Below some more specific comments for each section.
Abstract:
It is recommended to incorporate additional quantitative results within the abstract. Moreover, the abstract should underscore the significance of the proposed approach and emphasize its optimal application scenarios, while also addressing the computational demands it entails.
Introduction:
An essential point to highlight in the introduction is that the input data utilized deviate from the conventional expectation of satellite images. Additionally, an interesting inclusion would be to mention the historical context of fire towers, traditionally situated in expansive forest plantations.
Methodology:
In the methodology section, it is crucial to provide an exhaustive description of each step. To enhance clarity, consider incorporating section numbers into the primary flowchart.
Results and Discussion:
To enhance readability, it is recommended to divide both the Results and the Discussion sections into two distinct parts. Within the Discussion section, supplement the existing content with additional comments, references to pertinent literature, and potential avenues for forthcoming research endeavors. You also may explore supplementary material.
Conclusion:
Within the conclusion, ensure that the findings are situated within a broader context. This serves to underscore their relevance and potential implications beyond the immediate scope of the study.
In summary and to conclude my report, the manuscript demonstrates substantial promise and aligns well with the thematic focus of the targeted journal and special issue. By addressing the highlighted structural aspects and incorporating the suggested content revisions, the manuscript stands to become an even more compelling contribution to the field of early wildfire surveillance.
Author Response
The letter of response has been attached as below,
Author Response File: Author Response.pdf
Reviewer 3 Report
Accuracy Assessment of Drone Real-Time Open Burning Imagery Detection for Early Wildfire Surveillance
The paper introduces an accuracy assessment of the Drone Real-time Open Burning Imagery Detection (Dr-TOBID) system, which utilizes a combination of the YOLOv5 object detector and a lightweight version of the Long-Short Term Memory (LSTM) classifier for detecting smoke and burning in drone imagery. While the research yields valuable outcomes, there are certain deficiencies in the current document that necessitate enhancements to ensure a scholarly outcome equivalent to the significance of the publication.
1. In lines 65-67, the paper highlights three deep learning techniques viable for aiding Unmanned Aerial Vehicles (UAVs) in precisely detecting fire and smoke: image classification-based methods, object detection-based methods, and semantic segmentation-based methods. The rationale behind choosing the object detection method as the focus of this research is expounded upon.
2. Briefly mentioning recent developments in the YOLO family of algorithms (such as YOLOv6, v7, and v8) would provide additional context for the study. For instance, Classical Image Processing and YOLOv7 Fusion Algorithm is already proposed for rapid automatic Camellia oleifera fruit detection.
3. The authors may add more state-of-art application articles for the integrity of the manuscript (Rachis detection and three-dimensional localization of cut off point for vision-based banana robot; Computers and Electronics in Agriculture. Detection and counting of banana bunches by integrating deep learning and classic image-processing algorithms; Computers and Electronics in Agriculture. Fruit detection and positioning technology for a Camellia oleifera C. Abel orchard based on improved YOLOv4-tiny model and binocular stereo vision; Expert Systems with Applications.).
4. Figure 6 should be presented with enhanced clarity and comprehensiveness.
5. A more comprehensive explanation of Figure 6 is required, detailing the process of acquiring RGB and HSV images, elucidating why RGB images are employed for smoke assessment, and clarifying why HSV images are utilized for burning detection.
6. The article should incorporate additional datasets related to smoke and burning.
7. It is imperative to verify the accuracy of Figure 12c in comparison to figures 14c and 18c.
8. A meticulous elucidation of the causes and trends of data variations based on variables is necessary, rather than solely presenting data in a cumulative manner across figures 11, 18, and Table 3.
9. The conclusion necessitates further expansion, providing additional insights and proposing areas for prospective improvement.
Author Response
The letter of response has been attached as below,
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The paper notably improved after its revision, and I have no further comments.
Author Response
The letter of response has been attached,
Author Response File: Author Response.pdf
Reviewer 2 Report
I am pleased with the revisions in response to the rebuttal letter and the manuscript. However, I would like to emphasize the importance of enhancing the discussion section further.
Given that this is an Article rather than a Technical Note, I cannot recommend it for publication like it is. However, with significant improvements to the discussion section and the inclusion of more pertinent references, it has the potential to meet the criteria for publication as a Scientific Article.
Please consider searching for more references in specific databases such as Scopus and Web of Science. Using specific words from the title, I found the following papers [10.3390/rs15030720, 10.1016/j.ecoinf.2021.101397,10.3390/su14105927, 10.1109/LGRS.2022.3229173, etc].
Mainly for the editor, I do not benefit from the abovementioned references. This is just a suggestion using the term "Wildfire Surveillance" in Scopus. A better and more detailed list of manuscripts is possible using more words and some strategies.
To conclude, as mentioned before, I will still insist. Authors must explore more references and enhance discussions to consolidate the manuscript as a Scientific Article. That is why I am recommending Major Revisions.
Author Response
The letter of response has been attached,
Author Response File: Author Response.pdf
