Automated Detection of Animals in Low-Resolution Airborne Thermal Imagery

Round 1

Reviewer 1 Report

Dear authors,

the topic of your manuscript is fascinating and again shows the versatility of thermal infrared remote sensing methods and the need of improvement for optimized application possibilities. The manuscript is well structured, the approaches are based on the state of the art findings. Your analysis parameters are reasonable combined. The benefit of your work comes out and extends the tools for thermal and automated animal detection.

The structure of your manuscript is well done and there are no major comments on analytical and method proceedings. However, I would like to recommend, that the paper has to be improved in cases of more detailed explanations, extended content on method aspects and critical discussion, as well as result visualization:

1. Please check if every abbreviation is defined with the long name at the first use in the text or figure (e.g. F1, F2 and F3 in the caption of figure 4).
2. Please include units wherever you show data, e.g. in line 122 at the feature map types.
3. The paper focuses of the development and application of your OYOLO-Algorithm. For a better traceability of your method, please improve your figures. The thermal images in Figures 1, 2, 8 and 10 needs a temperature scale. Partly they look like screenshots with possibly unnecessary information resp. displayed aspects. In some cases, the resolution is very low: the thermal image is limited because of camera specifications, but the surroundings and labels (e.g. Fig. 8 and 10) are in a low quality as well. A larger resp. smaller temperature range in the scale could also help upgrading the benefit of these figures (e.g. Fig 8). Additionally please give more details on the shown pictures in the text: more information on the shown scenarios in the text could help the reader for a better understanding of the method and results as well as assessing your approach.Please include the missing units at the axis of diagrams (e.g. Fig 7).

4. Possibly it would make more sense to include the facts, shown in lines 204-208 into a table, thus it is easier to compare these important results. Figure 9 visualizes the results, but wouldn´t it be more correct, to show these facts as a bar chart? The results you present are absolute numbers, thus graphs are wrong, showing a trend from species to species. I hope that I understand your results correctly.
5. Please extend the text on more details:
   a) Please reason the selection of thermal camera types and give more information on the effect of the selection concerning the method development and results resp. discuss this point in the results critically.
   b) Why there are three calculated types of feature maps (line 122). Is this a result of the method or a chosen aspect? If yes, please reason this.
   c) How is the splitting in “zoom in” and “zoom out” group, mentioned in 2.3 respected in the following method steps. Possibly, it would be reasonable to create a figure on the data handling, including aspects such as these two groups, camera types with data volume (N), the data splitting into training, evaluation and testing datasets and the corresponding N, et cetera.
   d) Please extend the method critic at the end of chapter 2: are there more method limiting factors or aspects, which complicate the results or the assessment of OYOLO? What would be needed for an optimized method application: better data, other or additional thermal cameras, more or other image filtering steps, more data (N), et cetera?

In detail:

• Line 24: space character is missing
• Figure 1: lowercase Kangaroos (rabbits, pigs and Kangaroos)
• Line 60: thermals
• Line 71: “This paper used the Convolutional…” Better “In our work we used the…”?
• Line 72: delete the point, lowercase Data
• Line 73/74 and others: Probably “department of primary industry” is a proper name and should be capitalized; applies also for other names.
• Line 75: of
• Table 1:
  needs a title/headline
  2. Probably it is more reasonable to renounce the column of the costs, because with one item in this column the comparison between different camera types is not possible. Same situation with the column “Image”.
  3. Short note within the table, how “Fixed ring” (Pigs) and “Drone” (Kangaroo) works is reasonable in this collection of system data.
• Line 89-90: Sentence structure is a little confusing
• Line 94: Why no use of a normal reference for the link http://labelme.csail.mit.edu?
• Figure 2: Relocate figure 2 into chapter 2.2.
• Table 2: There is no reference to this table in the text. Please add some explanations.
• Figure 3: Space character is missing in the title caption after “explained”
• Line 121-122: Please include units
• Line 126: “This architecture is shown in Figure 3” can be deleted. This information is included at the beginning of the passage.
• Line 129: at instead of ar?
• Figure 4:
  Please correct the caption concerning 3(A) and 3(B).
  2. Use left and right or (A) and (B). When you use (A) and (B) then place the letter in the figure as well.
  3. F1, F2 and F3 are mentioned for the first time in this caption. Please explain it more in detail or reference to a chapter or passage below.
• Line 143-146: Please correct to figure 4(A) and (B)!
• Figure 6:
  Relocate figure 6 into chapter 2.4.
  2. What is the reference of the Google areal images? It must be included in the caption. When the system is connected to google, there should be the chance for you to use Google-pictures for this paper. However, please check terms of copy right; Google has strict rules for the use and referencing of their data.
• Line 162: please change this phrase: “This section describes the description…”
• Line 165: Table 3
• Table 4: table is not shown completely, thus a review was not possible
• Lines 179-187: text is missing, thus a review was not possible
• Figure 7: Please include axis units
• Figure 10 should be renamed to figure 8 and should be located near to line 198/199.
• Line 200: pest
• Line 205: Delete additional space chapter.
• Line 207: Space chapter is missing.
• Figure 9: Include unit at y-axis. Please consider the note to this figure in the text above.
• Figure 8: There is no direct reference to this table in the text. Please extend the text on this.
• Line 211: for

Author Response

Dear Reviewer,

We have addressed the valuable comments mentioned in the review report to improve the quality of the paper.

Thank you.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and suggestions:

1. Consider to rename your algorithm: "Optimized YOLO" sounds a little bit arrogant;
2.  Table 1 has no caption. Moreover, consider to revise the format as several entries overlap;
3. Rephrase the caption of Figure 4, also explaining the meaning of the various items found in the figure;
4.  Line 138: it should be "let ... be". Also, "function" is missed after "discrete";
5. In formula 2 " _*dt "should be  " d*t ";
6. In line 104 the text refers to Figure 4, not 3. Check all the other keys;
7. Line 148: it is not clear to me if the dilatation coefficient is fixed or trainable;
8. Page 8: the final line is unreadable;
9. Section 2.5: Results. The adopted loss function is not given. Moreveor, it is not clear how the accuracy is computed;
10. Line 200: "pst habitats" stands for "pest habitats", I suppose;
11. Please, review all the document for other typos.

Author Response

Dear Reviewer,

We have addressed the valuable comments mentioned in the review report to improve the quality of the paper.

Thank you.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear authors,

my aspects and comments of the review 1 were appropriately adapted or rejected with good reasons. Figures were upgraded and content reasonable extended.

If possible, with little effort, you could change the scale (percentage) of figure 8 (b) to show the differences more clearly and replace the legend.

Best wishes.