Automated Hyperspectral Feature Selection and Classification of Wildlife Using Uncrewed Aerial Vehicles

Round 1

Reviewer 1 Report

In this study, wildlife classification was carried out with hyperspectral aerial images. Feature selection was applied to identify suitable bands. It is an interesting study for the use of hyperspectral data. However, major revisions are needed. Here are my suggestions:

1) A few more sentences about the novelty of the work should be added to the Introduction.

2) Based on what criteria did you determine the number of bands you chose with PCA and LDA? How does using more or fewer bands affect accuracy?

3) It is recommended to use popular ML algorithms such as RF and SVM as classifiers. There are many examples in the literature of the successful use of these algorithms for hyperspectral data.

4) PCA and LDA are out-of-date methods for feature selection. There are also deep learning-based methods and filter-based methods in the literature. I recommend implementing additional feature selection algorithms. For example, RelieF, Information Gain.

5) Abbreviations can be used to make the tables clearer.

6) The distribution of training and testing data can be presented as a table.

7) At the end of the study, suggestions for future studies should be added.

8) Table names could be more descriptive. More information about the context of the table should be provided.

9) PCA and LDA should be explained mathematically.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This manuscript (remotesensing-2722124) explores the use of uncrewed aerial systems (UAS) and hyperspectral imagery (HSI) for identifying and classifying invasive animal species. Focusing on spectral data, this study involved creating a hyperspectral library for four large mammal species and testing the effectiveness of HSI for classification using neural networks. The results showed high accuracy (over 90%) in identifying these species using spectral data, despite challenges such as intraclass variation and practical difficulties in data collection and analysis. The study suggests potential in using spectral data for wildlife management but highlights the need for further research in this area.

The manuscript is not suitable for publication in its current form. The introduction is informative. However, the authors have no clear hypotheses or objectives. The conclusions are generic. The figures are of low resolution and the tables are not formatted correctly. Therefore, lateral lines should not be used.

In the Materials and Methods section, references have been missing for support. In the results section, there is no mention of tables or figures. However, such comparisons are lacking. The discussion section is inadequate. There needs to be better discussions with data from the literature as well as citations.

Check the Instructions for Authors in Remote Sensing, especially regarding references.

L128. Scientific names. Please check the manuscript throughout.

Figure 1, why not create a map? The images were of low quality.

In the Results section, the images need to be clearly indicated.

The captions of tables and figures need to be described completely, highlighting all elements.

Table 1. Not suitable; it has a division. Are these two tables? Redo.

However, this discussion is inadequate. There are no conflicts of interest to declare. Moreover, it does not meet these objectives. What hypotheses were tested? How do the data correlate with the existing models in the literature? In addition, it is unsuitable for RS. In many cases, applied modelling is poor and biased. For example, in tables 3, 4, and 5. Precision: 21.93%; recall: 90.36%? This is just one example, but it has been repeated throughout the manuscript. This is unsuitable for an efficient and precise model. What corrections are necessary to obtain the spectra? I am sure that many filters were used, but why were they not described? For example, in Table 3, why was the accuracy 91.05% higher and recall lower at 39.05% for the PCA data?

The language is okay; however, some sentences and paragraphs are very long.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Please see attached document.

Comments for author File: Comments.pdf

Please see attached document.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Authors in this paper presented techniques of identification of invasive animals using hyperspectral images and the latest classification methods. Article is well-written. All data and analysis in this article presented appropriately. Methods used in this research are original and new. Methods described with enough details. Presented research is interesting and I think paper will attract a wide readership. 

The explanations under Table 1 are duplicated

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Dear Authors, I appreciate your comments. The manuscript has been improved in some aspects. However, why is the discussion presented as section 3.2? The discussion still contains elements of the results. There is no discussion yet based on the literature. Why do the authors not discuss these data? Note that from lines 456 to 550, there are no references, which is completely unacceptable in a scientific manuscript.

Minor change in grammar and spelling.

Author Response

"Please see the attachment."

Author Response File: Author Response.docx