Early Detection of Dicamba and 2,4-D Herbicide Drifting Injuries on Soybean with a New Spatial–Spectral Algorithm Based on LeafSpec, an Accurate Touch-Based Hyperspectral Leaf Scanner

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1、Some of the writing and formatting of this article need to be corrected, such as the punctuation of the abstract.

2、This article uses the matlab toolbox, whether the author has the license of the software.

3、The method of data establishment, the training network built and the influence of data quality on the results should be analyzed and discussed.

4、The author's spectral data is two-dimensional data, and the recognition accuracy of similar data after processing is 80%. Is there any comparison with similar algorithms? It is suggested to add in the article.

Comments on the Quality of English Language

Minor editing of English language required

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript utilizes a hyperspectral leaf scanner for early detection of herbicide induced losses, which has certain research significance. The main recommendations are as follows:

1. In the abstract, the method introduction is not detailed, and it cannot be seen how your method was carried out;

2. A. The title emphasizes early detection, B. The review summarizes two problems in  existing studies, C. The actually solved problems in this manuscript. Are these three consistent?

3. The diagram should be self explanatory, such as what is marked on the diagram? Should be annotated

4. How applicable is the method? How do the results reflect regional issues?

5. What are the main innovations in this manuscript? What is the difference with previous research methods?

Also, please provide a brief explanation: Which field does this study belong to? Is it related to remote sensing?

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors,

after reviewing this paper, the only thing i would add is some more recent relevant references - some are from 1979- and something more about the methodology section (eg why do you choose anova and t-test over other statistical tools).

The overall work is satisfactory.

Best,

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript present an interesting an relevant study to support the potential of High-Precision Spatial and Spectral Imaging Solution for early detection and classification of soybean damage caused by herbicides. The overall structure and writing style is of good quality. However, at several points in the methodology a better description needs to be provided including supporting arguments on the choice of the methods. More detailed suggestions can be found in the points indicated below.

Line 85-86: could you summarize previous research including methods where spectral and spatial features were adopted to characterize damage at leaf scale. This would support your choice on experimental set-up and methods for leaf level analysis

Line 102: could you specify which soybean variety was adopted for this study. Varieties might vary in their damage effect at leaf and canopy level

Line 119: could you provide more details on the visual assessment. Which protocol was used, how many observers, can you provide a reference figure/photos with varying damage levels and associated scores for reference.

Line 132: not clear what is the physical unit of the indicated threshold level: is this reflectance or a DN value? In latter case how do you deal with radiometric calibration of the images.

Line 140-146: Have you considered the evaluation of other VIs then NDVI like more green pigment related. NDVI is now averaged over complete leaf, this will result in a dilution effect. Alternatively you could also develop another indicator which also takes variation of reflectance within leaf into account.

Line 202-203: could you elaborate how these morphological features were included in the analysis: which method was used?

Line 212-218: could you in the discussion reflect on the automatization of the developed methodology and especially focus on those parts of the method where visual evaluation/processing is required to prepare image datasets or intermediate products

Line 232-233: to which extent would other limiting growth or stress factors influence the visual assessment

Line 241-242: it would be especially important how the visual score between 0 and 100% is being made: is this at leaf level or canopy level? How are the % scores estimated? Needs elaboration in methods section with photo illustration

Figure 4: what is the field of view for the NDVI calculation: it seems canopy scale as the NDVI is developing over time. However, other evaluation methods are based on leaf level. How can this be related in an overall method comparison: in this case visual vs. image based.

Line 281-282: which ground truth was used for the accuracy calculations: can you add the formula for the accuracy calculations to the methods section.

Figure 5: a more detailed explanation needs to be provided on the meaning of the pixel values and the resulting colours in the images. What is the meaning of low vs. high values in relation to leaf damage. As this is a prediction, do the original images also already provide insight into the damage patterns?

For the discussion, I would suggest to add references to result figures, and extend the discussion on field based assessment (line 443-444) as all kind of factors (varying illumination, other leaf damage factors, …) will influence crop reflectance. This also includes generalization of methodology to larger range of crop varieties.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The author made corresponding modifications to the article as required, including:

1.Regarding writing and formatting issues, a comprehensive review of the manuscript was conducted, and the abstract was thoroughly revised and errors were corrected;

2.Corresponding analysis and discussion have been added in the methods of data establishment, training networks established, and the impact of data quality on results;

3. Added explanations of control experiments and related content, and further supplemented a benchmark for evaluating the performance of the author's proposed method, particularly highlighting the superior performance of the author's proposed method in terms of recognition accuracy;

However, there are still some minor issues, such as the clarity of the image in Figure 4 being significantly lower than that in Figure 5. It is recommended that the author improve the quality of the image in the future to ensure the quality level of the entire text image.

Through the above modifications, the content of this paper has significantly improved in terms of accuracy, rigor, and completeness of experiments, which meets the level of journal inclusion. It is recommended to accept this article after optimizing some details.

Reviewer 4 Report

Comments and Suggestions for Authors

The authors have carefully made use of the provided suggestions and elaborated and improved the manuscript on several points. One textual point: title of section 3 can be reduced to Results. Discussion is presented in section 4.