Early Detection of Change by Applying Scale-Space Methodology to Hyperspectral Images

Round 1

Reviewer 1 Report

In my opinion, the manuscript applsci-695923, presented to Applied Sciences Journal has to be rejected. The scientific level of the presented paper is, in fact, too poor for publication.

I will offer here some of the main reasons, leaving out minor points like the text style, imperfections, typos.

Hyperspectral imaging “was” an emerging technique while, now, is offering quite robust tests and applications spanning over many application areas. Several reviews are available to guide within hundreds of scientific papers.  In the meanwhile, the Authors simply ignore such a vast amount of references available today, totally missing to achieve a scientific level adequate to show an updated understanding of this field. In this particular case, water pollution has been chosen as a case study. The topic is interesting, but for water pollution, it is usually meant a biological contamination (e.g., Legionnaires’ disease, or many others), chemical pollution (e.g., by heavy metals, such as lead) or other practical and real cases. The possible impact of an “alcohol” contamination is not a practical or interesting case: a useful analytic method can be validated only if it is proven to discover the minimum amount of pollutant that defines the possible risk, which, in turn, depends on the chemical (biological) species of the pollutant itself. The reason to select “alcohol” requires at least an explanation. I’m discouraged also by the choice of “alcohol” as the water contaminant: which “alcohol” are we dealing with? Are we dealing with ethanol? Methanol? A mixture? In this case, the precise composition of the chemical agent used as a pollutant, by using an IUPAC nomenclature, is mandatory to allow experiment reproducibility. Moreover, a comparison of the statistical tools used in this case with other techniques should be introduced ( for instance, PCA analysis) as well as citing  other more advanced tools currently used in the hyperspectral framework, like deep learning techniques, even if maybe too much advanced in the presented example.

Author Response

• We have changed the perspective of the paper to an improvement of an existing method.
• We have also obtained a new data set with a higher scientific quality. 

Reviewer 2 Report

The manuscript titled "Early detection of contamination in water by applying a scale-space methodology to hyperspectral images" by Uteng et al. described a method in processing hyperspectral data and proposed usage in the detection of water contamination. The paper has good potentials in future applications, however, it must be significantly improved before publishing. I would also suggest rewriting the introduction and conclusions.

First, the experiment design is questionable or needs more refinement. The manuscript used 10 repeating measurements of 100% water as the "standard" and compared other measurements with the standard. However, is this standard representative? The measurements performed in glasses may introduce spectral signatures of glass, thus really fine spectral features of water may be masked by the glass features. There may not be a good spectral featureless container suitable for liquids, but some cuvettes may help as physics/chemists use these for absorption measurements. Water has very low reflectance (less than 5%, check ASTER spectral library), especially for long wavelengths, the reflectance values shown in Figure 2 are abnormally high, making me believe that these curves in Figure 2 are not true reflectance of the liquids. Probably the subtraction between two very well-registered measurements of only the container and then the liquid-filled container may show the actual liquid reflectance. The curves in Figure 2d are very different from the others, and 2a, 2b, 2c, and 2e are showing similar spectral features. The measurement of 97.5% water may be wrong. Are the spectral differences showing the actual compositional variations? Is it possible that instrumental errors are classified as different? As we know that water-alcohol mixtures are transparent/have low reflectances, the subtle change between liquids of different compositions may not even be detectable by current instruments. A better-designed experiment is needed.

Second, assuming the experiment design is good, the result is not properly described. The authors didn't explain how to really understand and interpret Figure 3. Nor did the authors explain the significances of the range/mean/standard deviation analyzed in the tests, why would they perform as shown? Even if we put the results described in the methods section into the results, the length of results is very short, and the manuscript is not balanced. 

 

Comments for author File: Comments.pdf

Author Response

• We have changed the perspective of the paper to an improvement of an existing method.
• We have also obtained a new data set with a higher scientific quality. 

Round 2

Reviewer 1 Report

The newly submitted paper “Early detection of change by applying scale-space methodology to hyperspectral images” by Dr. Uteng et al., appears to be a fair analysis of the potentiality of hyperspectral imaging by using one specific statistical method within multiscale modeling, and can be accepted for publication.
However, it is required an extensive revision of the English text, possibly with the aid of a native English speaker. Some examples of text areas requiring extensive corrections are given below:

Lines 3-4: “We acquire hyperspectral images from one application: Monitoring of freshness of fish. Finally, The proposed method showed a rather low false positive rate.”;

Lines 10-12: “However, a method which reports many false positives is not that useful. This is what we are improving in the method developed in this paper, a sensitive enough method not reporting many false positives.”;

Lines 19-20: “and is now a household (?) method in many areas” ;

Lines 146-147: “The fish application is useful, however, to be practical different fish have to be tested. Not the same fish as in our example.”

Moreover:

The optical illumination makes use of 150 W lamps, using 1800 W of  electrical power, in total. However, it is well known that only a very small part of such a power becomes something related to the optical intensity, mainly because a W lamp “wastes” a lot of such a power for IR region (it is hot !), and, eventually, only a portion of the lamp useful spectrum is really ported at the output of the fiber bundles: can the Authors report the optical power actually imping on the sample ? That one is a much more interesting number.

Author Response

We have answered reviewer 1's comments in the uploaded document.

Author Response File: Author Response.pdf

Reviewer 2 Report

The revised paper "Early detection of change by applying scale-space methodology to hyperspectral images" by Uteng et al. provided lots of improvement comparing with the initial submission about applications in detecting water contamination. However, the writing (not only the English language but also the writing style) needs significant improvement before adequate for publication. 

The introduction was not appealing, and the logic flow was not established. The authors were showing comparison results of their method with previous methods in Fig. 1. And their comment on the highlighted changes describing Fig. 1 was not even a complete sentence. In your introduction, you should mention the name of the method that you cited in [4], and the name of your method.

The methods section cited Fig. 3, but Fig. 3 appeared in the results section. You should not cite the figure here as the figure is showing the result. There is a comma at Tabs distance in Equation 5, please reformat.

A lot of suggestions I gave in the first round were ignored (you may disagree but please give your reasons), or the questions were not solved:

1) This is what I commented last round: "Second, assuming the experiment design is good, the result is not properly described. The authors didn't explain how to really understand and interpret Figure 3 (now new Fig. 4). Nor did the authors explain the significances of the range/mean/standard deviation analyzed in the tests, why would they perform as shown? Even if we put the results described in the methods section into the results, the length of results is very short, and the manuscript is not balanced."

2) Listing order of citations;

3) Give a citation for Line 19 "The choice of wavelength range often includes parts of both the infrared (IR) and ultraviolet (UV) regions."

4) Deleting redundant phrases. 

5) Labeling order of subfigures, if you cite Fig. 2b first, you should put that subfigure as 2a;

6) Line 110 "When all 216 images are collected" makes me feel that this is not push-broom (taking all 216 channels for a line and then next line), but taking an image for one channel and then another image. Please rewrite this.

7) Labeling panels in Fig. 4.

 

Author Response

We have answered reviewer 2's comments in the uploaded document.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

This second revision has improved a lot compared to the previous two submissions. I would suggest some changes before publishing. Please see the attached comments in pdf.

I would ask another question about the applicability of the method. Is this method sensitive to spatial differences? For example, if I scan the skin mole of a patient a few times as the training dataset. A year later the same patient came again, but I cannot make sure that the patient is sitting on the exact same spot to be scanned in the exact same pixels in the scene. The illumination may be different, the patient may have had plastic surgery, etc., all factors may cause spatial differences. Is the method sensitive to these?

An experiment would be, cut a line (301 pixels) from Line 1 in the scene and put it as the new Line 301, and make the old i lines (i = 2:301) as the new i-1 lines. Would the method detect significant changes?

Comments for author File: Comments.pdf

Author Response

We have changed and improved the manuscript according to the reviewers notes.

Author Response File: Author Response.pdf