Hyperspectral Imaging and Hierarchical PLS-DA Applied to Asbestos Recognition in Construction and Demolition Waste
Round 1
Reviewer 1 Report
Abstract – include spatial resolution of system
L93 – define C&DW, perhaps at first instance in the abstract
L97 – how much asbestos? Concentration?
L101 – what was the size range?
L120 – 1cm FOV?
L132 – unsure where bracket ends
Fig 2 – what is the proportion of ACM v’s CDW pixels used in the calibration set? It seems quite low, based on Figure 2. In Figure 2b, how were the colours assigned? This should be explained more clearly in the caption.
Figure 5 - There seem to be different preprocessing methods depending on which part of the hierarchical model is considered – can you explain how these were selected?
Please show the effect of the various pretreatments on the spectral variability within each class.
Figure 7 – there seems to be a high degree of misclassification of the ACM samples – can the authors clarify if this is the case and why, e.g through inspection of misclassified spectra ? Please also show the ground truth and incorrectly classified pixels as additional sub-figures in this plot
Author Response
Please see attachment!
Author Response File: 
Author Response.docx
Reviewer 2 Report
The manuscript deals with a relevant application of hyperspectral imaging in order to identify asbestos contaminated materials in construction and demolition waste. The paper is generally well structured and easy to follow. In addition, the hierarchical classification model is strongly validated and the results are thoroughly discussed.
In my opinion the manuscript is suitable for publication after some minor revision that may further improve the work.
In the Introduction section I suggest to mention also the challenges derived from the high amount of data contained in hyperspesctral images, together with pertinent literature references Section 2.1: please add information about the illumination system of the hyperspectral camera and the procedure adopted to perform reflectance calibration When discussing Figure 2.b, please better specify how the pixels related to asbestos fibres were identified in the hyperspectral images in order to set the corresponding class In Section 2.1, please add a paragraph describing the hierarchical classification strategy together with proper literature references Page 7, lines 225-231: this paragraph is confusing, please rephrase it to make it more understandable for the reader Please add a paragraph describing the spectral preprocessing methods used for each classification rule or summarize them in a table Page 9, line 282: the “and” at the end of this sentence should be removed Page 10, line 313: “true negatives” should be “false negatives”
Author Response
Please see attachment!
Author Response File: Author Response.docx
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
I believe that this is a very interesting manuscript. It aims at development of a method that could be very useful on health and safety of demolition and renovation workers. In the following years it is expected great volume of ACM. Unfortunately, there is not yet specified an appropriate way of workers safety and ACM waste management.
I would like to kindly ask from the authors to consider the following:
line 44: "...producing dust" it would be interested to explain brittle nature of asbestos fibers to be understood the risk for workers.
line 75: "...chemometric techniques" I would like to be clarified the used techniques. Please provide details and justify the selection of specific ones, regarding their compatibility with PLS-DA.
Materials and methods: It s not clear how identified the different types of asbestos (amosite, crocidolite and chrysotile) presence in the ACM samples. Additionally, I would like to be clarified if your proposed method recognize the different type of asbestos fibers (probably due to their different optical characteristics i.e. colour, length curvity etc).
Figure 4: which analytical technique was used to obtain this diagram?
Additionally, I would like to ask you about specific results. According the results illustrated in Figure 7 and Table 3, it can be assumed that the most difficult recognition is between ACM and aggregates/mortar/bricks, probably due to the similarity of raw materials. Do you already propose a possible solution or approach regarding this problem in order to minimize false positives and true negatives?
Reviewer 2 Report
Specific comment to authors:
Page 1 line 10-24: Mention your key findings in the abstract. What is the minimum amount of asbestos that you can detect with your technique? What are the repeatability, reliability and efficiency of your technique?
Page 2 line 62-71: How is your work different from Ref 11 (doi:10.1016/j.jhazmat.2017.11.056)? What is the main difficulty that was not resolved in previous work?
Page 4 Figure 2: You need to define hypercube images and how you performed mosaicking.
Page 4 line 108-110: “Spectral data…………Inc.” Explain is detail how you performed spectral data analysis using chemometric techniques. Show example analysis in supplementary information.
Page 4 line 110-111: “A single mosaic…..Figure 2.” In previous section (Page 2 line 89-90) , you mentioned that four acquisitions were made. But now why you are using only two acquisitions to develop Figure 2?
Page 4 line 120: How did you define the calibration dataset?
Page 5, Figure 4: You should mark the Fe-OH, Mg-OH combination bands in the figure.
Page 5, Figure 5: Fig. 5 is very confusing. It seems like after pre-processing, you cannot see all four classes in a single figure. Currently ACM is showing two different spectra in Fig. 5c and 5d. Therefore, which pre-processing would then be used for your modelling? Also details of all pre-processing techniques should be provided in supplementary information.
Page 6, Figure 6: What were the objects and variables in the PCA score plots? I presume your pre-processed spectral values were used as variables. Were those values standardised for each object to ensure Gaussian distribution? What were separated in Fig. 6d? You need to list and define each rule before PCA?
Page 8 line 195-196: “Mortars……model.” Where did you demonstrate this spectral variance?
Page 8 line 203: What is your discussion on the specificity and sensitivity for four different rules in the calibration and validation sets? What significance do the specificity and sensitivity have in terms of your overall outcomes?
Page 8 line 214: It is not clear how you evaluated efficiency and reliability of your model.
Page 9 line 222-228: Author contribution to be stated in this section.
