Next Article in Journal
Construction of a Gothic Church Tower: A 3D Visualisation Based on Drawn Sources and Contemporary Artefacts
Next Article in Special Issue
Towards Using Digital Technologies to Balance Conservation and Fire Mitigation in Building Heritage Hosting Vulnerable Occupants: Rapid Evacuation Simulator Verification for the “Omero Museum” (Ancona, Italy)
Previous Article in Journal
False Promise: World Heritage, Ecotourism, and the Local Community of Strahan, Tasmania
Previous Article in Special Issue
Mortar Characterization of Historical Masonry Damaged by Riverbank Failure: The Case of Lungarno Torrigiani (Florence)
 
 
Article
Peer-Review Record

Deep Learning Enhanced Multisensor Data Fusion for Building Assessment Using Multispectral Voxels and Self-Organizing Maps

Heritage 2024, 7(2), 1043-1073; https://doi.org/10.3390/heritage7020051
by Javier Raimundo 1,*, Serafin Lopez-Cuervo Medina 1, Julian Aguirre de Mata 1, Tomás Ramón Herrero-Tejedor 2 and Enrique Priego-de-los-Santos 3
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Heritage 2024, 7(2), 1043-1073; https://doi.org/10.3390/heritage7020051
Submission received: 2 January 2024 / Revised: 1 February 2024 / Accepted: 2 February 2024 / Published: 17 February 2024
(This article belongs to the Special Issue Conservation Methodologies and Practices for Built Heritage)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper presents an innovative contribution for the analysis of the state of conservation of built surfaces based in a voxelization approach for multi-sensor data fusion.
The overall quality of the paper is good, but I think some aspects should be complemented and clarified.

I know that the main purpose of the paper is not to discuss how data is generated but I think that some clarifications are needed regarding this aspect to convince the reader of the appropriateness of the process.

In 2.2, for the sake of consistency could you:
a) Indicate how many targets were used?
b) Indicate how many targets in the ground?
c) Detail the "satellite positioning techniques"?
d) Provide a table with the marker’s coordinates.

In 2.3:
a) when you refer that the point clouds are georreferenced, i would suppose that ground markers were used for that purpose. Is this correct or was it done in a different way (e.g. the laser scanning point cloud providing the coordinates for the alignment of the other models and a final georreferencing in block with all the models)? If so, is the GPS precision good enough to ensure a correct registration (overlap) between the point clouds? Can you clarify this process?
b) And the targets in the walls, how were they used? As tie points only? Are these targets easy to identify in all the types of images?
c) Can you also add an overall short description of what software tools were used to generate the point clouds? Are all the point clouds generated the same way (excluding the laser one)? Are the thermal point clouds just a mapping of the thermal imagery onto the photogrammetric point cloud? This should be clarified.
d) In figure 7 you should also have an image of the laser scanning point cloud.
e) In table 5, would it make sense to indicate and average spatial resolution for each point cloud?

In 2.4:
a) You say that the RGB point cloud was used because it provides the most comprehensive coverage of the building. But it is not clear if interiors were considered. Maybe they are represented in the laser scanning point cloud. Would not this point cloud (laser scanning) be the most appropriate, if interiors ought to be considered? If only exteriors are considered, please declare so.

In 2.5.
a) Is orientation of the voxels adjusted to the main orientations of the building (e.g. voxel faces paralel to the main planes of the building)?
b) Figure 10 is not referred nor explained in the text and it should be. Because the voxel quantities are very disparate maybe the vertical scale should be nonlinear or, alternatively, you could display the absolute count on top of each bar.
c) In 2.5.1 you use the concept or notion of "full voxels" but do not explain what a full voxel is. This may be evident to an expert, but it is not for a person of a different area. I think that you should explain what a full voxel is.

In 3 and 4 you only discuss the 5cm voxel. Can you justify this choice? I think that it would be interesting to extend the discussion a bit in order to see if the results are consistent when considering other voxel sizes. Can we state that there is an optimal voxel size to perform these analyses from a state of conservation point of view? Is there any correlation between the spatial resolution of the point clouds (that was not mentioned) with the choice of an appropriate voxel size?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The paper is well written but needs few improvements.

1. In the photogrammetric process, more information are needed: setting parameters of camera, both UAV's and terrestrial, GDS, overlapping in the UAV flight, why did you use a 16mm lens...?

2. Have you integrated laser scanner data with photogrammetry? if yes, why? which accuracy did you get? if no why did you need also a laser scanner survey?

3. You stated you used Self-Organizing Map (SOM). It is quite clear why, but did you try other ways?  did you try other algorithms? have you done comparison?

4. in the conclusion you can improve the critics results you obtained

Comments on the Quality of English Language

English is quite fine, maybe check few sentences are long

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Back to TopTop