Regional Mapping and Spatial Distribution Analysis of Canopy Palms in an Amazon Forest Using Deep Learning and VHR Images
, Ricardo Dalagnol 2, Ximena Tagle Casapia 3,4, Annia S. Streher 2, Oliver L. Phillips 5, Emanuel Gloor 5 and Luiz E. O. C. Aragão 2,6Round 1
Reviewer 1 Report
Dear Authors,
your submitted manuscripts reads well and the presentation of the distribution map of canopy palms in a tropical rainforest using a very high multispectral remote sensing image and the U-net convolutional network was impressive. While I have no further comments on either methods, presentation of results, discussion and conclusions, I was a bit disappointed to detect some form of plagiarism or you may call it to a lesser extent also inappropriate citation of literature. I am referring to line 168-176, 3.1 Model Architecture, which reads almost identical to a paragraph (section: Methods, U-net model, Architecture) in "Wagner et al. (2020): Mapping Atlantic rainforest degradation and regeneration history with indicator species using convolutional network, PLOS One". You may argue that you still cited the references, however, consider to either rewrite this section or instead of citing the references clearly indicate by using ".c.f." that you basically copied text from the aforementioned publication.
Author Response
"Please see the attachment."
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper take advantage of 0.5m spatial resolution images and deep learning methods in order to mapping canopy palms.
The paper has enough cuality to be published, but before, conclusions must be improved. More specific comments are included in the attached pdf.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
In the paper, the authors deploy U-net architecture for the segmentation of satellite images of vegetation. Some points need to be addressed on the paper:
- In the introduction, it is necessary to resume the state of art in 4-5 sentences. What other deep learning techniques used to address the same problem in your field ....
- It is necessary to have a schematic describing the u-net architecture.
- The dataset needs to be described in terms of pixels for every class. Moreover, if the dataset is unbalanced, did you opt for any enhancement of the architecture to correct the deviation of the results.
- The chosen metric is insufficient to assess the performance for pixel segmentation algorithms
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The comments are well adressed.
