Concatenated Residual Attention UNet for Semantic Segmentation of Urban Green Space

Round 1

Reviewer 1 Report

The Authors present an improved fully convolutional neural network based on the encoding and decoding structure for urban green space detection applied to Gaofen-1 remote sensing images. The topic is of great interest to the scientific community. However, some issues need to be addressed before considering for publication.

The paper has to be carefully revised because many sentences are unclear and confusing. Here are just a few examples

Line 14-15: “not fully represent the spectral and spatial information of the green spaces”. Classification methods use spectral and spatial information to classify the target, not vice versa. Please clarify the meaning of the sentence.

Line 75: “vegetation is affected by …. water bodies”? Rewrite because it is confusing

Line 77-81: Revise. “Such as” and “are widely used” refer to the same subject, making the sentence confusing.

Line 81: “foreign object with the same spectrum”. It is not clear, especially in reference to the high spectral resolution. See also Line 404-405.

Line 84-89 and Line 91-94: Unclear. Rewrite to improve understanding

Line 97: “feature extraction” repeated

Line 99: What does "mainly " stand for? Add a simple description of the role of the different layers.

Line 117-119: Confusing.

Line 129-130: “Compared with other classical …..”, confusing.

Line 130: “fully”?

Line 141-143: rewrite.

Line 144-148: “remote sensing image” is the subject?

Line 182: “He and others”, write the reference appropriately.

Line 191: “the figure” and Line 195: “the graph”. Add the figure number to the reference.

Line 201-205: Revise

Further:

1 - There is no reference to the software implementation of the algorithms used for data processing. Has any software been developed ad oc? in which language? in what environment?

2- in paragraph 4.1 it is not clear the connection between table 4 and figure 8.

3- paragraph 4.2 seems to propose only a visual comparison, indeed very difficult due to the high similarity between the various images. A clear reference to the applied metrics is needed to support the conclusion that "Figure 10 shows the best performance for CRAUNet"

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Dear authors,

you have presented Semantic segmentation of urban green space. I find this article easy to follow and it is interesting to see this results. 

Among some technical issues (glued words) rest of the text is fine.

I would like if you could update Figure 1 so it shows China and Shenzhen. Although many readers know how to locate China and part of China, I feel this will give better visibility for broader audience.

Also I feel like discussion part is a bit weak. Please in discussion part reflect on cited authors in introduction, a compare your results to other authors results. Highlight novelty of your work in discussion part.

Conlusions should be supported by results, not only "This is better than this", but for example "Based on Overall accuracy CRAUNet is better than FCN8s, UNet and DeepLab V3+ with overall accuracy of  97.34%"

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors revised the article taking into account all relevant issues.