Research on Optimization of Processing Parcels of New Bare Land Based on Remote Sensing Image Change Detection

Round 1

Reviewer 1 Report

This paper uses a simple Siamese CNN to obtain the probability distribution maps of dual-temporal images, and then designs a new process to obtain the final change detection results. The article describes in detail a processing parcels of new bare land method, in which the main innovation is to propose three metrics, Confidence-based semantic similarity, Spatial proximity, and Spatial proximity. The paper is innovative and the experimental demonstration is sufficient. The main problems are as follows:

1) Why do you think and can represent semantics, and what is the relationship between this confidence score of the parcel and semantics

2) Would it be better if the three coefficients were set to =1, =0, and =0. If better, are the other two indicators meaningless? The effect of different coefficients on the results should be further discussed

3) In Equation 3, the format of  = 0.5 is different from the above of  =1

4) Deep learning usually uses metrics such as precision, recall, etc. The impact of the processing on these metrics can be further discussed in the text

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The proposed manuscript presents a method for automatic feature extraction, using a general deep Siamese CNN change detection model, applied to the detection of new bare land parcels. The proposed aggregation is compared with the ArcGIS aggregation tool and illustrated with multi-sensor mosaic images of the experimental area. The authors presented the results of both their approaches, initial and enhanced, showing a decrease in the total number of extracted parcels by more than 50% and the false detection rate (FDR) by approximately 30%. However, the FDR is still pretty high (44-47%), as well as MDR (30-33.5%), which, as the authors frankly admitted, leaves plenty of room for further improvements.

The paper is clearly and comprehensibly written. Figures need to be organized so that all parts of the same figure are on the same page. Figure 8 should have additional explanations of what each column represents. The meaning of Figure 9 is not very clear at the time it appears. Complete the image label with data from the text describing it.

Table 2 should be reduced to fit on one side, or adequately divided into two, but so that the header can be seen on the other side, as well as the continuation of the label.

 

Some tiny typing errors:

ARCGIS -> ArcGIS (line 329)

MIOU -> MIoU  (line 353)

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper presents the authors’ work on developing a deep-learning-based framework for detecting bare land for the purpose of land use monitoring. Overall, this paper presents an important research topic. I have no question about the methodology of this paper, but the presentation of this paper should be improved to increase readability. The detailed comments are as follows.

1) Explain ‘constructed bare land’ and their possible sources.

2) Figure 7: Figure 7(a) and Figure 7(b)

3) Figure 8 and Figure 11:  explain what these images are, consider using a table of figures

4) The first time an acronym is used, give the full version, e.g., MIoU

5) Line 337: ‘As shown in (13)’, ‘(13)’ should be ‘(7)’

6) Explain a new concept the first time it is introduced, e.g., resnet101 and resnet152.

7) ‘Change information extraction’ can be replaced by ‘Change detection’

8) The overall writing quality should be greatly improved. For example,

a) Line 30: ‘illegal land occupation for natural resource is becoming increasingly prominent.’ -> ‘illegal land occupation is becoming…’

b) Line 43: ‘satellite images at different times’ -> ‘satellite images at different points in time’

Author Response

Please see the attachment.

Author Response File: Author Response.pdf