Siam-EMNet: A Siamese EfficientNet–MANet Network for Building Change Detection in Very High Resolution Images

Round 1

Reviewer 1 Report (Previous Reviewer 1)

The paper can be accepted in its current form.

Author Response

Thank you for your valuable comments on this article.

Reviewer 2 Report (Previous Reviewer 2)

The authors have addressed all my concerns. This manuscript can be published as it is.

Author Response

Thank you for your valuable comments on this article.

Reviewer 3 Report (Previous Reviewer 4)

You must take all suggestions into consideration. If you disagree with any suggestion, please present the reasons in an extra document.

 

Please, ask a native English speaker to review your article.

 

The aim of the paper is to describe a siamese EfficientNet B4-MANet network (Siam-EMNet) for building change information extraction of VHR images. The paper contributes to help to better predict building change regions and acquire higher quality prediction results, meanwhile pre-trained weights are used to make the experimental results converge more stably. The paper also contributes providing a change detection network for which the details are recovered and improved in the up-sampling process, the edge information of the changing regions can be detected and the missed detection of small regions can be effectively avoided. Moreover, the research reduce the detection error caused by the imbalance of change and unchanged samples.

 

 

All the following comments are my suggestions to the authors.

 

 

Lines 14 and 15: Please replace:

 

 

“As very high resolution (VHR) remote sensing technology and deep learning, building change detection methods have made great progress.”

 

with

 

“As well as very high resolution (VHR) remote sensing technology and deep learning, methods for detecting changes in buildings have made great progress.”

 

 

 

 

Line 75: Please replace:

 

 

“Peng et al. [27] Since”

 

with

 

“Peng et al. [27]. Since”

 

 

 

Line 88: Please replace:

 

 

“The DASNet [34]network introduces”

 

with

 

“The DASNet [34] network introduces”

 

 

 

 

Line 121: Please replace:

 

 

“explain”

 

with

 

“explains”

 

 

 

Line 129: Please, consider reviewing (English language grammar): “compound scaled based.”

Are you referring to the “compound scaling method” here?

 

 

 

Line 158: If you mean “only one among the three”, so please replace:

 

 

“only one of the width, depth and resolution”

 

with

 

“only one out of the three either width, depth or resolution”

 

 

 

 

Lines 160 and 161: Please replace:

 

 

“network width, depth and image resolution”

 

with

 

“network width, depth, and image resolution”

 

 

 

 

Line 162: Please replace:

 

 

“EfficientNet B7 .”

 

with

 

“EfficientNet B7.”

 

 

 

Lines 175 to 179: You could add some comments about Layers and Channels here.

 

 

Line 183: Please, consider replacing “,” with “;” in “resolution, its”

 

 

 

Line 233: Please replace:

 

 

“respectively. its”

 

with

 

“respectively. Its”

 

 

Lines 244 and 245: please, pay the attention to some absences of spaces.

 

 

 

Lines 245 to 246: Please, make the following sentence clearer:

“, and (...), and (...) is”.

 

 

 

Lines 285 and 286: Please replace:

 

 

“Mobilenet V2, XCEP, Resnet34 and VGG13”

 

with

 

“Mobilenet V2, XCEP, Resnet34, and VGG13”

 

 

 

Lines 301 and 302: Please replace:

 

 

“TN, TP, FN, and FP in (15), (16) and (18) are the numbers of true negatives, true positives, false negatives and false positives, respectively.”

 

with

 

“TN, TP, FN, and FP in (15), (16), and (18) are the numbers of true negatives, true positives, false negatives, and false positives, respectively.”

 

 

 

 

Line 311: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 2. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

 

Line 327: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 3. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

 

 

 

 

 

Line 333: Please replace:

 

 

“precision, accuracy and F1-score”

 

with

 

“precision, accuracy, and F1-score”

 

 

 

Line 336: Please replace:

 

 

“which has a large area missed detection”

 

with

 

“which has a large area of missed detection”

 

 

 

Line 345: Are [a] and [b] 1024×1024 or 256×256 pixels images? Please, add this information here.

 

 

 

 

Lines 345 and 346: Please replace:

 

 

“[a] T1 image [b] T2 image [c] Truth map [d] BIT [e] CDNet [f] DSIFN [g] L-Unet [h] P2V-CD [i] SNUNet and [j] Siam-EMNet(ours).”

 

with

 

“[a] T1 image, [b] T2 image, [c] Truth map, [d] BIT, [e] CDNet, [f] DSIFN, [g] L-Unet, [h] P2V-CD, [i] SNUNet, and [j] Siam-EMNet(ours).”

 

 

 

Lines 351 and 352: Please replace:

 

 

“BIT, DSIFN and P2V-CD models significantly missed detection. In contrast, the performance of L-Unet, P2V-CD and SNUNet”

 

with

 

“BIT, DSIFN, and P2V-CD models significantly missed detection. In contrast, the performance of L-Unet, P2V-CD, and SNUNet”

 

 

 

Line 355: Please replace:

 

 

“of proposed method”

 

with

 

“of the proposed method”

 

 

 

Line 359: Please replace:

 

 

“precision, accuracy and F1-score”

 

with

 

“precision, accuracy, and F1-score”

 

 

 

 

Line 363: Are [a] and [b] 1024×1024 or 256×256 pixels images? Please, add this information here.

 

 

 

 

Lines 363, 364, and 365: Please replace:

 

 

“[a] T1 image [b] T2 image [c] Truth map [d] BIT [e] CDNet [f] DSIFN [g] L-Unet [h] P2V-CD [i] SNUNet and [j] Siam-EMNet(ours).”

 

with

 

“[a] T1 image, [b] T2 image, [c] Truth map, [d] BIT, [e] CDNet, [f] DSIFN, [g] L-Unet, [h] P2V-CD, [i] SNUNet, and [j] Siam-EMNet(ours).”

Lines 385 and 386: Please replace:

 

 

“recall, accuracy and F1-score”

 

with

 

“recall, accuracy, and F1-score”

 

 

 

Line 391: Are [a] and [b] 1024×1024 or 256×256 pixels images? Please, add this information here.

 

 

 

Lines 391 and 392: Please replace:

 

 

“.[a] T1 image [b] T2 image [c] Truth map [d] BIT [e] CDNet [f] DSIFN [g] L-Unet [h] P2V-CD [i] SNUNet and [j] Siam-EMNet(ours).”

 

with

 

“. [a] T1 image, [b] T2 image, [c] Truth map, [d] BIT, [e] CDNet, [f] DSIFN, [g] L-Unet, [h] P2V-CD, [i] SNUNet, and [j] Siam-EMNet(ours).”

 

 

 

Lines 406 and 407: Please replace:

 

 

“A higher the Recall value represents a low miss rate of the model, and a more complete detection of the change regions.”

 

with

 

“The higher the Recall the lower the missed rate of the model, consequently more complete the detection of the change regions.”

 

 

 

Line 409 “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 7. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

Line 421: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 8. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

Lines 433 to 454: please, review (English language grammar) all lines of the section “Conclusions.”

Comments for author File: Comments.pdf

Please, ask a native English speaker to review your article.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report (Previous Reviewer 4)

The aim of the paper is to describe a siamese EfficientNet B4-MANet network (Siam-EMNet) for building change information extraction of VHR images. The paper contributes to help to better predict building change regions and acquire higher quality prediction results, meanwhile pre-trained weights are used to make the experimental results converge more stably. The paper also contributes providing a change detection network for which the details are recovered and improved in the up-sampling process, the edge information of the changing regions can be detected and the missed detection of small regions can be effectively avoided. Moreover, the research reduce the detection error caused by the imbalance of change and unchanged samples.

 

 

All the following comments are my suggestions to the authors.

 

 

Lines 120 and 121: if you really refer to the “compound scaling method” here, please, replace:

 

“To improve the accuracy of model, the structure is compound scaling method on the network's width, depth, and resolution.”

 

with

 

“To improve the accuracy of the model, the structure uses the compound scaling method on the network's width, depth, and resolution.”

 

 

Otherwise, make clearer the sentence “To improve the accuracy of model, the structure is compound scaling method on the network's width, depth, and resolution.”

 

 

 

 

Line 330: Please replace:

 

 

“(256 pixels × 256 pixels) [a] T1 image”

 

with

 

“[a] T1 image (256 pixels × 256 pixels)”

 

 

 

 

Line 349: Please replace:

 

 

“(256 pixels × 256 pixels) [a] T1 image”

 

with

 

“[a] T1 image (256 pixels × 256 pixels)”

 

Line 376: Please replace:

 

 

“(256 pixels × 256 pixels) [a] T1 image”

 

with

 

“[a] T1 image (256 pixels × 256 pixels)”

 

 

Lines 419, 420, and 421: Please replace:

 

 

“To solve the problems of incomplete detection of change areas and rough edges in the field of building change detection. A VHR remote sensing image building change detection network based on Siamese EfficientNet B4-MANet (Siam-EMNet) is proposed.”

 

with

 

“A VHR remote sensing image building change detection network based on Siamese EfficientNet B4-MANet (Siam-EMNet) was proposed to solve the problems of incomplete detection of change areas and rough edges in the field of building change detection.”

 

 

 

 

Line 426: Please replace:

 

 

“by skip”

 

with

 

“by skipping”

 

 

 

Line 428:

 

When you write “The experimental on the LEVIR-CD dataset results of Siam-EMNet are effective”, do you mean “Results obtained using the LEVIR-CD dataset show that Siam-EMNet is effective” ? If so, please replace:

 

“The experimental on the LEVIR-CD dataset results of Siam-EMNet are effective”

 

with

 

“Results obtained using the LEVIR-CD dataset show that Siam-EMNet is effective”

 

 

Otherwise, make clearer the sentence “The experimental on the LEVIR-CD dataset results of Siam-EMNet are effective”.

 

 

 

Line 432: Please replace:

 

 

“the WHU-CD dataset is used to verify”

 

with

 

“the WHU-CD dataset was used to verify”

 

 

 

 

 

Lines 437, 438, and 439: Please delete (remove):

 

“In addition, two published data sets are used in this paper, which have been pre-processed such as atmospheric correction to eliminate atmospheric effects. We will also verify this model on the large data sets in our subsequent work.”

 

Comments for author File: Comments.pdf

Author Response

Please see the 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

In this paper, the Authors investigate on an encoder-decoder architecture relied upon siamese EfficientNet B4 for building change detection of very high-resolution remote sensing images. The addressed problem is certainly a relevant issue for the Community.

In the opinion of this reviewer, methodological speaking, the novelty of the paper can be considered limited. The encoder of the proposed network is inspired by EfficientNet B4 [38] and the decoder is the same as in [44]. Instead, the loss function is a combination of two well-known loss functions. Thus, from this point of view, the novelty is limited. Anyway, the other side of the coin is represented by the proper combination of these modules to solve a problem that has not been considered in the previous works [38] and [44]. The paper is generally well-organized, but some typos and mistakes can be found along the whole paper (see, just for an instance, Page 3, Line 127: “The encoder-decoder based network is highly accuracy in feature extraction [37]. The …”). The experimental analysis is solid enough to support the proposed approach. Hence, even considering the promising performance, this work can meet the highest standards for a publication on MDPI Remote Sensing.

A couple of concerns to be addressed are as follows:

-        Page 3, Lines 105-106. The Authors report as contribution of their paper: “A decoder with integrated PAB and MFAB to up-sampling the feature mapping of the encoder.”.  This is not a contribution of the paper because it has already been proposed in [44], i.e., in another architecture using a different encoder. Please, delete this sentence. Thus, the contribution for Point (2) is just the Siam-EMNet change detection network.

-        The benchmark to assess the performance of the proposed approach should be improved. More specifically, please delete FC-Siam-conc [28] and FC-Siam-diff [28] that get results far from the optimal ones, substituting them with two recent state-of-the-art change detection approaches published on top journals.

The Authors should polish the paper removing typos/mistakes.

Reviewer 2 Report

In this work, a building change detection method of VHR RS images is proposed based on a Siam-EMNet network. Overall, the work is soundness, technically correct, and the results seems convincible. I have the following concerns.

1.      In section 2.1, the overview of former works should be reorganized in Introduction.

2.      The statement of “The best performance is obtained when β= 0.5 and α=2”. Please explain in detail with experimental analysis.

3.      Grammatical slips should be corrected in the whole paper, for example, “is compound scaled”, “is highly accuracy”, etc.

4.      What does SOTA mean? Please provide the full name of abbreviation when it first appears.

5.      According to the summarized deficiencies of former works, please provide the discussion of the proposed work regarding the overcome of atmospheric condition, inadequate feature extraction, and insufficient spatial information.

Grammatical slips should be corrected in the whole paper.

Reviewer 3 Report

 

In this work, the authors  propose a siamese EfficientNet B4-MANet network (Siam-EMNet) to build change information extraction of VHR images which uses bitemporal images as inputs and outputs binary detection maps.

 

General remarks concerning this article:

-       The abstract give an overview of the methodology used and the authors discuss the results correctly

-       The contribution is compared with other survey methods

-        The references respect the structure of research paper and highly updated.  

-       The quality of figures is high

-       The some expressions of the paper are hoped to improve in the future works.

-        In summary, the paper is very convincing. The authors propose a theoretical and practical model.

There are many grammatical errors, confusing and repetitive sentences, which degrade the readability of the paper

Reviewer 4 Report

 

The aim of the paper is to describe a siamese EfficientNet B4-MANet network (Siam-EMNet) for building change information extraction of VHR images. The paper contributes to help to better predict building change regions and acquire higher quality prediction results, meanwhile pre-trained weights are used to make the experimental results converge more stably. The paper also contributes providing a change detection network for which the details are recovered and improved in the up-sampling process, the edge information of the changing regions can be detected and the missed detection of small regions can be effectively avoided. Moreover, the research reduce the detection error caused by the imbalance of change and unchanged samples.

 

 

All the following comments are my suggestions to the authors.

 

 

Careful not to confuse "Accuracy" and "Precision" in the article, mainly in the section “Experimental results and analysis”:

 

Accuracy measures the efficiency of results and is represented by (TP+TN)/M, for which M is either the number of images or pixels.

 

Precision measures the relevancy of results and is represented by TP/(TP+FP).

 

 

 

 

All lines: Please, do not activate the tool to hyphenate.

 

 

Line 15: Please, consider replacing “leaps and bounds” with more formal terms.

 

 

Line 37: What do you really mean here: “can detect changes” or “can be used to detect changes”?

 

 

Lines 45 and 46: Please replace:

 

“where the first method utilizes the difference of bands between images and obtains the change results by clustering or thresholding”

 

with

 

“the former method using difference between images and obtaining change results by clustering or thresholding”

 

 

Line 47: Please replace:

 

 

“to processing of the images”

 

with

 

“to process images”

 

 

 

Line 69: Please, consider replacing “injected new exuberant life” with more formal terms.

 

 

Line 72: Please, consider reviewing (English language grammar): “Peng et al. [27] Since”

 

 

Line 85: Please, consider separating “[34]network”

 

 

 

Line 113: Please replace:

 

 

“explain”

 

with

 

“explains”

 

 

 

Line 118: Please, consider eliminating the dot in “Figure 1.”

 

 

Line 121: Please, consider reviewing (English language grammar): “compound scaled based.”

Are you referring to the “compound scaling method” here?

 

 

Line 127: Please, consider reviewing (English language grammar): “is highly accuracy”

 

 

Line 144: Please, consider reviewing (English language grammar): “In most of the past research works, only one of the width, depth and resolution was”

 

 

Lines 146 and 147: Please replace:

 

 

“network width, depth and image resolution”

 

with

 

“network width, depth, and image resolution”

 

 

Line 144: Please, make the following sentence clearer:

“, and a total from EfficientNet B0 to EfficientNet B7 in total, 8 versions are proposed.”

 

 

 

Lines 161 to 165: You could add some comments about Layers and Channels here.

 

 

Line 169: Please, consider replacing “,” with “;” in “resolution, its”

 

 

Line 182: Please, cite references here “which has been maturely used for semantic segmentation of medical images”

 

 

 

Lines 189 and 190: Please replace:

 

 

"The structure of the two modules is shown below, respectively."

 

with

 

"The structures of the two modules are shown in Figure 3, PAB in (a) and MFAB in (b)."

 

 

 

 

Lines 218 to 220: Please, consider reviewing (English language grammar): “are achieved using the global average pooling, statistics are denoted as S1 and S2, respectively. its kth pixel can be expressed as:”

 

 

 

Line 225: please, pay the attention to some absences of spaces.

 

 

 

Line 226: Please, consider reviewing (English language grammar): “, respectively. the”

 

 

 

Lines 230 and 231: please, pay the attention to some absences of spaces.

 

 

 

Lines 231 to 233: Please, make the following sentence clearer:

“, and (...), and (...) is obtained from Vk and the feature mapping (...) between obtained by multiplying them channel by channel.

 

 

Line 226: What do you mean here: “connecting with (...) and (...) this paper”?

 

 

Lines 236 and 237: please, consider reviewing (English language grammar): “change and unchanged samples.”

 

 

Lines 254 and 255: Please, consider reviewing (English language grammar): “used. [30] This”

 

 

 

Lines 268 and 269: Please replace:

 

 

“Mobilenet V2, XCEP, Resnet34 and VGG13”

 

with

 

“Mobilenet V2, XCEP, Resnet34, and VGG13”

 

 

 

Lines 268 and 269: Please, cite references here: “Mobilenet V2, XCEP, Resnet34, and VGG13”

 

 

 

Lines 270 and 271: Please replace:

 

 

“SNUNet [29], BIT [10], DSIFN [30], CDNet [47], FC-Siam-conc [28] and FC-Siam-diff [28].”

 

with

 

“SNUNet [29], BIT [10], DSIFN [30], CDNet [47], FC-Siam-conc [28], and FC-Siam-diff [28].”

 

 

 

Line 279: Please, add “Accuracy” as an evaluation index.

 

 

Line 280: Please replace:

 

 

“formulas”

 

with

 

“equations”

 

 

 

Line 280: Please, add the equation: Accuracy = (TP+TN)/M, for which M is the number of images.

 

 

 

 

Line 281: Please replace:

 

 

“TP, FN and FP”

 

with

 

“TP, FN, and FP”

 

 

 

 

Line 287: Please replace:

 

 

“accuracy”

 

with

 

“precision”

 

 

 

Line 291: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 2. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

Line 294: Please replace:

 

 

“SNUNet, BIT, DSIFN, CDNet, FC-Siam-conc and FC-Siam-diff”

 

with

 

“SNUNet, BIT, DSIFN, CDNet, FC-Siam-conc, and FC-Siam-diff”

 

 

 

Line 296: Please replace:

 

 

“A comparison of Table 3”

 

with

 

“A comparison in Table 3”

 

 

 

Line 296: Please, eliminate the dot in “Table 3. shows”

 

 

 

Line 307: In Table 3, referent to the Recall of FC-Siam-diff, please replace:

 

“7545”

 

with

 

“75.45”

 

 

Line 307: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 3. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

Line 309:

In lines 254-256, you wrote: “The LEVIR-CD of the building change detection dataset (...). This dataset consists of 637 pairs of 1024 pixels × 1024 pixels VHR Google Earth images.” Besides writing “three scenarios”, please, write here also how many pairs of 1024×1024 pixels images from the 637 ones of the dataset you used for performing your experiments.

 

 

 

Lines 315 and 316: Please replace:

 

 

“SNUNet, BIT, FC-Siam-conc models building detection is more complete”

 

with

 

“SNUNet, BIT, and FC-Siam-conc models building detections are more complete”

 

 

 

Line 323: Are [a] and [b] 1024×1024 or 256×256 pixels images? Please, add this information here.

 

 

Line 328: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 4. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

Line 335: Please, replace:

 

 

“SNUNet, BIT, DSIFN and CDNet”

 

with

 

“SNUNet, BIT, DSIFN, and CDNet”

 

 

 

Line 336: Please replace:

 

 

“accuracy”

 

with

 

“precision”

 

 

 

Line 344: Are [a] and [b] 1024×1024 or 256×256 pixels images? Please, add this information here.

 

 

 

Line 351: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 5. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

Line 356: Please, replace:

 

“6”

 

with

 

“six”

 

 

 

 

Line 359: Please, replace:

 

 

“SNUNet, BIT, CDNet, FC-Siam-conc detection has”

 

with

 

“SNUNet, BIT, CDNet, and FC-Siam-conc detections have”

 

 

 

Line 360: Please, consider replacing “information, different” with “information. Different”

 

 

 

Lines 263 and 267: Please eliminate all the redundancies in:

“Siam-EMNet method effectively avoids the influence of roads and shadows in the background, and detects the building contours. The Siam-EMNet method effectively avoids the influence of roads and shadows in the background, detects the complete edge information of buildings, has a strong anti-interference property, and has better edge detection results.”

 

 

Line 368: Please, replace:

 

“7”

 

with

 

“seven”

 

 

 

Line 369: Please, replace:

 

“methods”

 

with

 

“method’s”

 

 

 

 

Line 370: Please, replace:

 

“6”

 

with

 

“six”

 

 

 

Line 371: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 6. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

Line 376: Are [a] and [b] 1024×1024 or 256×256 pixels images? Please, add this information here.

 

 

 

Line 385: Please, write here also how many images from the WHU-CD dataset you used for performing your experiments.

 

 

Line 389: Please, consider reviewing (English language grammar): “the proposed method of Recall and F1-score”

 

 

Lines 390 e 391: Please, consider reviewing (English language grammar): “The higher the Recall value represents the low missed rate of the model, and the more complete the detection of the change regions.”

 

 

Line 393: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 7. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

Line 396: Please, replace:

 

“8”

 

with

 

“eight”

 

 

 

Line 405: “Accuracy” and “Precision” are different metrics. Please, add a new column entitled “Accuracy” to Table 8. Please, fill in the column “Accuracy” with the respective and correct values.

 

 

 

Line 414: please, consider reviewing (English language grammar): “change and unchanged samples.”

 

 

 

Lines 417 to 434: please, review (English language grammar) all lines of the section “Conclusions.”

Moderate editing of English language required.

The article must be proofread by a native English speaker.