Bilateral Attention U-Net with Dissimilarity Attention Gate for Change Detection on Remote Sensing Imageries

Round 1

Reviewer 1 Report

1. Figure 1: Quality of images can be improved

2. Preprocessing methods not presented in paper. When we take any image, due to environmental disturbances, noise will be occurred. How to handle noises not presented 

3. How spatial and spectral information are differentiated not presented in paper 

4. Spectral information means , wavelength information not presented in paper 

5. Computational time not presented in paper when compared with other existing methods 

6.higher computational burden will be occurred compared with attention U-Net owing to additional processes of bilateral dissimilarity encoding and DAG

Over all the work is good. The paper can be approved for publication 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

 

Review：

This work proposes a bilateral attention U-Net with a dissimilarity attention gate (DAG) for change detection on remote sensing images.

1、 The references are old, and some of the references have years with bold font, but some are not, the author should correct them with consistent format.

2、 The title of introduction misses the 1.

3、 Figure 5 is unclear, please replace it.

4、 The hardware and software configuration of the model training and test must be added

5、 Table2-5, the best scores should be marked with bold.

6、 According to Table 2 and Table 3, along with Table 4 and Table 5, I understand that the proposed method shows better performance under the condition when the input images are reversed, which is the core idea of this manuscript. Does the reference methods share the same training dataset with the proposed method? In the training process, how about reversing the input dataset as the online data augmentation method? It seems a simple data augmentation approach can achieve the similar performance.

 

7、 In line 358, the author claims: Conventional methods, such as U-Net, ATTUNet, and Modified-UNet++ using… I think these methods are not conventional methods, it should belongs to Ablation experiments. The classic conventional segmentation methods, such as FPN, Linknet, PSPNet, should be included to verify the performance of the proposed Unet based architecture. And the existing experiments should be rewritten as ablation studies.

Author Response

Please see the attachment.

Author Response File: Author Response.docx