论文修改建议（WangLR 20211217细节问题）

1. Examples include multi-channel singular spectrum analysis (MSSA) [37,38], damping order reduction method (DRR) [39,40]
   $\to$
   method (DRR) [39,40]后缺句号

2. This makes it possible for CNN to denoise seismic data [56–59].
   $\to$
   这个不确定，用It makes会不会好点

3. According to different network models, the existing denoising
   algorithms can be roughly divided into three categories. The
   first is
   $\to$
   现有去噪算法分为三类，第二类三类分段，第一类不单独成段会不会不太对称

4. The decoding part converts the generated fixed vector into an output sequence. Autoencoder(AE) [60,61] is a typical encoding-decoding framework with numerous variations.
   $\to$
   介绍编码解码框架的部分，解释完编码解码，直接就摆出几种具体框架，会不会不太友好。可不可以加一句，目前流行的这三种框架，然后再具体说明每种。

5. Becuase the only connection of the existing encoding-decoding framework is
   $\to$
   Because 拼写错误

6. The third is a denoiser based on the Adversarial Generative Network (GAN).
   $\to$
   Adversarial Generative Network 前两个写反了哦

7. A denoising device that uses the powerful learning ability of GAN to construct a denoising device [64–67].
   $\to$
   此句表述有误

8. The other uses GAN to model the noise distribution and generate rich training data to generate rich training data to achieve blind denoising [68].
   $\to$
   ① The other特指两个或两部分中的另一个或另一部分，表示两个中“一个……另一个……”，常用结构式为：one…the other…
   此处是说了两种，可否改成上述结构更好。
   ② generate rich training data to generate rich training data 写重复了

9. However, the only connection of the encoding-decoding framework is a fixed-length vector,
   $\to$
   个人感觉，这个however有点突兀。前面只说提了一个新的网络，也没铺垫什么，然后就转折。

10. However, the only connection of the encoding-decoding framework is a fixed-length vector, making it difficult for the decoder to obtain sufficient information from the input data. ①
    $\to$
    Becuase the only connection of the existing encoding-decoding framework is a fixed-length vector, making it difficult for the decoder to obtain sufficient information from the input data.②
    这里的第①句与前文中介绍第二类编码解码框架部分的最后一句②几乎完全相同，没问题吗

11. patching, residual channel attention block and encoding-decoding sub-network, and supervised attention module.
    $\to$
    这里的 residual channel attention block到底是blocks还是block，前面介绍的时候是blocks。(后面也有这个问题，请将所有确认一致)
    闵老师说不滥用and，这里的多个and是否合适

12. The patch technology uses a H×W fixed window to slide 2-D
    $\to$
    这里的符号系统是否要跟前面的（d=s+n） 保持一致

13. H and W denote the width and height of the patch, respectively
    $\to$
    先说的H，应该先是height
    but the only difference is that there is no overlap
    $\to$
    这句怪怪的

14. The second part, the excitation operation includes two 1×1 convolution operations as follows:
    o=δ(Conv(ReLU(Conv(z)))),(3)
    $\to$
    ReLU也算是function，要不要和前面的函数的符号系统一致，δ在图中是sigmoid不一致

15. C. Encoder-Decoder Subnetwork
    $\to$
    Subnetwork 中间有没有-，文中有些地方有，有些没有，请确认保持一致

16. The input patch is first compressed to extract hidden features, and
    then expanded to reconstruct the input patch.
    $\to$
    and可否去掉

17. 式(5)的符号系统与前面一致否

18. The RCAB module weights the input X∈RC×H×W to obtain the weighted feature map Xfeat∈RC×H×W, （此句附近的X都有下面类似问题）
    $\to$
    这里的X与前面的符号系统不一致

19. H and W denotes height and weight,
    $\to$
    denotes ->denote

20. 式子(6)UpSampling应该是down吧，前面说的Xfeat作为down-sampling的输入。
    where bilinear up-sampling with a scaling factor of 0.5 is
    used instead of max-pooling,
    $\to$
    这句中是up也应是down
    scaling factor 还是 scale factor
    此部分叙述中的up-sampling是否应改成式子中一样的形式

21. Among them, the input of the last two RCAB modules is composed of the output of the convolutional layer and the output of the corresponding RCAB module in the encoder.
    $\to$
    two RCAB modules is composed of中的 is->are

22. 式子（8）中的符号系统与前文一致否

23. The module provides useful ground truth, and the generated attention maps further suppresses useless information.
    $\to$
    suppresses->suppress

24. The input of SAM is the output of the decoder Fin∈RC×H×W, the output includes denoised data and attention enhancement featuresFout∈RC×H×W, whereC is the number of channels,HandWare height and width respectively.
    $\to$
    the output includes -> include

25. First, Fin generates residualRS∈R1×H×Wthrough 1×1 convolution… attention enhancement feature Fout.
    $\to$
    此部分的符号系统，跟图2不太匹配，还有些正体斜体，是否粗体，请自己核对一下。
    Fin generates 单数还是复数

26. Before the algorithm starts, the input of the two stages is preprocessed.
    $\to$
    two stages is ->are

27. 式子（9）第一个和最后一个符号错了, l对应l, r 对应 r
    且请确保此处的符号系统与前文一致

28. The spliced data is decoded to get the output
    $\to$
    这里的data是单数还是复数

29. 式（16）与前面符号系统一致吗
    Con正体，前面是斜体，concat与前面的Cat属同一类吗，为什么这里首字母不大写

30. 式（20）和（21）中的(ˆIi,j,k−Ii,j,k)中是否都该有平方，是对（20）中的进行了更改，还是（20）写漏了。且式（21）中增加的那部分中的 SNR(ˆI−I) 中的I是符号使用是否有误

31. In this section, We conduct three experiments:
    $\to$
    We -> we

32. The experimental test data does not include training data, it includes four synthetic data and two field data.
    $\to$
    does
    it includes four
    复数还是单数

33. One field data (field-data-1) is selected from Stratton survey
    The other field data (field-data-2) uses the seismic data we collected, which contains 350 traces and 5,400 time samples (Figs.5(j)).
    $\to$
    is
    uses
    contains
    这两处单数还是复数

34. One group (without RCAB) removes RCAB and penalty item, and the other (with RCAB) only removes penalty item.
    $\to$
    前边用的term，是否要一致（后文也有此情况）

35. We observe that in two datasets, the algorithm with RCAB suppresses noise while protecting the effective signal well. ①
    The results show that our method can reduce the loss of effective signal while suppressing noise.②
    $\to$
    感觉while前后动词应形式一致

36. Results show that the RCAB module is very effective in protecting effective signals.
    $\to$
    一个句子出现两个effective，可否换个词

37. The test data are selected from the additional three synthetic data.
    $\to$
    前文中的data后一般用的单数形式，请确认其后一般用单数还是复数。确定后，所有的地方应一致。

38. Hence we will setα= 0.6 in the following experiments.
    $\to$
    Hence后是否缺少逗号

39. The wavelet transform has the worst denoising effect, and there is a large amount of noise leakage in the removed noise.
    $\to$
    此处表述是否有误，去除的噪声中有效信号泄漏。

40. The denoising data obtained by DnCNN and the proposed method removes most of the noise, and the signal becomes clearly visible.
    $\to$
    removes->remove

41. The RCAB module can protect the loss of effective signals
    through feature weighting.
    $\to$
    signals到底应该是复数还是单数，整篇文章都有此问题，有时候单数，有时候复数。请将所有用到的地方保持一致。
    且前文使用的是 feature weighted。

42. In addition, in complex field data, DDAE-GAN is significantly better than the denoising effect of the three.
    $\to$
    DDAE-GAN->DDAE-RCAB

43. It suppressed irrelevant information and protected valid signals through feature weighting.
    $\to$
    前文使用的是 feature weighted

44. This paper took advantages of two technologies…One was to …The second is to …The third was to avoid excessive denoising by adding penalty items after the loss function.
    $\to$
    含义不明，先说了本文利用两种技术的优点。后面有1 2 3，前2个是针对技术，应该与第3有区别。
    且此处，3个地方的时态不一致（请确认要使用何种时态）：
    One was
    The second is
    The third was

45. Experiments showed that the penalty
    $\to$
    实验结果如果是客观存在的结果，是否用一般现在时

46. The proposed method had better denoising performance than the three four benchmark algorithms in both synthetic and field data.
    $\to$
    the three four benchmark 写错了
    另外，请确保时态正确

47. Fig. 2. Residual channel attention block (RCAB).
    $\to$
    图3中的Residual channel attention block(RCAB)也是，确认每个单词首字母大写吗，全文保持一致

48. Fig. 3. Encoder-decoder subnetwork.
    $\to$
    subnetwork中请确认是否使用-，全文保持一致
    图3中的符号正斜体和文中有些不一致
    文中使用的是UpSampling和DownSampling，与图中的不一致
    请确认此图中的符号系统和文中解释的一致

49. 所有图中使用到C×H×W的地方，确认是否加粗，与文中一致

50. Fig. 4. Supervised attention module (SAM).
    $\to$
    确认此处每个单词首字母大写吗
    图4中的符号系统与文中不一致，是否加粗，是否斜体。（几乎每个符号），请自己确认一下。

51. Fig. 11. 图注中，(e) wavelet transform ,多了空格

PS：
由于是先写在记事本的，粘贴过来公式没有改，但不影响修改哈。
参考文献未检查，太多了，请自己看看哇。