Deep Learning-Based Layer Identification of 2D Nanomaterials

Round 1

Reviewer 1 Report

The authors of this study applied deep learning to identify 2D nanomaterial layers using a semantic segmentation model. It is very important to study 2D materials to conduct research of this type with very high accuracy. This study examined 16 semantic segmentation models and analyzed their specific effects on 2D nanomaterial OM images.  As a result of this manuscript, some exciting findings have been presented. I recommend this paper for publication after the following comment is addressed.

1. Since the authors claimed in the conclusion, “these trained semantic segmentation models can also efficiently identify 2D nanomaterials other than those used for training, with good generalization ability and high accuracy.” What is the model's generalization capability with other 2D materials like MOFs, COFs, MXenes, etc.? It is crucial if this model can be applied to various types of 2D materials. Did the authors conduct any preliminary experiments? Having some results and discussion would enhance the impact of the paper, although accuracy is likely to be lower.

Author Response

Firstly, we would like to thank the reviewers for giving us the opportunity to revise the manuscript. Secondly, we have carefully read the questions and suggestions given by the reviewers and have thought and analyzed them in depth. Finally, we have carefully revised the manuscript according to the reviewers' questions and suggestions.

We greatly appreciate the reviewer’s comments, which will help us in our next research work. For 2D materials that are chemically close and belong to the same general class, the model can show a strong generalization capability with good data and labels. The model is able to fit the data well and achieve the desired segmentation results for new data images of the same 2D material that appear in the actual task. For different classes of 2D materials with widely varying chemical properties, the generalization ability of the model is weak. But this can be alleviated by adjusting the model structure and optimizing the model training and deployment.In conclusion, these semantic segmentation models based on deep learning can have good generalization ability for the layer recognition of 2D materials

These deep learning-based semantic segmentation models can be applied to most different kinds of 2D materials. Firstly, these models achieve good segmentation results on a variety of large datasets and are able to segment dozens of different classes of objects. As a result, these models have the ability to cope with complex scenarios and complex problems. This has been demonstrated for medical image segmentation, novel material prediction and large molecule structure prediction. Secondly, with good data and labeling of different types of 2D materials, it is possible to achieve segmentation capabilities that are far superior to those achieved manually. For 2D materials with no or few labels, unsupervised or semi-supervised learning can be used for training, a process that can be useful in the study of layer identification of 2D materials. For 2D materials with difficult and small data acquisition, semantic segmentation models based on small samples can also solve this problem to a certain extent. In conclusion, deep learning-based semantic segmentation models can solve the problem of layer recognition for different kinds of 2D materials, as demonstrated in some recent studies.

Finally, the difficulty of layer recognition in 2D materials lies in the collection, labeling and processing of data samples. At present we have only collected data images related to graphene and molybdenum disulphide. Therefore, we have not yet conducted experiments on other types of 2D materials. However, we are in the process of collecting data on more types of 2D materials and intend to test the model's generalization capability.

Reviewer 2 Report

The manuscript is well structured and the approach is suitable for the aim. The study is of good relevance since two-dimensional (2D) nanomaterials due to their high anisotropy and chemical functions have attracted increasing interest and attention from various scientific fields, including functional electronics, catalysis, supercapacitors, batteries and energy materials. In this study, authors analysed sixteen semantic segmentation models that perform well on public datasets and applied them to the layer identification and segmentation of graphene and molybdenum disulfide. The results showed that deep learning-based semantic segmentation methods can greatly improve efficiency and replace most of the manual operations, and different types of semantic segmentation methods can be adapted to different properties of 2D nanomaterials, thus promoting the research and application of 2D nanomaterials.

I have no objections or suggestions to the manuscript and I propose the manuscript for publication in present form.

Author Response

Thank you very much for your affirmation.

Reviewer 3 Report

The paper by Yu Zhang et al  presents novel and interesting results that deserve to be published afterminor improvement:

The titles of references have a different format, the title of the article is written in capital letters at the beginning of words, others only in lower case. Also, the standardizedformat of presentation in the journal's name. Becase names have written in a different.

Author Response

Many thanks to the reviewers for their comments. We have gone through the manuscript carefully in the light of the comments and have made careful revisions to the formatting of all the references------Please see Page 16-19.

The title of the article has been revised------Please see Page 1

The journal’s name has been put on the right place-----Please see Page 1