A New Deep Learning Model for the Classification of Poisonous and Edible Mushrooms Based on Improved AlexNet Convolutional Neural Network

Round 1

Reviewer 1 Report

Please define clear what is a goal of the paper, you propose a new model or you do comparison of the existent models.

Please, check englisch grammar.

Please also think about structure of the paper, in the section 2 you describes the basics of CNN, then in section 3 compares nets with yours. All these schould be in my opinion in "Materials and methods" section in different subsections. In section 3 should be results with mashroom dataset.

Make also images sharper, by zooming they are blured. You can use vector format to save images.

The figure 13 is also saved in a bad format, the image is blured by zooming. It is hard to see the axis numbers.

Tables are also copied and have bad quality. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The article presents a valuable addition to CNNs to the food industry. The application itself is quite challenging. The results are convincing and valuable. The proposed algorithm is simple, but the accuracy is suitable for real-time monitoring. The manuscript is well structured. The reviewer, therefore, recommends publishing this paper in the journal after answering and correcting the comments below:

1. Throughout the paper, it was noticed that the image quality is extremely poor. The authors either copy-pasted it from different sources or saved low-quality images.
2. Section 3 should only consist of the proposed method, which is what authors have proposed in their research, so it makes 3.1,3.2, and 3.3 irrelevant in this section, and the reviewer suggests moving these in section 2.
3. Lines 115 to 120 should be summarized as a table.
4. Section 3.6 is very vague. More explanation should be given about the confusion matrix and its formulation.
5. Figure 13 is one of the research's primary results, and the reviewer failed even to understand the scale. Kindly adjust all the images.
6. There is no clear explanation of how the authors and their proposed algorithm is differentiating between poisonous and edible mushroom. The results are not sufficient. There should also be a comparison of errors in the form of an error bar plot. Without an expression of uncertainty of the fit parameters (and error bars in the plot), it is impossible to tell if the discrepancy between experimental data and the fitted line is substantial or compatible with the measurement error. The text is not enough to support the claims of the authors.
7. Please revise English throughout the article.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The article proposes a new deep learning model for the classification of poisonous and edible mushrooms based on improved AlexNet CNN

AlexNet CNN is probably one of the simplest methods to understand deep learning concepts and techniques.
AlexNet is not a complicated architecture compared with some state of the art CNN architectures that have emerged in the more recent years, so it is exciting for a broad community to see how we can improve the model to deal with new tasks. Practical examples of using AlexNet CNN are always welcome. 
AlexNet is regarded as one of the most influential papers published in computer vision. It has stimulated the production of many more research papers using CNNs and GPUs to accelerate deep learning. Considering the importance of every new investigation in this domain, the article presenting the research needs to be clear and precise.

My concern begins with the introduction of the paper.

Pooling is a key-step in convolutional based systems that reduces the dimensionality of the feature maps, as you stated but:
there are more types of pooling than max and average pooling. Those types are most commonly used, but you forgot to mention other popular methods like Mixed pooling, ?? Pooling, Stochastic Pooling, Spatial Pyramid Pooling, and Region of Interest Pooling. There are also other novel pooling methods like Multi-scale order-less pooling, Super-Pixel Pooling, PCA networks, Compact Bilinear Pooling, Lead Asymmetric Pooling, Edge-aware Pyramid Pooling, Mixed Pooling, Spectral Pooling, Row-wise Max Pooling, Inter-map Pooling, Rank-based Average Pooling, Per Pixel Pyramid Pooling, Weighted pooling, and Genetic-based Pooling, among others. 

In the section 2.1 I advise you to add references covering the pooling topic since it is crucial for your work and explain why you have not evaluated other pooling methods.

In the section 2.2. of your paper R-CNN is proposed for object detection and drawing boxes around objects. Other methods are mentioned, but it is not clear why the R-CNN is chosen as it still takes a considerable amount of time to train the network as you would have to classify 2000 region proposals per image. Another problem of the R-CNN is that the selective search algorithm used is a fixed algorithm; therefore, there is no learning happening at that stage that could lead to the generation of lousy candidate region proposals. 

You should explain the methodology that led you to choose the proposed R-CNN. 

In section 3.4. You describe the proposed model, but you do not mention the transfer learning technique you are using. You should clarify that the transfer learning technique is a machine learning technique where a model is trained and developed for one task and then re-used on a second related task. 

It is not clear from the description how have you chosen the proposed CNN architecture. For example, what was the chosen methodology and why?  

Comparing obtained results is therefore not relevant.

Taking all mentioned into account, I don't find your paper ready for publication at this moment.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The figures are still in a bad quality, there is no improvements of them.

Please, check English grammar through the paper.

Author Response

All images have been changed to higher resolutions and successfully corrected and checked English grammar, and check through English ending service. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

I accept the publication 

Author Response

The authors thank you for reading and comment that is helpful in improving our work. And successfully corrected and checked English grammar, and check through English ending service. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Thank you for the changes made in the manuscript.

Author Response

The authors thank you for reading and comment that is helpful in improving our work. And successfully corrected and checked English grammar, and check through English ending service. Please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Thanks for the chnages. The images are better but not really in a sufficient quality. My suggestion is to use vector format for the images in further publications.

Author Response

Thank you for your useful suggestion in better quality of all images. I've made the requested edits, I hope the manuscript is now satisfactory. All image in this manuscript were changed into vector format already.