Lightweight Isotropic Convolutional Neural Network for Plant Disease Identification
Round 1
Reviewer 1 Report
The manuscript "A Lightweight Isotropic Convolutional Neural Network for Plant Disease Identification" by W. Feng, Q. Song, and G. Sun is well-written. It presents an intriguing approach utilizing a convolutional neural network (CNN) model to predict plant disease identification. The authors have successfully developed a model that effectively distinguishes between healthy leaves and various diseases, enabling visual diagnosis using laboratory images and images captured from real-life scenes. The manuscript comprehensively describes the entire procedure, encompassing image collection for training and validation, image augmentation techniques, and the training and fine-tuning of the deep CNN. To the best of my knowledge, no other models are exhibiting the same level of accuracy as the one presented in this study, particularly considering the extensive database that encompasses various plant varieties and diseases.
All the acronyms should be defined the first time it appears in the text.
Regarding the review of Manuscript agronomy-2464356, where more specific comments were requested and certain topics were suggested to be addressed, upon revisiting my comments, it appears that all the points have been adequately addressed in the original revision.
It presents an intriguing approach utilizing a convolutional neural network (CNN) model to predict plant disease identification.
The authors have successfully developed a model that effectively distinguishes between healthy leaves and various diseases, enabling visual diagnosis using laboratory images and images captured from real-life scenes.
The manuscript comprehensively describes the entire procedure, encompassing image collection for training and validation, image augmentation techniques, and the training and fine-tuning of the deep CNN.
To the best of my knowledge, no other models are exhibiting the same level of accuracy as the one presented in this study, particularly considering the extensive database that encompasses various plant varieties and diseases.
This part could be added: “The conclusion appears to resemble a summary rather than providing a comprehensive analysis. To enhance it, the authors could emphasize the unique strengths of this model compared to others available and provide a more elaborate discussion on the specific areas they believe need improvement, offering detailed insights and explanations from their perspectives”.
References cover a wide range of sources, demonstrating diversity, considered appropriate to the context. Also, they are up to date, being 16 out of 48 from 2022.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors propose a Lightweight isotropic convolutional neural network based on FoldNet for plant disease images of the PlantVillage and FGVC8 dataset by adjusting its width h, depth n, and fold length d.
The author stated that the model could achieve 97.62% recognition accuracy, but its limitations are the limited data variety and the inability to detect apple leaf diseases in real-time. The method is only applicable to winter wheat spike images.
They recommend that the model effectively capture minute features of plant diseases and enhance the ability to characterize diseases. There are still many aspects that can be improved. They plan to collect more realistic scenario data to improve the model's performance further.
From a technical perspective, the work is relevant, and the results support the proposed objectives.
In order to enhance the paper, given the relevance of the subject matter and the promising results presented, I invite the authors to make the following changes in order to achieve a publication of their proposal and results:
- add a last paragraph in the introduction where the article's sections are described to orient the reader better.
- in the introduction itself, please separate some paragraphs into sections that are more accessible to the reader; separate them into more specific and related ideas, as this makes reading very difficult; although there is a lot of very relevant background information, it is relegated to the size of the paragraphs.
- As for the structure of the paper, I recommend that the authors separate the results and discussion sections.
- considering the previous comment, it is necessary, in the discussion, to mention the eventual weaknesses associated with the study carried out and to see what measures are considered to mitigate their effect.
- beyond what is stated in section 2, it is recommended to add a section on related works to show the foundational proposals of other authors and highlight the common and differentiating aspects of your proposal.
- although section 3 describes the defined work regarding the technical support and the experiments to be performed, a definition of the methodological support on which the defined work is based would be missing.
- Fig 9 requires an improvement in its resolution.
- references are current and relevant; 27/48 of the cited papers are from 2018 or later.
- links are provided to the data presented in the proposal; thinking about the study's replicability is very relevant.
The English requires a revision, aiming to facilitate the reading:
E.g. (taken from the abstract and introduction, please revise all the text)
chain of same blocks ---> chain of the same blocks
This facilitates precision agriculture applications on mobile, low-end terminals. ---> This proposal (or model) facilitates precision
agriculture applications on mobile, low-end terminals.
Precision agriculture, or smart farming, as a new technology, aims to improve the efficiency and sustainability of agriculture by using data and advances technologies to make more informed, data-driven decisions about crop production and resource management [2]. ---> As a new technology, precision agriculture or smart farming aims to improve agriculture's efficiency and sustainability by using data and advanced technologies to make more informed, data-driven decisions about crop production and resource management [2].
...the use of... ---> using
In general, early, rapid, and accurate disease identification and control can reduce the negative impact on the environment, and using fewer chemicals and pesticides can reduce water and soil contamination. ---> Early, rapid, and accurate disease identification and control can reduce the negative environmental impact, and using fewer chemicals and pesticides can reduce water and soil contamination.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This is a well written interesting paper, that should be of great interest to agricultural producers.
Throughout the paper there is mention of machine learning models, with very little description, perhaps some description as to whether they are ANN, KNN, SVM etc. In addition there is extensive use of abbreviations which are not written in the long form in the first instance.
Some suggestions:
Line 32: advanced not advances
Lines 45 - 67: abbreviations not in long form in first instance
Lines 73 - 98: Description of models may improve the paper
Line 92: remove "equally"
Line 97 [11] should not be superscript
Line 228 - 230: Remove we in all instances
Line 269: The influence of structural features of the residual neural network on its performance, is ........mapping, ...........network,
Line 273: such "a" mapping
Line 276: such "a" mapping
Line 296: change to "The residual neural network folds the backbone
Line 303: change "such a" to the
Line 317: remove "we"
Line 323: remove "we"
Line 325 remove "we"
Line 343: remove "a lot Lighter" to reduced
Line 344: remove Therefore
Figure 8 (a) labeling within folding blocks not easily readable
Line 358: GELU not in long form Gausian error linear unit
Some references to setting epsilon, should e(-2 and -5) be superscript??
Line 632: remove "to"
English is generally very good, overuse of We and therefore and minor typos and punctuation noted and in comments to authors.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf