A Survey of Crop Disease Recognition Methods Based on Spectral and RGB Images

Round 1

Reviewer 1 Report

Comments paper: “A survey of crop disease recognition methods based on spectral and RGB images”.

Zheng et al., 2026, J Imaging 4085618.

Crop disease recognition is of major concern in agriculture and horticulture. In this paper, the authors give a detailed survey of the several methods used to detect and recognize crop diseases based on spectral and RGB images.

The paper is well organised, it delivers detailed information on the existing reviews, on the publicly available datasets (table 2) and on the commonly used evaluation metrics including the formulas (table 3). Moreover, in extensive tables, detailed information is available on the spectral based traditional machine learning crop disease recognition methods and on RGB image based deep learning crop disease recognition methods. In these table the corresponding references are mentioned and also which methods deliver the best results (table 4 and 5).

The figures has been taken care off, the references are clear.

However, there are some minor remarks:

• In the introduction, in the literature overview, the first author should be mentioned before the reference number; this number can also be located at the end of a sentence or a small paragraph. The same is on p. 8 in line 225 and p.13 in line 351.
• 5, table 2 “publicly available datasets”: explain all the abbreviations used. For a reader not yet specialised in this domain and especially researchers who want to start in the topic, it is very fuzzy and certainly not informative. It is a review paper, so it is meant to be read by many people.

I recommend a minor revision before it can be accepted for publication.

Comments for author File: Comments.pdf

Author Response

Comment 1:
In the introduction, in the literature overview, the first author should be mentioned before the reference number; this number can also be located at the end of a sentence or a small paragraph. The same is on p. 8 in line 225 and p.13 in line 351.
Response 1:
Thank you for this helpful comment. We agree with the reviewer’s suggestion. Accordingly, we have revised the citation style in the Introduction (Literature Overview) section by explicitly mentioning the first author before the reference number. The same modification has also been applied to the text on Page 8 (line 225) and Page 13 (line 351). These changes improve readability and align the manuscript with standard citation conventions.
Comment 2:
p.5, Table 2 “publicly available datasets”: explain all the abbreviations used.
Response 2:
Thank you for pointing this out. We have revised Table 2 by adding a clear explanation of all abbreviations used (e.g., RGB, hyperspectral, and dataset acronyms) in the table caption/notes. This revision improves the clarity and accessibility of the table, especially for readers who are new to this research field.

Reviewer 2 Report

This paper systematically reviews crop disease recognition methods based on spectral and RGB images from 2020 to 2025. The literature collection is comprehensive, and the data summary is substantial, offering significant reference value for researchers in agricultural information technology. However, there is room for improvement in logical coherence, technical depth, and writing standards.

Specific Comments 

1. The abstract mentions an in-depth discussion of State Space Models (Mamba) and Generative AI, but the main text lacks corresponding content on these topics. Generative AI is only briefly mentioned in the final chapter.
2. The omission of "spectral images" from the keywords section is unreasonable.
3. In Figure 1, the RGB and spectral image schematics appear to be arranged by wavelength from short to long waves; however, the red band is incorrectly placed before others in the diagram.
4. In Table 3, some variables are defined in footnotes while others are defined in the main text. Please unify the definition method for all variables.
5. The structure of Chapters 4 and 5 is unclear. The paper discusses traditional methods for spectral images and deep learning methods for RGB images separately, but they should not be discussed in isolation. Both traditional and deep learning methods exist for spectral images, RGB images, and even their combination. The discussion structure should be reorganized.

Author Response

Comment 1: 
The abstract mentions an in-depth discussion of State Space Models (Mamba) and Generative AI, but the main text lacks corresponding content on these topics. Generative AI is only briefly mentioned in the final chapter.
Response 1: 
We agree that the original abstract overstated the extent of discussion on State Space Models and Generative AI. To address this issue, we revised the abstract to accurately reflect the actual scope of the manuscript. The wording has been adjusted to indicate that these topics are reviewed as emerging research directions rather than extensively discussed. This revision ensures consistency between the abstract and the main text.
Comment 2:
The omission of "spectral images" from the keywords section is unreasonable.
Response 2:
Thank you for this comment. We have added “Spectral images” to the keywords to ensure that they accurately represent the two primary image modalities discussed throughout the manuscript.
Comment 3: 
In Figure 1, the RGB and spectral image schematics appear to be arranged by wavelength from short to long waves; however, the red band is incorrectly placed before others in the diagram.
Response 3:
We appreciate the reviewer’s careful observation. Figure 1 has been corrected by reordering the bands according to increasing wavelength. Specifically, the RGB bands are now arranged as Blue–Green–Red, and the multispectral bands as Blue–Green–Red–Red edge–Near infrared. No other elements of the figure were modified.
Comment 4:
In Table 3, some variables are defined in footnotes while others are defined in the main text. Please unify the definition method for all variables.
Response 4:
We agree that the variable definitions were previously inconsistent. In the revised manuscript, all variables appearing in Table 3 are now uniformly defined in the table footnote, and the corresponding explanatory paragraph in the main text has been removed. This ensures a clear and consistent definition strategy.
Comment 5: 
The structure of Chapters 4 and 5 is unclear. The paper discusses traditional methods for spectral images and deep learning methods for RGB images separately, but they should not be discussed in isolation. Both traditional and deep learning methods exist for spectral images, RGB images, and even their combination. The discussion structure should be reorganized.
Response 5: 
We appreciate this important structural suggestion. Following the reviewer’s comment, we revised Chapters 4 and 5 to avoid discussing learning paradigms in isolation.
Specifically:
In Chapter 4, we explicitly acknowledge that traditional machine learning methods based on hand-crafted features have been explored for RGB image-based disease recognition, while clarifying why deep learning methods have become dominant in recent studies.
In Chapter 5, we strengthened the discussion by explicitly introducing a method-level perspective, clarifying the applicability of traditional machine learning and deep learning methods under different image modalities and disease stages.
These revisions ensure a coherent integration of image modality and learning paradigm, addressing the reviewer’s concern without altering the overall structure of the survey.