Identification of Cabbage Seedling Defects in a Fast Automatic Transplanter Based on the maxIOU Algorithm

Round 1

Reviewer 1 Report

This paper shows an automatic method for the identification and classification of cabbage seedling using computer vision algorithms with real-time constraints.

In my opinion, the authors made a great effort in their work. However, there are some small mistakes and some doubts which are detailed in the following list:

 

Abstract:

For me, the abstract is extremely long. It is well detailed but unnecessarily long. It can be shortened.

 

Keywords:

I suggest deleting the word maxIOU in keywords since this is a specific metric and it is not at the same level of abstraction than machine vision or image segmentation. Also, the style of this word in the text is different.

 

Introduction:

Line 86: There are too many "area" words in the same sentence.

Section 2.2. Image Acquisition:

Line 101: There is a point between two brackets that seems wrong.

 

Section 2.3. Color space analysis:

Line 116: I suggest changing "seedling-the" for "seedling: the..." Line 130: I suggest changing the sentence "In this article, a grayscale histogram to determine the color space." to "In this article, a grayscale histogram is used to determine the color space." Line 134: The reference to "figure 2A1" needs a space between "2" and "A1". Throughout the document, I can see many references to the figures that happen the same, or lack a space.

 

Section 2.5. Identification of seedling defects

Lines 255 to 258: I suggest putting this as a list of items to improve the reading of the text.

 

Section 2.6. Identification of cabbage seedling defects prodedure

Line 304: I suggest changing "prodedure" by "procedure". Line 307: The style of "Step1" word in the text is different from the rest. Also, space is needed between "step" and "1". Line 315: There is a number "9" that I do not understand.

 

Section 3.1. Image segmentation threshold calculation and result analysis

Lines 338 and 339: I suggest to change this part "[0,24]" and "[21,255]" with brackets since the reference style of the journal is with "[reference number]" and it can be confusing for a reader.

 

Section 3.3. The result of seedling defication identification

Line 397: The style of the text is different from the other sections. Table 4: The word "precious" must be changed for "precision".

 

Other considerations:

It is not necessary, but I suggest making public the dataset that the authors used to allow other authors to compare in the future works given the difficulty of finding this type of dataset. Also, I suggest giving more importance in the article to the contribution of the dataset, since the dataset is a contribution in itself.

 

In my opinion, the graphics would be much cleaner if the green background is removed. Also, in the case of certain figures, the "threshold" text appears outside this green background.

 

I recommend that Table 4 compare the proposal presented in this paper with those previously discarded of the use of HSV, etc. This would help reinforce the necessity of the present work.

 

After reading the introduction, the necesity of using maxIOU is not completly clear. I can imagine that it is because you need a "reliable" method to use it in real-time, but I cannot completely appreciate in the text. For this reason, and to reinforce the need for the use of IOU and maxIOU metrics/methods, I recommend including these current references of works that make use of them:

@misc{Angus_2019,
title={Efficacy of Pixel-Level OOD Detection for Semantic Segmentation},
author={Matt Angus and Krzysztof Czarnecki and Rick Salay},
year={2019},
eprint={1911.02897},
archivePrefix={arXiv},
primaryClass={cs.CV}
}

@article{Rodriguez_Lozano_2019,
doi = {10.3390/s19194096},
url = {https://doi.org/10.3390%2Fs19194096},
year = 2019,
month = {sep},
publisher = {{MDPI} {AG}},
volume = {19},
number = {19},
pages = {4096},
author = {Francisco J. Rodriguez-Lozano and Fernando Le{\'{o}}n-Garc{\'{\i}}a and M. Ruiz de Adana and Jose M. Palomares and J. Olivares},
title = {Non-Invasive Forehead Segmentation in Thermographic Imaging},
journal = {Sensors}
}

@article{Caicedo_2019,
author = {Caicedo, Juan C. and Roth, Jonathan and Goodman, Allen and Becker, Tim and Karhohs, Kyle W. and Broisin, Matthieu and Molnar, Csaba and McQuin, Claire and Singh, Shantanu and Theis, Fabian J. and Carpenter, Anne E.},
title = {Evaluation of Deep Learning Strategies for Nucleus Segmentation in Fluorescence Images},
journal = {Cytometry Part A},
volume = {95},
number = {9},
pages = {952-965},
keywords = {fluorescence imaging, image analysis, deep learning, nuclear segmentation, chemical screen},
doi = {10.1002/cyto.a.23863},
url = {https://onlinelibrary.wiley.com/doi/abs/10.1002/cyto.a.23863},
year = {2019}
}

Author Response

Thank you for give me precious advises for my article.My repliy is in the attachment text.

Author Response File: Author Response.docx

Reviewer 2 Report

The article focuses on image processing and related algorithm for segmentation applied on a specific feature of planting mechanism.

Accuracy and precision are theoretically defined but do not reflect an assessment of the tested transplanting device under field operation. The question arises whether the algorithm performances with which quality under changing conditions.

Excessive abbreviations without definitions are used MRA, YCrCb color space, otsu later Otsu , line 83- 84, IPC. Concise use of technical terms is poor.

l 56, 58 etc reference format is not standard, later reference by numbers only

Fig 1 Fuzzy photos, details to small. Reader is not able to recognize what is written line 103 ff

Transplanting operation is not explained, please describe the mechanized operations of the transplanter to understand where and when the detection take place

The paper is acceptable if it is clearly stated that the results are from laboratory conditions with artificial light and are not proofed under field conditions. This should be mentioned in the abstract and in conclusion otherwise the title and results suggests that the system is operable on field level. The variance of plants, species, soil -wet and dry, sand and clay may lead to other results and are not defined in the paper as test conditions.

It is doubtful to use data for training the segmentation algorithm from the same data set the image analysis is applied on, l 342 - 345

L 444 detection of missing seedling in a continuously working planter is a necessary step but not complete. To optimize a transplanter the correction of empty cups is a request to improve accuracy and precision of planting.

 

Author Response

Thank you for give me precious advises for my article. My reply is in the attachment text.

Author Response File: Author Response.docx