Growth Analysis of Plant Factory-Grown Lettuce by Deep Neural Networks Based on Automated Feature Extraction
Round 1
Reviewer 1 Report
The manuscript illustrated the application of deep neural networks for growth analysis of lettuce. The title is quite interesting, however, the context must be improved. Most of context is unclear and very difficult for readers to follow. All comments and suggestions must be clarified in the revised version before getting publication.
1. Authors mentioned several times about growth analysis, however, it is unclear what is growth analysis ??? is it planting date ? growth stage ? type of lettuce ?... Please clarify the aim of the study clearly in the introduction.
2. What is the color indices did authors used in the data analysis. e.g. RGB, HSI, other indices ? please clarify and mention them clearly in the manuscript.
3. The data analysis might be influence by region of interests (ROI) on the image. From the context, it has been assumed that only 1 single area of image was used as a representative area of the object image. Is this adequate to represent an overall object image ? Please clarify and support the evidences.
4. There is no validation results, please support the information as it is important to express the feasibility of your model.
5. What is the label for x- and y- axis of figure 4 ? It is assumed that information from different part of lettuce used to predict the same indices on the growth of lettuce. Why different part of plants give very different numeric value range ? If the model is sufficient, all part should give the similar results ?
Author Response
- Authors mentioned several times about growth analysis, however, it is unclear what is growth analysis ??? is it planting date? growth stage? type of lettuce?... Please clarify the aim of the study clearly in the introduction.
- Added relevant information (L64-67).
- What is the color indices did authors used in the data analysis. e.g. RGB, HSI, other indices? please clarify and mention them clearly in the manuscript.
- Added relevant information (L108).
- The data analysis might be influence by region of interests (ROI) on the image. From the context, it has been assumed that only 1 single area of image was used as a representative area of the object image. Is this adequate to represent an overall object image? Please clarify and support the evidences.
- Since the target task of this study was not a kind of object detection, the ROI was not applicable. Instead, relevant information was added in Results and Discussion section (L205-208).
- There is no validation results, please support the information as it is important to express the feasibility of your model.
-Figure 4 represents the validation result. Added relevant information (L219-220).
- What is the label for x- and y- axis of figure 4? It is assumed that information from different part of lettuce used to predict the same indices on the growth of lettuce. Why different part of plants give very different numeric value range? If the model is sufficient, all part should give the similar results?
- Added relevant information (L218).
Reviewer 2 Report
Please find document with my comments attached.
Comments for author File: 
Comments.pdf
Author Response
- As feedstock for the experiments, lettuce grown were used. One of my initial concerns was whether it would be possible to accurately identify and pinpoint an effect of suggested methodology in comparison with well known deep learning procedures. The authors do not address this point in a sufficient manner in the Introduction, although they discuss about this in the Results and Discussion chapter. This will make the motivation for this experiment easier to understand in my opinion and introduce the reader better into the topic.
- Added relevant information (L67-69).
- Something else missing from the Introduction is a correlation with papers based on deep learning applied to lettuce grown in plant factory along with a description of cover all key aspects of the process, which are: (1) Product quality; (2) Individual process time and total throughput; (3) Process/equipment design, operation and control, (4) Energy use and consumption and consequently (5) Process economics.
- We are afraid that the evaluation of the PFAL itself could deviate from the logical context of our insist. We originally focused on the usage of the deep learning algorithms. In this perspective, PFAL was just a tool to grow material plants.
- Other computational information was explained in Materials and Methods section.
- M&M. Why do authors used a convolutional neural networks (ConvNets) FFNN and LSTM as a model structures? What was the base for such choice. Please add your explanation into the introduction section.
-Added relevant information (L123, L129-130).
- LinReg model should be also introduced in 2.3. Model structure section.
-Added relevant information (L130).
- Discussion section is very poor. Please add your comments of further model improvements and ideas for validation improvement.
- Added relevant information (L197-204).
Round 2
Reviewer 1 Report
It was dramatically improved.
The manuscript was accepted in current form.
Reviewer 2 Report
The authors have carefully revised their manuscript according to my
comments.
