A Method for Identifying Picking Points in Safflower Point Clouds Based on an Improved PointNet++ Network

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The authors wrote an article titled: "A Method for Identifying Picking Points in Safflower Point Clouds Based on an Improved PointNet++ Network". The goal is to improve the localization accuracy in challenging conditions (e.g., complex plant morphology), which is essential for the development of robotic picking.
Suggestions for improvement:
- A very powerful computing machine was used (RTX3080, 32GB RAM, i9). Suggest an optimization so that this can also be used on more common machines. I recommend discussing the possibilities of optimizing the computing time or deploying the method in real field conditions.
- Lighting is always the most important factor when taking pictures. Discuss different lighting conditions.
- Try to add an experimental comparison with some well-known RGB-D approaches (e.g., YOLOv5 + depth fusion).
- The mIoU for fruit ball is only 44.14%. Although it is an improvement, it seems small to me. Discuss how to improve this to at least be above 50%.
- Table 4 is not needed. It is unnecessary to make a table of one row. It is enough to list it in text.
- The list of references mainly includes authors from Asian countries. Have you tried searching elsewhere?
After incorporating these comments, the article will be suitable for publication.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this work, the authors propose a three-stage 3D point cloud analysis framework incorporating modified previously developed algorithms. As the authors describe the workflow, the aforementioned stages of analysis are: capturing and processing multi-view RGB images; segmenting the images; and identifying optimal picking coordinates. According to the authors, testing the proposed method results in improved picking point localization accuracy.  Their results could effectively address morphological variations and ultimately provide high-accuracy visual guidance for the potential use of robotic harvesting systems.

From the imaging viewpoint, the paper is interesting because it combines different algorithms to study an agronomical problem. However, the discussion on how such algorithms are actually useful needs to be improved.

1. Due to the journal's scope, provide more details on why this study is relevant to agronomy. Also, provide more details on why the safflower is important (from the agronomy viewpoint) and add related references (try to use references that demonstrate that this could be a globally interesting problem, instead of just a regional study).
2. Section 2.2. Please provide details on the hardware employed for 3D model construction. Did you use the same hardware (lines 240-245) during the entire workflow?.
3. From the text, it is unclear if each flower was captured individually or extracted from an overall photo. In either case, please discuss the limits of the method from the viewpoint of a possible future application.
4. Describe your training set and how you employed it in more detail.
5. Does your model provide a way to measure the “quality” of the flowers (other than just their geometry)? Due to the scope of the journal, this must be addressed in your discussion. Can a related “vegetation index” be adapted to your model to provide an adequate agronomical result?

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors,

I appreciate your efforts. I think that the article is good , and I have the following recommendations for improvement.

I understand that you are presenting an application for “visual perception” using “3D point cloud analysis.” The idea of the article is common; I see it in other articles; therefore, the novelty is reduced. You propose a new scheme for safflower picking point recognition. The term “scheme” is ambiguous. You can find another term.

Also, subchapter “2.6. Experimental Evaluation” should be moved to the chapter “3. Result and Discussion.”

The abstract is ok, and I have only one comment: please present the novelty.

In the chapter Introduction, you present the safflower, some issues of harvesting, some algorithms that can be used (YOLO, Faster R-CNN, CR-YOLOv5, etc.), and some methodologies for picking point localization. You emphasize the importance and the advantages of cloud-based localization.

The text is bulky, and to be easy to read and understand, you can split the text into more paragraphs. You can organize the paragraphs according to the ideas presented. For example, you can start a new paragraph after line 42.

Also, you cannot jump from one idea to another. For example, you can insert a phrase before you are presenting YOLOv3.

In the final part of the chapter “Introduction,” you are presenting the methodology of your work.

Overall, the first chapter is good, the ideas are clear, and you are using enough and suitable references.

In the second chapter, you start with the overall method presentation using appropriate diagrams and information. You continue to present in an interesting way and with enough details each component of your system.

A suggestion is to present the parameters xi, yi, and zi and index i. How do you acquire them? How many values does “index i” have? You can discuss more about the OBB algorithm.

I have similar suggestions regarding equations 9-11.

The chapters “Results and Discussion” and “Conclusions” are satisfactory and include interesting findings, and it is easy to read them.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The new version of the paper has significantly improved, especially regarding the "agronomical" implications and usefulness of the proposed method of analysis. However, I still have a significant concern. While I recognize the interest this analysis could raise, it is far from an immediate, practical, and useful application in the agronomical field. The improved discussion somehow addresses this, but more details should be provided in this regard throughout the text. 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf