Research on 3D Phenotypic Reconstruction and Micro-Defect Detection of Green Plum Based on Multi-View Images
Round 1
Reviewer 1 Report
Thank you for the opportunity to review this article.
The authors evaluated the Research on 3D Point Cloud Construction and Micro Defect Detection of Plum Based on Similar Graph Clustering Matching.
The manuscript is generally well written in good order.
Some minor remarks follow.
Please write all the references with the same way (both at the end and in the text) and according journal's guidelines.
Line 135.[16]...Please without color.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
The organization of the paper should be improved. Currently, the introduction is mixed with a related work section. I would recommend splitting it into two sections: Introduction and Related Work. In Introduction clearly describe the motivation, why this research exists, what are deficiencies of existing methods and what are contributions of this work. And in Related Work section, discuss the relevant related work.
The language side of the paper must be improved. Many sentences are impossible or difficult to understand or sound unnatural. Some examples:
"Aiming at the dense point cloud model of green plums, through point cloud pre-processing, the improved adaptive segmentation algorithm based on Lab space realizes the effective segmentation of the point cloud of green plums micro-defects"
"After clustering the micro-defect point cloud, the micro-defect information of green plums was extracted based on Random Sample Consensus (RANSAC) plane fitting, which provided a theoretical model for further improving the accuracy of sorting the appearance quality of green plum"
"SFM algorithm has stable and superior performance in three-dimensional reconstruction [7-9]." - superior to what?
"The coarse matching of features in traditional SFM reconstruction will cause a large number of repeated matching." - not sure what 'repeated matching' means.
"The set of SIFT features ... extracted from the image were ascending dimensions transformed and normalized." - I'm not sure what 'dimension' authors refer to? What value is used to sort SIFT features?
"Inter-class fusion of the image classes t, R is performed via minimum spanning tree to obtain global cyanotype sparse reconstruction results. " - It's impossible to understand this sequence. I thought t and R are translation vector and rotation matrix, not “image classes”? What is 'cyanotype'?
How was the number of "Error points" in Table 1 calculated? It's unclear why authors claim that many matches will result in larger reconstruction error. I would argue for the opposite; the more correct SIFT matches we have, the lower the reconstruction error we'll get.
It's difficult to understand what authors' contributions are. Arguments for the advantages of the proposed improvement (building an image similarity graph using a Fisher vector) are not convincing. The authors compared results obtained using OpenMVG package with their improvement to results obtained using different software packages: Bundler and VisualSFM. But the differences in Pose and Range Error metrics could be due to implementation differences or different settings of run-time parameters in these packages. The authors should compare their method with the baseline version of OpenMVG.
Discussion in lines 262-270 is unnecessary. The authors discuss and illustrate the fact that PMVS reconstruction generates more points than sparse SFM. But SFM methods by design generate sparse point clouds, and dense stereovision methods, such as PMVS generate dense point clouds. It's a well-known fact, not deserving a dedicated discussion or illustration in the paper.
The author hasn’t mentioned what is could be the practical applicability of the proposed approach. The reconstruction and analysis of a single plum take over 2 minutes, not counting the time required for image acquisition.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Title of manuscript: Research on 3D Point Cloud Construction and Micro Defect Detection of Plum Based on Similar Graph Clustering Matching
Review comments:
1. Title should be redefined to make it more precise for the work presented.
2. Abstract needs to be rewritten by removing the grammatical error.
3. Introduction part missing the introduction about subject of research and their applications.
4. Some literature about the latest development in the field of research must be provided which is missing in the manuscript. Add the separate section for Literature Review.
5. Material and methods section describes about the hardware but does not specify the method for carrying out the process for finding the output.
6. High quality images should be used like in Figure 1 a, the image is not that much clear. Figure 3 is also not clear.
7. Figure 4 should be nomenclature as “Block schematic of improved plum 3D reconstruction algorithm”.
8. Statement on line number 220 needs to be reconstructed to have a proper meaning.
9. Figure 6 should be nomenclature as “Block schematic of the detection of micro defects in green plums”.
10. In Table 2, Instead of using “ours’, use “Proposed Approach”
11. The comparison with the state-of-the-arts is missing.
12. Conclusion of the paper needs to be rewritten for making it more precise.
Overall suggestions:
Manuscript needs to be formatted properly as far as the placement of the text, equations and the figures are concerned.
English grammar throughout the manuscript need to be checked.
Motivation behind the research needs to be mentioned.
Check whether all the references are cited properly
Basically the manuscript of this type expected to have around 40 to 50 reference.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
My concerns from the review of the previous version of the paper were addressed. Therefore I recommend to accept the paper.
Reviewer 3 Report
The revised may be accepted for publication.
