Potato Visual Navigation Line Detection Based on Deep Learning and Feature Midpoint Adaptation

Round 1

Reviewer 1 Report

The authors proposed a novel approach to detect potato visual navigation lines with deep learning and feature midpoint adaptation which is a good topic. The authors leveraged U-net to generate a semantic segmentation and then combined the original pics to extract mid-points. The authors leveraged K-means to get the navigation line fitting. The results show that the proposed approach can improve the crop row segmentation accuracy. The reviewer has the following comments: 

 

1. What is the motivation of the paper? Why is the crop row segmentation is so important? The reviewer suggests that the authors reconsider the motivation and highlight the motivation and contribution in the abstract. 

 

2. The sentence "End-to-end potato crop row detection is proposed in this paper in order to solve the low navigation accuracy and poor robustness caused by traditional image-based autonomous navigation that are susceptible to field weeds, illumination, and plant appearance variances in different growth period." is too confusing. The reviewer suggests that the authors re-write this sentence to make it clear and concise. 

 

3. Why are there so many errors "Error! Reference source not found.]"? The reviewer suggests that the authors double check the original file and solve the problems before the next submission. 

 

4. The literature review for the peers' work is too weak. The reviewer suggests that the authors collect more references and have a comprehensive literature review. 

 

5. It is supposed to have an overview of the whole system. The authors do not provide that. The reviewer suggests that the authors read some more papers and learn how to have an overview of the system. 

 

6. Before you have research, do you have a target? Research is not a flash that can not help you to have any progress. 

 

7. Do the methods (Semantic Segmentation, Model Training and Data Augmentation, K-means clustering, and Least Squares Fitting) have any relationship with your final goal? Do you have any corresponding revisions to make the methods suit your case? 

 

8. Before you present your results, you are supposed to have a description of your platforms and configurations. The reviewer suggests that the authors make up the missing parts. 

 

9. The language of the paper shall go through polishment before the next submission. Otherwise, the reviewer DOES NOT recommend submitting again. 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The authors present a method for localization into a potato field based on visual data only. The method relies on a DNN to extract the semantic information and then on k-means algorithm to locate the mid point in each potato row, whilst a least-square method is used to find the lines. The proposed method is somehow, although more sophisticated techniques has been presented in the recent literature, yielding more efficient results (see e.g. Fei, Zhenghao, and Stavros Vougioukas. "Row‐sensing templates: A generic 3D sensor‐based approach to robot localization with respect to orchard row centerlines." Journal of Field Robotics (2022).), though not always purely visual. 

 

Few problems have been spotted by the reviewer, as follows:

 

i. The state of the art is simply written, only referencing some previous works, with no actual intention to highlight the problem and reveal the necessity of the proposed method.

 

ii. The dataset collected in minimal, resulting to the evaluation results being unreliable. The authors should somehow justify their results with such limited data.

 

iii. The proposed method is rather primitive. K-means, followed by least square is a good, yet simplistic,  approach  to tackle the problem. A method relying on SVM-based local path planning algorithm (see e.g. Charalampous, Konstantinos, Ioannis Kostavelis, and Antonios Gasteratos. "Thorough robot navigation based on SVM local planning." Robotics and Autonomous Systems 70 (2015): 166-180.) would have allowed the authors more agility and flexibility to the proposed algorithm and allow for even better accuracy. 

iv. English need some polishing. Also some references are missing (see p. 2 "Error! Reference source not found ")

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The revised version is a mess. Please resubmit the revision.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Authors have now prepared an extensive and thorough revision of their manuscript. For some comments they were not able to address, they have provided sufficient reasoning. 

Author Response

Dear Reviewer:
Thank you for your review of my paper， I wish you all the best！

Round 3

Reviewer 1 Report

This paper presents an interesting work. Overall the work is solid and complete. However, this paper didn't present the state of the art of relevant topics. Some important references are missing. A more comprehensive literature review of relevant topics is desired. Some example references are "Dynamic inversion of inland aquaculture water quality based on uavs-wsn spectral analysis", "Bio-inspired routing for heterogeneous Unmanned Aircraft Systems (UAS) swarm networking", "A decade survey of transfer learning (2010–2020)".

Author Response

Please see the attachment

Author Response File: Author Response.docx