An Improved YOLOv5-Based Tapping Trajectory Detection Method for Natural Rubber Trees
Round 1
Reviewer 1 Report
The manuscript is written with clear understanding of the project addressed. However, there are major concerns that need to be addressed to enhance the quality of the manuscript. My specific comments are as follows:
Introduction:
“More than 98% of natural rubber comes from Hevea brasiliensis.” Add citation
Based on your objectives, please compare how your study is different from those that have already been published
Materials and Methods:
How many samples used?
“A total of 2000 images is collected for training and testing.” How do you get this value? Split ratio?
“Then, the images are randomly selected according to 8:1:1 for network training…” in what split ratio?
Results and discussion:
The findings lack in terms of justification
Instead of mentioning the results, the authors should justify/explain the findings
Conclusions:
Simplify the main finding of this study
Add recommendation for future studies.
General comments:
Please check the reference styles and grammar of the manuscript.
Author Response
Dear Reviewer,
we quite appreciate your favorite consideration and the insightful comments. Now we have revised our manuscript entitled "An Improved YOLOv5-based Tapping Trajectory Detection Method for Natural Rubber Trees"(ID: agriculture-1867431) exactly according to your comments, and found these comments are very helpful. we hope this revision can make our paper more acceptable. The main corrections in the paper and the responds to your comments are as follows:
Point 1: “More than 98% of natural rubber comes from Hevea brasiliensis.” Add citation.
Response 1: We are very sorry for our negligence. Reference has been added to the revised manuscript. The revised sentence is “More than 98% of natural rubber comes from Hevea brasiliensis [7].”
Point 2: Based on your objectives, please compare how your study is different from those that have already been published.
Response 2: In the revised manuscript, comparisons between our study and published studies have been added in paragraphs 2 and 4 of the Introduction. In paragraph 2, the comparison between our objectives and traditional object recognition algorithms is added. In paragraph 4, the comparison between our objectives and the published two-stage object detection algorithm is added.
Point 3: How many samples used?
Response 3: The number of original samples collected is 2000, and a total of 5000 samples are obtained after image enhancement.
Point 4: “A total of 2000 images is collected for training and testing.” How do you get this value? Split ratio?
Response 4: This has been modified in the revised manuscript. The 2000 original samples have not been processed by image enhancement to increase the dataset size, so the dataset is not classified.
Point 5: “Then, the images are randomly selected according to 8:1:1 for network training…” in what split ratio?
Response 5: This has been modified in the revised manuscript. The revised sentence is “Then, the images are randomly divided into training set, test set and validation set according to 8:1:1 for network training and parameter verification to avoid overfitting of the training model.”
Point 6: The findings lack in terms of justification. Instead of mentioning the results, the authors should justify/explain the findings
Response 6: It has been modified in paragraphs 3 and 4 of subsection 4.1 in the revised manuscript. Firstly, after the CBS module of the neck network is replaced with the CGB module, the reason for the parameter changes is that the amount of model parameters is reduced, which leads to a decrease in the extraction performance of the network model for feature information. Secondly, after adding the CA mechanism to the Backbone network of the original YOLOv5 model, the reason for the parameter changes is that after the attention mechanism is added, the number of model layers increases, which makes the calculation of the model and the FLOPs value increase. Finally, using the EIoU loss function to replace the CIoU loss function of the original YOLOv5 model, the reason for the parameter changes is that the CIoU loss function has shortcomings. However, EIoU loss function comprehensively considers the overlapping area, the distance between the center points, and the real difference between width and height, which makes the model converge faster and the regression process more stable.
Point 7: Simplify the main finding of this study. Add recommendation for future studies.
Response 7: In the revised manuscript, the conclusion section has been revised, and the description of the model improvement section has been reduced.
Point 8: Please check the reference styles and grammar of the manuscript.
Response 8: Grammar and reference formatting have been corrected in the revised manuscript.
Special thanks to you for your good comments.
We tried our best to improve the manuscript and made some changes in the manuscript. These changes will not influence the content and framework of the paper. And here we did not list the changes but using the “Track Changes” function of MS Word.
We appreciate for you and Editors warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.
Kind regards,
Zejin
Author Response File: 
Author Response.docx
Reviewer 2 Report
This paper discuss about the use of image processing technique using YOLOv5 and YOLOv5-improved version in predicting the tapping trajectory in the rubber tapping mechanism. The method of prediction the trajectory will be useful in developing the automatic mechanization and robotic tool in the rubber plantation industry as such backbone task is very tedious work, thus automation would be one of the solution.The specific comments as below;
Comments:
Abstract:
Missing the full name for this abbreviation; CGB, CSP, YOLOv5-CCE, FLOPs, which should be mentions only once, at the first time mention.
Introduction:
Need to define the term ‘tapping trajectory’ in the introduction section. How this trajectory is important in latex tapping operation? More latex or the depth of cut of the plant xylem?
Typo “Labelimg Image Annotation software is used to annotate the data” replace with Labelling …..
Need to improve the writing on this sentences.
“ . On the one hand, adopting sliding windows for region selection leads to high time complexity and window redundancy [17]. On the other hand, the variety of appearance, the uncertainty of illumination changes, and the diversity of background lead to low robustness and poor generalization of traditional object detection methods. And the complicated algorithm steps lead to slow detection effi-ciency and low accuracy [18].”
Overall, its not clear where this wil lead into interm of ground application, either for robotic development or automation in the farming activities.
Method:
Under 2.1 can be improve by establish the tapping section so called “ tapping” such as start, end, area to be tapped, and existing tap line. Graphical information would be useful. Which way is should be the starting point? Typically, the top would be the start point during the manual operation, not the bottom section of the tapping area.
Image as the input. What is the minimum images per plant or tree required for processing? At certain angle, you may see the trunk is very narrow, thus will affect the trajectory accuracy and area for tapping such as Fig. 1 c vs Fig. 1 d. How the algorithm would address this issues?
Section 2.0. the value for TP, FP and FN is unit less, should have mention in in the document.
Section 3.1 Need to establish the table for all the abbreviations used in the flowchart or process in Fig. 3.
Section 3.2.1 Fw and Fh is not define, may be can define h and w separately along with other alphabets
Results
Fig 7, Fig 9 resolutions need to be improved.
Conclusion:
Typo “ paper. the precision rate of the improved YOLOv5 network model reaches 96.0% and the mAP (0.5) value reaches 95.1%. “
References:
Must include the latest research in YOLOv5 or rubber tapping method for 2022.
Author Response
Dear Reviewer,
we quite appreciate your favorite consideration and the insightful comments. Now we have revised our manuscript entitled "An Improved YOLOv5-based Tapping Trajectory Detection Method for Natural Rubber Trees"(ID: agriculture-1867431) exactly according to your comments, and found these comments are very helpful. we hope this revision can make our paper more acceptable. The main corrections in the paper and the responds to your comments are as follows:
Point 1: Missing the full name for this abbreviation; CGB, CSP, YOLOv5-CCE, FLOPs, which should be mentions only once, at the first time mention.
Response 1: We are very sorry for our negligence. The full names of these abbreviations have been added to the Abstract of the revised manuscript.
Point 2: Need to define the term ‘tapping trajectory in the introduction section. How this trajectory is important in latex tapping operation? More latex or the depth of cut of the plant xylem?
Response 2: The definition of the tapping trajectory has been added to the Introduction of the revised version. Accurately detecting the rubber tapping trajectory is the primary condition for realizing intelligent rubber tapping. The rubber tapping trajectory is a spiral line formed by the rubber tapping knife, which is composed of the tapping area, start point and end point. Whether the rubber tapping trajectory is appropriate will affect the raw rubber output and rubber tapping life.
Point 3: Typo “Labelimg Image Annotation software is used to annotate the data” replace with Labelling ….
Response 3: We are very sorry for our incorrect writing. The revised sentence is “In addition, according to the shape of the rubber tapping trajectory, Labeling Image Annotation software is used to annotate the data set and generate the corresponding label files to ensure the accuracy of the parameters.”
Point 4: Need to improve the writing on this sentences. “ . On the one hand, adopting sliding windows for region selection leads to high time complexity and window redundancy [17]. On the other hand, the variety of appearance, the uncertainty of illumination changes, and the diversity of background lead to low robustness and poor generalization of traditional object detection methods. And the complicated algorithm steps lead to slow detection efficiency and low accuracy [18].”
Response 4: The revised sentence is “Firstly, region selection with sliding windows results in high time complexity and window redundancy [21]. Secondly, the variety of appearance, the uncertainty of illumination changes, and the diversity of background lead to low robustness and poor generalization of traditional object detection methods. Finally, the complicated algorithm steps lead to slow detection efficiency and low accuracy [22].”.
Point 5: Overall, it’s not clear where this will lead into in term of ground application, either for robotic development or automation in the farming activities.
Response 5: The development of intelligent tapping equipment can replace manual rubber tapping to solve the current predicament of the natural rubber industry. Accurately detecting the glue dispensing trajectory is the primary condition for realizing intelligent rubber tapping. At present, YOLOv5 model has been widely used in agriculture, such as fruit, crop and plant disease detection. The YOLOv5 model is improved to achieve a lightweight design while ensuring the accuracy, so that it can be applied to rubber tapping robots.
Point 6: Under 2.1 can be improve by establish the tapping section so called “ tapping” such as start, end, area to be tapped, and existing tap line. Graphical information would be useful. Which way is should be the starting point? Typically, the top would be the start point during the manual operation, not the bottom section of the tapping area.
Response 6: Manual rubber tapping knives are mainly divided into push-type rubber tapping knives and pull-type rubber tapping knives. Push-type rubber tapping knife taps rubber from bottom to top, pull-type rubber tapping knife taps rubber from top to bottom. At present, rubber tapping workers mainly use push-type tapping knives for rubber tapping operations. This article refers to the rubber tapping operation of the push-type rubber tapping knife, so the starting point is at the top and the end point is at the top.
Point 7: Image as the input. What are the minimum images per plant or tree required for processing? At certain angle, you may see the trunk is very narrow, thus will affect the trajectory accuracy and area for tapping such as Fig. 1 c vs Fig. 1 d. How the algorithm would address this issue?
Response 7: Processing each tree requires at least 2 images. Our team has developed the mechanical part of the tapping robot, whose rubber tapping module consists of a spiral rail and a tapping knife holder. The tapping knife holder has a profiling structure that can profile tree trunks of different sizes. When the rubber tapping operation is performed, the rubber tapping knife holder drives the rubber tapping knife to move upward along the rail. It only needs to detect the starting point and tapping area to cut a complete rubber tapping trajectory through the rubber tapping module
Point 8: Section 2.0. the value for TP, FP and FN is unit less, should have mention in in the document.
Response 8: We are very sorry for our negligence. The units for TP, FP and FN have been added in the revised manuscript.
Point 9: Section 3.1 Need to establish the table for all the abbreviations used in the flowchart or process in Fig. 3.
Response 9: In the revised manuscript, a table for full names of all abbreviations has been created below Figure 3.
Point 10: Section 3.2.1 Fw and Fh is not define, may be can define h and w separately along with other alphabets
Response 10: h, w, Fh, and Fw have been defined in the revised manuscript. The revised sentence is “where h represents the height direction, w represents the width direction, F1 is the batch normalization function, δ is the nonlinear activation function, σ is the sigmoid function, fh and fw are feature maps in the height and width directions, Fh and Fw repre-sent convolution operations in the height and width directions respectively, gh and gw are the attention weights in the height and width directions, xc(i, j)is the input of the original feature map, and yc(i, j)is the output of the feature map with attention weights in width and height directions.”.
Point 11: Fig 7, Fig 9 resolutions need to be improved.
Response 11: The original images of Figures 7 and 9 have been uploaded in the revised manuscript.
Point 12: Typo “paper. the precision rate of the improved YOLOv5 network model reaches 96.0% and the mAP (0.5) value reaches 95.1%. “
Response 12: We are very sorry for our incorrect writing. The revised sentence is “The precision rate of the improved YOLOv5 network model reaches 96.0% and the mAP (0.5) value reaches 95.1%.”.
Point 13: Must include the latest research in YOLOv5 or rubber tapping method for 2022.
Response 13: References related to YOLOv5 and rubber tapping in 2022 have been added to the revised manuscript.
Special thanks to you for your good comments.
We tried our best to improve the manuscript and made some changes in the manuscript. These changes will not influence the content and framework of the paper. And here we did not list the changes but using the “Track Changes” function of MS Word.
We appreciate for you and Editors warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.
Kind regards,
Zejin
Author Response File: Author Response.docx
