Classification of Tree Species in Transmission Line Corridors Based on YOLO v7

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript (forests-2700357) addresses the task of categorising individual tree species in transmission line corridors and highlights how effective tree management can mitigate damage to these lines. The presence of a dense forest canopy hinders the accuracy and speed of current deep learning models in identifying single tree species. This study employs aerial multispectral imagery from UAVs to generate a dataset of individual tree crowns, divided into training and verification subsets. It utilises the YOLO v7 and v4 networks for iterative training, and discovers that the YOLO v7 network can recognise single tree species with an accuracy of up to 85.42%. This method reliably aids in monitoring potential risks and conducting thorough surveys of trees in areas surrounding the transmission lines.

The Introduction and Results sections were satisfactory. However, the Materials and Methods section requires improved descriptions and additional information regarding the training, validation, and testing of the YOLOv7. The discussion section is inadequate. The conclusions are acceptable, but the references need to be checked throughout the manuscript and in the final section.

Keywords should be in alphabetical order.

Line 63 requires reference verification.

All figure captions need to be rewritten to fully convey the subject matter. As mentioned, it is challenging to discern all elements and their references.

Why were band combinations used, as noted in Section 2.2.2?

However, this discussion has been inadequate. The authors should compare the advances made with the YOLOv7 model with previous models. Moreover, why were YOLOv8 or other models not applied? What progress has been made in using RGB, RGB-NIR, RGB-SRIW, or hyperspectral imagery? What are the benefits of using fewer bands? Do the authors plan to train and validate the models using vegetation indices? How can these models be adapted for training and identifying other species?

The authors should conduct these tests in other areas or with other images to truly compare the adequacy of the training.

What is the actual sample size for the training?

Comments on the Quality of English Language

English requires major revisions in grammar, spelling, and verbosity.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper is an interesting study about the deployment of YOLO in the classification of tree species using multispectral imagery. It shows reasonable results for the classification of the species and in my opinion, it might be accepted for publication in Forests journal.  There are some drawbacks though, which have to be solved before publication. 

1 - first of all the the justification for trees in power line corridors is not epxlained. The importance of power line corridors is clear, but what is special in trees under power line corridors is not explained. Are the tree species under line corridors different than elsewhere? Do they behave differently? The importance of power line corridor management is understandable, but not the specificity of the species under it. The species classification is the same as in other areas. 

 

2- this phrase: "Forest areas have a significant negative impact on the power corridor transmission system, which must be avoided" is unacceptable. In no way this was ever proven. Oppositely many power companies grow small trees under the power lines to increase carbon fixation in these territories. Please distinguish between forest and trees. Second, simple solutions like removing all trees representing potential threat is oversimplification which is not part of this study and such claims are not justified and has nothing to do with deployment of YOLO AI to tree classification. So please try to as clearly as possible explain what is the importance of classification of tree species under power lines. (maybe faster growing species are more important than slow growing etc. )

3 - The discussion extent is unacceptable. And there is actually no discussion, it is just a repetition of results and general observations about own work. The discussion chapter is expected to be at least one or more pages, but mostly focusing and comparing with other research and papers. 

 

Conclusion should instead of a simple accounting of the performance of the paper focus more on general conclusion of the work, the weak and strong parts and propose some future work and framework. 

 

Small errors in text and English should be checked before publication. There is many of them at this moment, please ask English reviewer to check it. 

 

 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

We have improved the manuscript accordingly. However, the captions of the figures and tables still need attention and must be rewritten to highlight all relevant elements as well as to explain what they mean.

Comments on the Quality of English Language

The English language requires a moderate revision.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The paper was changed and all my comments were incorporated in the new version. I agree with publication of the article .. just small technical comments:

Row 129 with an altitude of 0 60m … please clarify which height exactly

 

Row 134 is about 5.73 hectares… if it is „about“ I guess you can say about 5 hectares

 

 

Row 283 memory is 16G… probably 16GB?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf