Transmission Line Segmentation Solutions for UAV Aerial Photography Based on Improved UNet
Round 1
Reviewer 1 Report
The manuscript "Research on transmission line image segmentation method under UAV aerial photography based on improved UNet" presents a deep learning model based on UNet for detecting transmission lines in UAV-acquired images.
The manuscript is well written except it needs to be properly formatted as there are at many places "[Error! Reference source not found."
Everything is well explained and the figures are also well explained.
1. Summary
In this paper the author extracts the power lines and towers by modifying the deep learning segmentation model UNet algorithm. The GhostNet module and a feature recombination operator CARAFE is integrated with the UNet algorithm to reduce the computational complexities and optimized feature recombination during the decoding stage. Asymmetric convolution is also incorporated for capturing long-distance targets in transmission lines to enhance the extraction capability of the model for target features. The author suggests that there is a substantial decrease in the number of model parameters and a fair improvement in inference speed delay along with improvement in segmentation metrics.
2. Issues
1. Please check Line 146 -149 “traditional upsampling leads to the loss of feature map information, and the deconvolution leads to the increase of computation”. Down sampling leads to loss of information rather than up sampling.
2. Line 150-152 needs rephrasing as it is unclear as to what the author is trying to suggest here.
3. In section 3.3 the CARAFE algorithm suggests feature compression in both modules. It is unclear how feature compression helps in feature reconstruction based on contextual neighbourhood pixel information.
4. It is suggested that model hyper parameters be listed in a tabular format with both initial and final values for ease of understanding.
5. It would be interesting to know how the model performs in highly dense urban areas with building in background.
6. Line 283 “model requires higher detail feature extraction in the recognition of power lines. and a” is there a full stop or it is by mistake?
7. Abstract can be rewritten and arranged in paragraph form.
8. Overall the paper needs an overhaul in usage of English especially punctuation. There are many silly mistakes across the manuscript.
.
Author Response
See pdf for details
Author Response File: Author Response.pdf
Reviewer 2 Report
The manuscript is well within the scope of the journal. The manuscript proposes a new method for extracting power lines and towers by combining the deep learning segmentation model UNet algorithm and light-weighting its features to extract backbone structure and reconstruct them with contextual information features.
· In most of places citation references are missing
· In many places, sentences start with “And” and “By”; try to avoid it.
· Lines 125-127, sentences are meaningless; rewrite the sentence.
· Similarly, lines 130-132 and lines 134-135.
· Details of dataset used should be included before methodology. Also, add citation from where dataset was taken and what were specifications.
· Figure 8, flow chart need to be improved. At many places text is going outside of the boxes.
· In figure 8, at many places long text is written, which should be avoided.
· Figure 9 and 10 have same caption, which should be avoided.
· English is very poor in the manuscript. At many places sentences are incomplete and meaningless.
Author Response
See pdf for details
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear Authors,
Please, recieve my review comments contained by the attachement.
Regards,
Reviewer
Comments for author File: Comments.pdf
Author Response
See pdf for details
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
Dear Authors,
Thank yu for correcting your manuscript also based o nmy comments. Please, recieve my recent comments and suggestion on the latest version.
Regards,
Reviewer
Comments for author File: Comments.pdf
Author Response
See PDF for details
Author Response File: Author Response.pdf