3D-GPR-RM: A Method for Underground Pipeline Recognition Using 3-Dimensional GPR Images
Round 1
Reviewer 1 Report
This manuscript proposed novel deep learning-based method for underground pipeline recognition using GPR images, where convolutional neural networks (CNNs) were employed for the task of interest. In the proposed method, to expand the number of samples used in dataset, data augmentation technique was introduced based on three-dimensional matrix rotation, together with wavelet-based denoising method to filter out the direct wave interference. The performance of the proposed method has been fully validated using experimental verification, with satisfactory results. Overall, the topic of this research is interesting, and the manuscript was well organised and written. The detailed comments are summarised as follows.
1. The main innovation and contribution of this research should be clearly clarified in abstract and introduction.
2. Please broaden and update literature review on CNN or deep learning to demonstrate its excellent capacity in resolving real problems. E.g., Automated damage diagnosis of concrete jack arch beam using optimized deep stacked autoencoders and multi-sensor fusion; Torsional capacity evaluation of RC beams using an improved bird swarm algorithm optimised 2D convolutional neural network.
3. The performance of proposed deep CNN models is heavily dependent on the setting of hyperparameters. How did the authors tune/optimise the network parameters to achieve the best classification accuracy in this research?
4. Figures 11, 14 and 17: the authors demonstrated the training process, in terms of training and validation samples. How about the test set? Test set should be different from training and validation sets, which could not be included in training process. Please give more details on how the authors determined training, validation and test datasets in this research.
5. The training time should be considered as one of evaluation metrics.
6. How about the robustness of the proposed networks against noise effect?
7. More future research should be included in conclusion part.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Good for the publication in the present form
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Figure 2 need to increase the caption in the figure due to illegibility.
Is there any difference in what environment the proposed pipeline is located?
As I understand from the article, most of the signal processing is based on a neural network, if the pipeline has a cross-shaped fork, what will such a picture look like?
At what depth can the pipeline be determined?
Will the picture of the received data change when the weather conditions on the roads (after rain) change, if so, how?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The article is rather speculative, since the effectiveness of the algorithms is tested either on model images or on real scans, which are quite well readable even with the naked eye.
The main difficulty in diagnosing urban communications, especially old and most prone to accidents, is the high noise level of the GPR images obtained. The article requires significant revision with using a sufficient number of images obtained in a real urban environment.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 4 Report
The scientific significance of the article, with low number of experimental samples, is largely limited. Because of this, it is difficult to assess the actual contribution of the authors to solving the important and urgent problem of detecting and diagnosing urban communications by GPR. In connection with this, the final decision on the publication of the article must be made by the editors of the journal.
Author Response
Please see the attachment.
Author Response File: Author Response.docx