Deep Learning-Based Method to Recognize Line Objects and Flow Arrows from Image-Format Piping and Instrumentation Diagrams for Digitization
Round 1
Reviewer 1 Report
The paper deals with the usage of AI for supporting image analysis. In the present form the paper needs improvements for being accepted in an international journal. Starting from the introduction aithors should provide a widerpicture on the usage of deep learning in the image analysis considering that many reseraches has been done in the medical domain where got perforances have been obtained working with very complex usages scenarios. Some more references shoudl be added as for instance the following ones:
Moreno, R., et al.(2020). Hierarchical fracture classification of proximal femur X-ray images using a multistage deep learning approach. European Journal of Radiology, 133, 109373.
Cao, F., et al. (2018). A new method for image super-resolution with multi-channel constraints. Knowledge-Based Systems, 146, 118-128.
Cirrincione, G. et al. (2021). Understanding Abstraction in Deep CNN: An Application on Facial Emotion Recognition. In Progresses in Artificial Intelligence and Neural Systems (pp. 281-290). Springer, Singapore.
Erickson, B. J. (2019). Deep learning and machine learning in imaging: Basic principles. In Artificial intelligence in medical imaging (pp. 39-46). Springer, Cham.
Regarding the methodological section authors should provide a global perspective of the proposed methodology for going further into the specific details of the specific methodology stages, introducing more information regarding parameters and strategie involved
Author Response
Please refer to the revision notes attached.
Author Response File: Author Response.docx
Reviewer 2 Report
The paper addresses the problem of line and flow arrows detection in piping and instrumentation diagrams.
The solution presented is very focused. The novelty lies mainly in the application, and in the resulting improvement from a correct association of the detected lines signs and flow arrows, to the final digital diagram.
Comments:
Are the problems with line detection listed in 3.2 the only problems that occur when performing such processes?
In line 192, when you refer graphics processing you are talking about Hough transform?
From Figs 4. and 6. one can assume the there are classes for the lines in the annotated files. Do the deep neural network also detects lines, or just signs and flow arrows?
Do the ratios of the bounding boxes (Fig. 5) will hold for different P&I diagrams?
Form Fig. 6 you can notice an imbalance on the data set (horizontal and vertical dotted lines samples are much more than samples from any other class). Do you think this may bias your results?
Do you think the time the method takes for detecting lines might be a drawback for real world applications?
Can you compare your results to any of the methods referred in 2?
Please check the text for small typos and mistakes.
Author Response
Please refer to the revision notes attached.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Authors have improved the scientific level of the paper.