High-Speed Object Recognition Based on a Neuromorphic System

Round 1

Reviewer 1 Report

The paper entitled "High-Speed Object Recognition Based on A Neuromorphic System" is very interesting and the content shows some experimental results about object recognition by using a designed system with different kinds of algorithms. However, there are some aspects that need to be addressed:

- It is not clear how many samples were acquired to make the different tests. This number is very important in order to assess the ability and feasibility of those methods to get the object recognition accuracy of 99%.

- Would be important to include a table with the overall information and the comparison of the traditional recognition systems using scaled ANN. 

- A cross-validation technique needs to be implemented with the data set.

-  Please, add more experiments with different rotating speeds in order to verify the results with 99.9% at all speeds using other kinds of letters.

- Describe the computational cost based on these tests.

 

 

Author Response

Thank you very much for your constructive suggestions that have helped us
improve the manuscript. Please refer to the PDF attachment for the point-by-point responses.

Author Response File: Author Response.pdf

Reviewer 2 Report

 

This manuscript benchmarks Object Recognition Based on A Neuromorphic System and explores fine-tuning techniques with several methods.

The novelty is a lack of this paper. All the experiments are interesting but need more analysis in the result discussion. After all the investigation, it is essential to highlight the contribution and conclusions of each one of them.

How High-Speed Object Recognition? Please summarize the answer to this question in the conclusion.

Please compare your metrics with state of the art. DVS and SpiNNaker methods need to discuss with proper metrics analysis.

Section 2 is very limit, Add more recent research in this area. Add the existing drawbacks in end of the section.

Are the parameter number reduced in the fine-tuning approach?

Add some mathematical model to show your proposed approach flow.

Please conclude about the metrics used. Which metrics are more suitable to be used?

Please include a comparison with the state-of-the-art.

Author Response

Thank you very much for your constructive suggestions that have helped us
improve the manuscript. Please refer to the PDF attachment for the point-by-point responses.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors proposed the neural network algorithm for high-speed object recognition using a neuromorphic system based on DVS and SpiNNaker. I suggest the authors fulfill the minor comments before considering possible publication. 

1) Figure 1 label is not clear enough. Please make them clear

2) Please increase the size of all the figures labels and axis numbers

3) Add full form for PSNR and SSIM, Add more information about analyzing PSNR and SSIM (Figure 8)

4) Figure 11 is not clear enough (increase the font size of all legends and labels

5) Conclusion should revised and improved 

Author Response

Thank you very much for your constructive suggestions that have helped us improve the manuscript. The point-by-point responses are shown below:

Point 1: Figure 1 label is not clear enough. Please make them clear.

Response 1: We have changed Figure 1 and make sure it is clear enough now.

 

Point 2: Please increase the size of all the figures labels and axis numbers.

Response 2: We have increased their size to make them more clear.

 

Point 3: Add full form for PSNR and SSIM, Add more information about analyzing PSNR and SSIM (Figure 8)

Response 3: We have removed the discussion of PSNR and SSIM, as well as the original Figure 8, because they are less relevant to the main body of the revised manuscript.

 

Point 4: Figure 11 is not clear enough (increase the font size of all legends and labels.

Response 4: We have increased its font size to make sure it's legible.

 

Point 5: Conclusion should revised and improved.

Response 5: We have revised the conclusion section and added some content. The revised conclusion is shown below:

This paper presents a neuromorphic high-speed object recognition method and system, which applies OR logic aggregation algorithm to acquire enough effective information, and utilizes SNNs to reduce computation in recognition, where the SNN uses LIF neurons and is trained through a surrogate gradient method. The asynchrony and sparsity of DVS and SpiNNaker are used in the implementation of the system to achieve both high-speed data acquisition and computation.

In the experiment with rotating speed from 900 rpm to 2300 rpm, the system achieves the detection rate of more than 99%, and the response time of each letter is within 2.5 ms. In addition, due to the computing mechanism of SNN and SpiNNaker, the number of FLOPs of the system reduce by 96.3% compared with that using the same scaled ANN. Furthermore, our system can be directly transferred to other neuromorphic simulators without retraining. In future work, FPGA is considered as a micro-controller to decrease data communication and power consumption further.

Round 2

Reviewer 1 Report

I am glad to announce the paper has been improved by the authors successfully, as they have addressed all of the questions totally.

Author Response

Thank you for your support!

Reviewer 2 Report

Recommended

Author Response

Thank you for your support!

Reviewer 3 Report

The manuscript is improved but still, the authors didn't marked or highlighted the revised text with updates. It would be great to receive the marked version from the authors. 

1) Still figure 11 resolution is poor. It would be great to see the figures with better resolution. 

2) All the figure's x-axis, y-axis, and label font sizes should be increased. I feel the authors didn't change it as well. 

 

Author Response

Thank you for your helpful comments. Because the manuscript has been significantly revised, the revised text has not been highlighted sentence by sentence.

Point 1: Still figure 11 resolution is poor. It would be great to see the figures with better resolution. 

Response 1: We have redrawn Figure 11, and now we make sure it clear enough for reading.

Point 2: All the figure's x-axis, y-axis, and label font sizes should be increased. I feel the authors didn't change it as well. 

Response 1: We have greatly increased all the figure's x-axis, y-axis, and label font sizes, and now we make sure they are easily recognizable.

Round 3

Reviewer 3 Report

I don't have any further comments. Good luck