Target Recognition Based on Singular Value Decomposition in a Single-Pixel Non-Imaging System
Yanfeng Bai
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsA single-pixel non-imaging target recognition scheme was proposed by using the singular values of the target objects in this manuscript. From the simulation and experimental results, it was shown that the target images can be quickly and accurately recognized after directly recording the single-pixel signals. The results are interesting, and my comments are as follows.
1. “One can quickly and accurately recognize the target images after directly recording the single-pixel signals.” Where does it reflect “quickly”? As mentioned in the introduction part, “this approach necessitates high-speed imaging techniques that can incur significant time costs and result in the generation of large volumes of data”. The authors should provide comparison results about the time costs by comparing with other non-imaging single-pixel target recognition methods.
2. In this scheme, the targets need be labeled, and SVD is performed on the target image, then one constructs the projection patterns by the left and right singular matrices corresponding to each singular value. Do the projection patterns need be reconstructed when the targets are changed? In addition, do we need to know the target information in advance?
Based on the above comments, I can recommend publishing this manuscript after
the authors response properly those comments.
Author Response
We thank the referee for the expert comments and valuable suggestions. In response, we address all the comments and concerns below.
Comments 1: “One can quickly and accurately recognize the target images after directly recording the single-pixel signals.” Where does it reflect “quickly”? As mentioned in the introduction part, “this approach necessitates high-speed imaging techniques that can incur significant time costs and result in the generation of large volumes of data”. The authors should provide comparison results about the time costs by comparing with other non-imaging single-pixel target recognition methods.
Response 1: "Quickly" is reflected in the recognition stage. After the bucket detection signal is collected by the single-pixel imaging system, target identification is successful by directly observing whether the bucket detection signal has a maximum value at the label position, without additional data calculation process. In contrast to the research results reported by Chen et al. [30], we save the time of compressed bucket detection signal, Fourier coefficient calculation and hash distance calculation, thus significantly saving the time cost. In accordance with the suggestion, we have included a concise discussion in the revised manuscript. This discussion is detailed in Lines 249 to 264 on Page 7 of the revised manuscript.
In reference to another article [32] on non-imaging single-pixel target recognition, the support vector machine (SVM) model is used as the classifier in the recognition stage, which requires a training phase. In contrast, our method does not require any training.
Comment 2: In this scheme, the targets need be labeled, and SVD is performed on the target image, then one constructs the projection patterns by the left and right singular matrices corresponding to each singular value. Do the projection patterns need be reconstructed when the targets are changed? In addition, do we need to know the target information in advance?
Response 2: In our scheme, each target image corresponds to a specific projection pattern, which is obtained by Eq. (9). Then all the projection patterns form a library of target features. If a new target is added into the target set, its specific projection pattern should be added into the feature library. In the revised manuscript, Eqs. (7) and (9) have been modified. Below Eq. (10), we have added two sentences to clarify this comment, “For instance, we choose j=2 in Eqs. (9) and (10) in our simulation and experiment. All the patterns form a feature library for target recognition.”
Therefore, when the targets changed, the projection patterns should be updated. The target information, especially the singular value decomposition of the target image is performed in advance.
We sincerely appreciate the time and effort the reviewers have dedicated to evaluating our manuscript. The manuscript has been revised according to these comments and suggestions. We hope now the manuscript can be considered for publication
Reviewer 2 Report
Comments and Suggestions for AuthorsSingular value decomposition is a widely used technique that is also used in image analysis. On the other hand, single pixel imaging enables image acquisition with a single pixel detector. Combining the above two tasks may renders the best of both in target image recognition. Thus the authors implement target recognition by single pixel imaging and using singular value decomposition method. Numerical simulations validate the method.
The paper is well organized, but I want to make some small suggestions in order to improve the manuscript readability. First it should be a rephrasing of the paragraph beginning with the line 117 and ending with the line 122, equation (5).
Secondly, the authors say in the Introduction and I quote that "To our knowledge, it is still a challenge to recognize targets in a non-imaging scheme, by combining SVD and SPI". Ok, the authors should clearly state what challenges they overcame in this paper.
So, I recommend the paper for publication if these two questions are soundly addressed.
Comments on the Quality of English Language
The English seems all right.
Author Response
We appreciate the referee for his/her professional comments on our manuscript. In response, we address all the comments below.
Comments 1: The paper is well organized, but I want to make some small suggestions in order to improve the manuscript readability. First it should be a rephrasing of the paragraph beginning with the line 117 and ending with the line 122, equation (5).
Response 1: We sincerely thank you for meticulous reading of our manuscript. According to your suggestion, we have rewritten the phrasing of formula (5) and marked it in the revised draft. We apologize for the careless typos.
Comments 2: Secondly, the authors say in the Introduction and I quote that "To our knowledge, it is still a challenge to recognize targets in a non-imaging scheme, by combining SVD and SPI". Ok, the authors should clearly state what challenges they overcame in this paper.
Response 2: In this paper, we carried out an innovation in the calculation and identification stage after collecting the bucket detection signal. We can judge whether the identification is successful by directly observing whether the bucket detection signal has a maximum value at the label position, without additional data calculation process. In contrast to the research results reported by Chen et al. [30], we save the time of compressed bucket detection signal, Fourier coefficient calculation and hash distance calculation, thus significantly saving the time cost. For further discussion of this part, we have added lines 249 through 264 on Page 7 of the revised manuscript.
We sincerely appreciate the time and effort the reviewers have dedicated to evaluating our manuscript. The typos have been corrected, and key materials have been added in the revised manuscript. We think the manuscript is more readable, and can be considered for publication.
