Visual Detection and Association Tracking of Dim Small Ship Targets from Optical Image Sequences of Geostationary Satellite Using Multispectral Radiation Characteristics

Round 1

Reviewer 1 Report

This paper proposes novel approach to visual detecting and association tracking of dense ship targets based 18 on the GF-4 image sequences. Here are some problems:

1. The motivation is missing. Who is the end user of the research or it is an academic research?

2. Figure 2 is too vague to identify the contents.

3. The font and size of the formula in Chapter 2 need to be adjusted.

4. The number of comparative experiments is too small to reflect the effectiveness of the proposed method. In addition, some comparative experiments based on deep learning can be added.

5. IMHO, the Conclusion should be re-written to 1) explicitly describe the essential features/advantages of the proposed method that other methods do not have, 2) describe the limitation(s) of proposed method, and 3) what aspect(s) of the proposed method could be further improved, why and how.

6. The English should be improved greatly.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors studied ship target recognition and tracking against complex background. Through visual significance and spectral radiation characteristics, a target detection and tracking method based on GF-4 image sequence is proposed. The authors analyzed the multi-vision salient features fusion strategy and obtained suspected target areas from the spectra. Based on the spectral displacement characteristics, the detection and filtering of moving targets are achieved, and the target position correction is compared with the data of automatic recognition system. The target tracking algorithm and motion state estimation are also studied. The experiment looks good, but there are a few questions.

 

1. GF-4 is a medium-orbiting satellite, belonging to the geosynchronous orbiting satellite, with a pixel resolution of only 50 meters. It should be clear to identify the target how large the vessel is. For cargo ships, it is generally only more than 100 meters long, less than 50 meters wide, and a total of less than 3 pixels, which is not enough to identify the target. Aircraft carriers are also more than 300 meters, with only six pixels. Does not meet Johnson's Goal Recognition Criteria. It is difficult to recognize the target of such a few pixels on an image with such a large width.

 

2. The author mentioned using spectral information to discriminate target location and size, but did not analyze the spectral characteristics of the target. Instead, he gave a graph explaining the effect of the spectrum on target recognition. This is not enough to explain that the methods used by the authors used spectral characteristics or how spectral information was used. In essence, the brightness information of the image is used, but only the NIR band image is used, and there is no innovation in the method.

 

3. In Section 2.3, the authors refer to multi-frame image target tracking, whereas the summary refers to single-frame motion detection using spectral shift characteristics, which is misleading.

 

4. Figures 8 and 9 in the experiment use pseudo-color diagrams to show the effect of target recognition. The pictures are blurry (including the MATLAB results below). And how to identify the salient map after it is not mentioned, is it just through the threshold?

 

5. In the quantitative analysis phase, only the table recognition rate is given. Have you experimented with only one moment in area 7? The experimental data includes different times and changing environments, such as clouds and islands, where there is more interference? No images such as RC curves are given to analyze the recognition rate. Comparing algorithm [3] through shape information analysis, there is no shape information at all for such a few pixels, but the experimental data in Table 1 has a high recognition rate, which I cannot understand here.

 

6. Moreover, the experimental comparison algorithm does not give the result image comparison of the three literature algorithms. In general, the experimental comparison data are insufficient.

 

7. Most of the studies cited in the references were analyzed a few years ago, but the results of the last two years were not used.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

1. The process from salience maps to candidate regions is very not robust if only gray thresholds are used for segmentation, and the author should think carefully about whether there is a better method.

 

2. Figure 14 does not give the true result. In my view, it looks like there are eight ships in figure e, but the algorithm only finds six. True values in a plot are more difficult to judge, and the advantages of the algorithm cannot be determined.

 

3. The comparative data of the literature [3] and [5] are given in the quantitative analysis, but the results are not given in Figure 14 of the qualitative analysis. Presentation of experimental data is the biggest problem with this manuscript.

 

4. The author mentions that data in complex environments is not readily available and therefore images of such scenes are not displayed. I think this is incomplete. Since the author has data for clean scenes, there should also be data for other scenes. Is it because the algorithm can't handle such scenes? The scenario to which the algorithm applies should be given as a precondition. The experimental data given is still too small to be convinced.

 

5. Since the author uses accuracy and recall as evaluation criteria, the PR curve should be plotted, but the author has not given it yet.

 

6. The author's innovation is to use spectral characteristics to track the target, but most of the author's processing is based on the technical means of processing the image, not on the principle of using spectral features, or on the difference between the brightness value of the target image and the background.

 

7. Authors always think about how to reject reviewers without making substantial changes that are inspired by their opinions. Overall, the author's revision is somewhat unsatisfactory.

Author Response

Please see the attachment.

Author Response File: Author Response.docx