Prototype Calibration with Feature Generation for Few-Shot Remote Sensing Image Scene Classification
Round 1
Reviewer 1 Report
The classification of remote sensing images is an important and hence a widely-researched area. A huge number of approaches have been studied in depth and some techniques have been reported to have really good classification accuracies. However, it has become really difficult to compare and assess these techniques with accuracy and fairness because performance is inevitably domain-dependent.
This paper concentrates on the application of few-shot scene classification - attractive because of the simplifications potentially possible in the training of the model. A feature encoder is developed and trialled on two existing remote sensing datasets. The classification results are compared with those obtained from several other classification techniques and the new approach appears to perform very well.
In general, I'm happy with this paper. It follows the standard structure with a discussion of the problem, related work, details of the new technique, results and assessment. I have a few relatively minor comments
The introduction contains a reasonable coverage of few-shot classification. It would be good to see something specific about the types of image that are well-suited to classification using this approach. The images in Figure 1 are too small to be useful.
Related work is short, but fine.
The methodology is well-presented and probably at the right level of detail. In the problem formulation, it would be helpful to ensure that it is stated that the number of samples values are per category. The equations all seem nicely presented.
In the experiments and discussions sections, it would be helpful to have references to the two datasets that are being used so that readers can experiment too. And in sections 4.1 and 4.2, some seemingly arbitrary decisions about parameter values are being made. It would be helpful to see some sort of justification, even if to say that the choice was made as a result of what seemed to work! In particular, I was puzzled by the choice of 84 pixels. Section 4.3 needs to be qualified a little since you aren't doing a proper statistical analysis of the results. Figure 6 needs the caption to be moved so that it doesn't cover the histogram bars. The analysis in the whole section is comprehensive and sensible.
References are good - they form a useful part of the paper.
The paper as a whole is well-written - the English is a little erratic in places, but the meaning is clear. It's not an easy read, but a great deal of information has to be conveyed in a limited number of pages.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The research titled “Prototype Calibration with Feature Generating for Few-Shot Remote Sensing Image Scene Classification” represents an interesting original scientific research. This research presents novel prototype calibration with feature generating model is proposed for few-shot remote sensing image scene classification. The manuscript should attract an audience in the field of remote sensing and image processing. The manuscript title is accurate and concise. In the entire manuscript, the authors use standard technical and scientific terminology. The manuscript is very well written. After the Introduction, the authors explained related works and developed methodology. The experiment and results were conducted according to the scientifically correct approach. The manuscript topics fit in the Remote Sensing journal and scope.
While the topic of the manuscript is new and interesting, the manuscript can be improved. I recommend this manuscript can be accepted after minor revision.
Comments for authors:
- I suggest to change section title “Experiments and Discussions” to the “Results and Discussions”.
- Please better explain the practical application in real cases of this research.
- Use MDPI standard font (Palatino Linotype) on figures if you can.
- Please provide adequate citations for all equations in the manuscript.
- All variables must be introduced and in the same font style and font size in the equations, tables, and manuscript text.
- Please, double-check all references and reference style.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
