Small-Target Detection Based on an Attention Mechanism for Apron-Monitoring Systems
Round 1
Reviewer 1 Report
It is generally a well-written and presented paper. Few suggestions to authors:
- There are multiple grammatical errors and typos throughout the manuscript. Those must be corrected.
- There are erros in symbols and numbering. For example: © instead of (c). Please correct.
- The authors are asvised to highlight the major drawbacks and limitations of the proposed approach and measures to correct those.
- Some more recent works and citations on the topic in Introduction and Related works sections would be great.
- Conclusions must be revised by highlighting the major outcomes/results (that are better compared to other available works), future perspectives, and general take-home-message from the present study.
Author Response
Please see the attachment
Author Response File: Author Response.doc
Reviewer 2 Report
The manuscript proposes a small-target detection model based on attention mechanism. Based on the experimental results, the effectiveness of the proposed method for detecting small targets is verified. But I have some questions about proposed method.
1. How is hyperparameter defined in comparison network?
2. If the detection accuracy of small targets such as pedestrians in the airport is high, does it mean that the application in other small target data sets also have good effects? Have you done any experiments on that?
3. Have weights of Mask R-CNN, Mask Scoring R-CNN and every part of your designed network been initialized using ImageNet pre-trained weights?
4. There are too few control networks, and are there any comparisons with classic networks such as YOLOX, SSD and Faster-RCNN?
5. When the accuracy is improved, whether the number of parameters, training time, occupied memory, reasoning speed and other conditions of the network are also considered?
Author Response
Please see the attachment
Author Response File: Author Response.doc
Reviewer 3 Report
This paper proposes a small-target detection model based on attention mechanism. A standard airport small-target dataset is established and an attention module was added to the feature extraction network to enhance feature representation and improve the accuracy of small-target detection. The experimental results demonstrate the effectiveness of the proposed method for detecting small targets. Here are some comments and suggestions for further revision before it can be accepted for publication.
The authors should make the abstract part more concise. According to the Instructions for Authors (www.mdpi.com/journal/applsci/instructions), the abstract should be a total of about 200 words maximum.
Section 2 introduces the related works of small-target detection, in which deep learning techniques play an important role. More relevant examples should be presented, such as: such as: doi:10.3390/app12094356; doi:10.1007/s00170-022-10335-8.
Section 4.2 lists the details of model training, e.g., optimizer, learning rate, epoch, etc. Have the authors tested the training performance of other training parameters? Or the authors should clarify the reason why these parameters are selected.
Overall, this paper is well-written with scientific soundness, and the experimental results are pretty interesting.
Author Response
Please see the attachment
Author Response File: Author Response.doc
Round 2
Reviewer 2 Report
The paper has been basically revised according to the revision suggestions. However, the reply to question 5 is not reflected in the revised paper, as shown in Table 1 in the comments reply.
Author Response
Many thanks for your commenting. We have put the results of metrics in the experimental section, as shown in table 6.