A Unified Framework for Anomaly Detection of Satellite Images Based on Well-Designed Features and an Artificial Neural Network
Round 1
Reviewer 1 Report
This is a paper with the title of “A Unified Framework for Anomaly Detection of Satellite Images Based on Well-designed Features and Artificial Neural Network”. In general, Satellite images is a combination of the intrinsic texture of scene and the texture due to noncoherent interactions (noise) that reflect erroneous information. These signal-dependent anomalies (anomalies) within the image present themselves in the form of pixel intensity alterations, which are the source of quality degradations and misinterpretations. Accordingly, authors present a unified framework based on Artificial Neural Network (ANN) to automatically identify such anomalies. In doing so, they used several datasets for verifications and analyzed them.
These kind of research studies are always interesting. However, the presentation method, the literature review, the contribution to the body of knowledge, and more importantly the novelties are playing major roles for such submission. Unfortunately, this paper suffers from Major shortcoming in terms of: Problem description (anomaly), literature review (previous works), Contribution to the body of knowledge (formulations and equations, modeling), novelties (ANN description and interrelations with equations) and verification scenario (proofs and evaluations). Hence, I would reject it Major. It should be noted that the English language was not proper and I highly recommend to polish it up professionally prior to resubmission.
Please consider the comments thoroughly in your manuscript:
1- Title: it is not informative. Please edit it in a way to cover your contribution and novelties. approach must be presented in your title.
2- Introduction: Unfortunately the introduction is not good. it suffers not only from poor English language, but also consistency between the sentences. Please keep text consistency while presenting your approach.
3- Keywords: please select 5 major keywords based on engineering taxonomy.
4- Introduction: it is also the same as previous items. Unfortunately, it suffers from poor English language and lacks text consistency. Please cite to highly accredited references while declaring your contribution to them. Your approach and novelties is not clear. However, in the present style, your technique is not an approach and it is just presented in a normal engineering report. please follow specific pattern of data presentation. A comprehensive literature review would help you to find out a proper way of presentation. please make a connection between items. The way you describe ANN is not suitable. You’d better to focus on the application and modification of the ANN in your research. What kind of ANN are you using? Did you do anything specific thing to it? How did you terrain it? How did you modify it? Did you do anything new to it? Is this a new topic for satellite application? What kind of satellites have used these techniques previously? What are their pros and cons? What are the advantageous and disadvantages of your method?...
Please use relevant citations rather than sporadic citations. Please fix it all. A good literature review would help you. Please see the item below and compare your work to them precisely. Please show your contribution to them while citing in your paper. please follow a specific pattern of data presentation
A) [DOI: 1109/TITS.2021.3055614]: in terms of anomaly detection and ANN application
B) [DOI: 1109/LGRS.2018.2869337]: in terms of anomaly detection and background presentation
C) [DOI: 1109/ACCESS.2020.2976815]: in terms of anomaly formulation and evaluation
D) [DOI: 1109/TGRS.2020.2985011]: in terms of filter application and verification
E) [DOI: 1109/TGRS.2016.2569450]: in terms of application with real data
5- Dataset: Please use similar dataset under different condition for the purpose of verification. Otherwise, you should present a model for anomaly (equation or specific pattern of presence) that can be used as a standard basis for extraction. This section really lack anomaly formulations. adverse effects of anomaly? Pixel behaviors?... please follow the references above.
6- Methods: this section lacks description and contribution declaration. the method has not been presented clearly at all! Equations are just simple equations without any contribution to the body of knowledge. this is also true about Section. 3.3. Please make a connection between anomaly description, anomaly formulation, ANN architecture, ANN modification and ANN operation!! This is the core section of the paper but suffers major problems. Please fix them.
7- Experiments and results: this section lacks verification. The results must be justified by the equations. You should present a global verification scenario. Please look at the sample references above for following a specific pattern of presentation.
In short, in the present style it is similar to a simple engineering report other than a scientific research item. please find a proper sample and follow suit. Hence, i would reject it major.
Stay safe,
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Good for the publication in the present form
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
I liked this paper very much. It is clear, concise, the study is performed properly and results presented descriptively and in a clear manner. One question tough, why do you repeat 300 times? Wouldn't 100 times be enough? 300 seems like an overkill in my opinion.
Minor remarks:
Line 122 is redundant since they are already placed in the introduction. Probably focus on the organization of the rest of the paper. Also in 120 the sentence looses its importance because it is being placed together with the organization sentences... Rephrase it and emphasise the lack of research and publications on this topic more.
- line 35: detective -> defective
- line 119: features suitable for our case is thus needed. -> are thus needed.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
The paper is well written, but it is unstructured: - Introduction is too long and difficult to follow. I recommend authors to split into two sections, the introduction (motivation, description, objectives, etc.) and state of the art. Do not use enumeration in the last part of the introduction. - The methodology of the article is not correctly described and it is disseminated along with the article, authors should rewrite the paper taking into account this point. - The experiments are good, but the discussion and conclusion should be compared with the state of the art, there are many articles in the information fusion area about this topic that should be taking into account. Also, there are many architectures to do so, that should be taking into account.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Despite the modifications which are minor improvements, the paper has changed to a slight extent, and it is far from an original research manuscript. As I commented the authors previously, formulations must justify the results and the results must be verified accordingly. Unfortunately, the authors claimed their research as an unique research, which is not convincing. An original research manuscript must have contribution to the body of knowledge. The authors should have worked on a novel method (in their field of study), which is not known throughout the entire paper! To make it short, the authors couldn't convince me that their approach is really something.
How did the authors work on anomaly while they do not have any specific model for it!! How can you verify your approach while there is no specific metric for it? How can you simulate the result while there is no pattern of anomaly behaviors? Unfortunately, using known modules of image processing is not considered a contribution. Regardless of your great attempt, i have to reject it. Please start from scratch.
stay safe,
Reviewer 4 Report
The article is good enough
