Improved YOLOv5: Efficient Object Detection Using Drone Images under Various Conditions
Round 1
Reviewer 1 Report
Authors proposed an approach to object detection from drone images based on Yolo v5 architecture. Authors proposed a several classes including person and car. They proposed NN model based on Yolo v5. The paper describes different types of detectors (one stage, two stage). There is a description of why one stage detector is selected. Also it includes a description of why the Yolov5 was chosen as a detector. The authors describe in sufficient detail why ELU was chosen as the activation function, compared with RELU and SiLU. Also in the paper there is a lot of reference information, for example in the form of metrics. Accordingly, a person who is familiar with this material for the first time will be much easier to understand everything in general.
However the scientific novelty of the paper is not completely clear. Replacement of the activation layer is a common practice in CV. It seems that it is not enough to put a lot of emphasis only on it. The paper has a lot of similar schemes that do not say about good scientific story.
Test sample is extremely small. So all the statistics mean little. One out of 10 times the results may be better, and the other times worse.
The article focuses on poor and difficult environments, but the authors only showed a couple of images with difficult conditions. Here it was necessary to concentrate more on bad conditions, augmenting the data accordingly for this. Adding rain, snow and so on.
The authors also mention that their drone has a 4K resolution camera. However, they do not report how they used this camera and whether they even trained a 4K neronet. Training at this resolution is a big problem, as there are a lot of calculations to be done.
Author Response
Please see the attachment. (Include Response to Academic Editor Comments, Response to Reviewer 1 Comments and revised Manuscript)
Author Response File: 
Author Response.docx
Reviewer 2 Report
The manuscript reports a methodologically oriented study. It is already in a mature status and matches the profile of this journal. There are however three points which the authors should consider when preparing a revised version of the manuscript:
The introduction begins very technical, with a definition of a drone. What I miss is some sort of introductory paragraph dealing with the societal relevance of drones. For what research fields are drones important and why? This could also be coupled with a strengthening of the literature body.
In terms of the study background, I miss more relevant research indicating the potential of the steadily growing and developing drone market. For example, recent studies highlight potentials of low-budget drones for spatially oriented citizen science applications, which emphasizes the public interest in the growing market and new approaches to optimizing drones (see for e.g. https://doi.org/10.3389/frobt.2022.886240 & https://doi.org/10.3390/heritage2020089)
A clear lack of this manuscript can be found at the end of the paper. Neither the (differntiated) discussion sections nor the conclusion section refer to any related study. In how far does your study contribute to an extension, back-up or even contradiction of previously published and established studies? Without this linkage, your paper would remain a solitary piece which is not connected to the international debates. This should neither be an achievement of the authors nor of this impact journal.
Author Response
Please see the attachment. (Include Response to Academic Editor Comments, Response to Reviewer 2 Comments and revised Manuscript)
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Authors fixed most of comments, however, please, highlight the following in the paper:
1. Formulate scientific novelty of the paper in introduction.
2. Extend the dataset more. In general you should have a t least 1000 images per class.
Author Response
Please see the attachment. (Include Response to Academic Editor Comments, Response to Reviewer 1 Comments and revised Manuscript)
Author Response File: Author Response.docx
Reviewer 2 Report
The authors re-submitted the manuscript which they changed according to the reviewers' comments. The response letter is quite short. However, it points to the changes made, and the argumentation is ok. Against this background, I would like to recommend this manuscript version for publication in this journal.
Author Response
Please see the attachment. (Include Response to Academic Editor Comments, Response to Reviewer 2 Comments and revised Manuscript)
Round 3
Reviewer 1 Report
Authors fixed all the problems.
