Next Article in Journal
Quantifying Hail Damage in Crops Using Sentinel-2 Imagery
Previous Article in Journal
Reconstruction and Characterisation of Past and the Most Recent Slope Failure Events at the 2021 Rock-Ice Avalanche Site in Chamoli, Indian Himalaya
 
 
Article
Peer-Review Record

Multi-Oriented Object Detection in High-Resolution Remote Sensing Imagery Based on Convolutional Neural Networks with Adaptive Object Orientation Features

Remote Sens. 2022, 14(4), 950; https://doi.org/10.3390/rs14040950
by Zhipeng Dong 1, Mi Wang 2,3,*, Yanli Wang 4, Yanxiong Liu 1,5, Yikai Feng 1,5 and Wenxue Xu 1
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Reviewer 4: Anonymous
Reviewer 5: Anonymous
Remote Sens. 2022, 14(4), 950; https://doi.org/10.3390/rs14040950
Submission received: 21 January 2022 / Revised: 6 February 2022 / Accepted: 14 February 2022 / Published: 16 February 2022
(This article belongs to the Section Remote Sensing Image Processing)

Round 1

Reviewer 1 Report

This paper proposes a high resolution image neural network target detection recognition method. The manuscript has a clear research background, detailed description and sufficient comparative experiments. However, it still needs to be carefully modified according to the following requirements:

  1. The fundamental principle of adaptive object orientation regression method should be described clearly ABSTRACT.
  2. The existing comparison methods used in the experiment (Faster-RCNN, CNN-SOSF, YOLOv2 and YOLOv3) should be described in the INTRODUCTION, and the significance of using these methods as comparison should be explained.
  3. In figure8, for ship targets, the PRC value of the algorithm proposed in this paper is not better than that of other comparison algorithms. The phenomenon in the figure is inconsistent with the description, please explain.
  4. In line 226-237, the determination method of confidence threshold should be given.
  5. The layout of Figure 9 is not reasonable (should be figure 11?). Please redesign the arrangement of experimental presentation and compare the experimental results of the same scene as much as possible.
  6. Conclusions are fragmented and need to be reorganized in a more logical way.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper presents a modification of a neural network. The algorithm marks the area in which the classified object is located taking into account its orientation. This is a very valuable part of the article. However, I have two questions:
Adding a parameter related to orientation affects the required number of examples in the training set, has this effect been studied?
Does the object recognition time increase?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

 

The submitted work presents an object detection method for remotely sensed data, based on CNN that benefits from adaptive object orientation features. The proposed pipeline starts with an adaptive object orientation regression followed with a CNN model for object detection.

My main concern is about the innovation level of the submitted work that does not seem adequate. The overlap between the proposed method and some other similar techniques is relatively high. Furthermore, literature review is week as several interesting works with related or same concerns are not addressed in the manuscript. One can mention tens of such works, but as an example I would cite ‘AIoU: Adaptive Bounding Box Regression for Accurate Oriented Object Detection’ by Wen et al. published in August 2021.

For a fair benchmarking, I would also like to check the results of the proposed method on other standard datasets like trainval (or trainval35k) as they are used by several other state-of-the-art approaches with the same concern.

Also, I suggest using a combination of different criteria (such as recall rate, missed detection rate, and confidence) as evaluation indicators.

Text needs significant level of improvement. There are several typos, impaired/unclear/vague sentences here and there throughout the manuscript.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

The manuscript is devoted to the application of convolutional neural networks to detect the multi-oriented objects in high-resolution remote sensing imagery with adaptive object orientation features. The topic is interesting and actual. The structure of the manuscript corresponds to requests for this type of publication. The annotation briefly reflects the manuscript content. The introduction contains the actuality of the problem and the analysis of current research in this subject area.  The Material and Methods section clearly described the used techniques and steps of the research. The images and charts to my mind are qualitative ones, and they reflect the content of the manuscript. The results also are described clearly and correctly. I think, that this manuscript can be accepted after minor revision. I have one remark that can be taken by the authors into account.

1. The paper will look better if after an analysis of current research before the main contribution of the paper to allocate the unsolved part of the general problem.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 5 Report

The paper presents an adaptive object orientation regression method to obtain object region in any direction. Results shows better performance as compared to existing methods. Paper is good written in a good English. Method is well presented and the comparison is well made.

To be honest I think that the article is quite good and well made. Therefore I have few comments and am inclined towards acceptance.

Here is my main comments:

  • The relation with Remote sensing - the main topic of the journal, should be better explained
  • I would suggest more literature review to be made in the introduction.
     

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The author has carefully revised and improved the comments in the first round review. The reviewer believes that the multi-oriented object detection in high resolution remote sensing imagery provided in the revised paper has reference value and could be published.

Back to TopTop