A Real-Time Negative Obstacle Detection Method for Autonomous Trucks in Open-Pit Mines

Round 1

Reviewer 1 Report

This paper proposed a new detection network based on YOLOv4. The detection ability of the new network for negative obstacle targets was improved by replacing the backbone ( RepVGG), adding an attention module (the SimAM), and optimizing the NMS method (the CS-NMS method). Furthermore, the performance of the proposed network was compared with several mainstream target detection networks based on the detection results of the new network. It presents obtained experimental results in a detailed and comprehensive manner. I have some concerns that I hope the authors can fully address.

(1).  Page6, This paper selects the SimAM attention module to mine the important features of each neuron between channels. Why did the authors choose such an attention module rather than other modules like CBAM, SE, etc? Please explain the advantages of the SimAM module.

(2).  Page10: The validation dataset is used to adjust the arguments in the training. For the dataset of small scale (below 10000), the scientific ratio of the training set, validation set, and test set is about 6:2:2 or 8:1:1. In TABLE Ⅰ, this ratio in the paper is not conventional, and please explain the basics of division.

(3).  The paper used VOC dataset in Section IV and used VOCTest in table III. Please solve this controversy.

(4).  In page 4, line 143, mentioning Fig. 2(b) is incorrect, not Fig. 2(b) alone.

(5).  Please change the caption of Fig. 15. Is it an algorithm?.

(6).  (6) Some recent references are useful for this paper, for example,

[1] F. Xia, R. Hao, et al., Adaptive GTS allocation in IEEE 802.15. 4 for real-time wireless sensor networks, Journal of Systems Architecture 59 (10), 1231-1242, 2013.

[2] P Kumar, R Kumar, et al., PPSF: a privacy-preserving and secure framework using blockchain-based machine-learning for IoT-driven smart cities, IEEE Transactions on Network Science and Engineering 8 (3), 2326-2341, 2021.

[3] C. Wu, C. Luo, et al., A greedy deep learning method for medical disease analysis, IEEE Access 6, 20021-20030, 2018.

[4] H. Cheng, Z. Xie, et al., Multi-step data prediction in wireless sensor networks based on one-dimensional CNN and bidirectional LSTM, IEEE Access 7, 117883-117896, 2019.

[5] Y. Yao, N. Xiong, et al., Privacy-preserving max/min query in two-tiered wireless sensor networks, Computers & Mathematics with Applications 65 (9), 1318-1325, 2013.

Author Response

Thank you for your careful review, which is very helpful to improve the quality of our paper. Our team has carefully considered and summarized each of your review comments, and carefully revised the manuscript. Please see the attachment for details.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper is interestign and of a very good possible practical aplicability.

I could not find any scientificly unsound things, only minor punctuation missing (for instance captions of figure not al end with a period . )

Also the references look like not the MDPI template is used for it.

As a suggestion, i would advise for more non-asian authors to be cited in the work, as this topic is not limited to literature only from China as far as I am concerned, but also in Europe (Germany, Romania, Poland) and the Americas there are scholars researching this topics in mining.

Author Response

Thank you for your careful review, which is very helpful to improve the quality of our paper. Our team has carefully considered and summarized each of your review comments, and carefully revised the manuscript. Please see the attachment for details.

Author Response File: Author Response.pdf