A Novel Object Detection Method of Pointer Meter Based on Improved YOLOv4-Tiny
Round 1
Reviewer 1 Report
This research paper proposes a YOLOv4-Tiny-based pointer meter detection model named pointer meter detection-YOLO (PMD-YOLO) for accurate reading of pointer meters in industrial settings. The researchers utilize a GhostNet feature extraction network with a channel attention mechanism to reduce the model's weight while ensuring high object detection accuracy. An improved receptive field block (RFB) module is added after the backbone network to enhance the feature extraction ability of small and medium-sized targets, and a convolutional block attention module (CBAM) is introduced into the feature pyramid network (FPN). The FPN is optimized to improve feature utilization and detection accuracy. The researchers experimentally verify the effectiveness and superiority of the proposed PMD-YOLO model using a constructed dataset of pointer meters. They compare the performance of PMD-YOLO with target detection algorithms such as Faster region convolutional neural network (RCNN), YOLOv4, YOLOv4-Tiny, and YOLOv5-s under the same conditions. The experimental results show that the mean average precision (MAP) of the PMD-YOLO is 97.82%, which is significantly higher than the other algorithms. Additionally, the weight of the PMD-YOLO is 9.38M, which is significantly lower than the other algorithms.
I have following Questions.
Can you explain in more detail how the constructed dataset of pointer meters was created? Were there any limitations or biases in the dataset creation process that could affect the generalizability of the results?
Can you provide more information about the practical applications of the proposed PMD-YOLO model? How well does it perform in real-world industrial settings with different lighting, shooting angles, and complex backgrounds?
Can you discuss the limitations of the proposed PMD-YOLO model? For example, are there any specific scenarios or situations where the model may not perform as well or could potentially misread the meter?
How does the proposed PMD-YOLO model compare to other state-of-the-art object detection algorithms that have been developed for reading pointer meters? Are there any specific advantages or disadvantages of PMD-YOLO compared to these other algorithms?
Can you explain the rationale behind the design decisions for the PMD-YOLO model, such as the use of GhostNet with channel attention mechanism, RFB module, CBAM, and FPN? Are there any potential trade-offs or drawbacks to using these specific components in the model?
How do you ensure that the results reported in the paper are statistically significant? Are there any potential sources of bias or error in the experimental methodology that could affect the validity of the results?
Can you provide more information on the computational resources required to train and run the PMD-YOLO model? Are there any practical limitations or scalability issues that could affect its adoption in real-world industrial settings?
Author Response
Thank you for your comments. Our detailed reply is in the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
1. Section 2.1, Line Number 168: What SE attention module adopted can be mentioned.
2. Section 2.1, Line number 180: What are different object categories identified has to described. Is it only one object category ? But , in real world the size and shape of pointer meter may vary.
3. Section 2.1, Line number 1801: CIoU was used to calculate the loss of the prediction frame and the true frame. Provide the description of CIoU metric.
4. Section 3.1, Line number 323: Web crawler was used to collect images of pointer meter. But, what keywords were used during web crawling process should be stated. Dataset collection process can be explained further in detail.
5. Section 3.1, Line number 324: LabelImg was used to create labels for YOLO . but what are different types meters that was labeled or it is only one category of pointer meter was labelled. Further description of labeling process should be provided.
6. Section 3.1, Line Number 338: Table 1, The first column heading can be changed accordingly as 'Component' instead of 'Experiment'.
7. During experimentation, how the different lighting conditions of the objects were taken care. Because pointer meters present in the industries which will be with different lighting conditions.
8. Section 3.3, Line number 450: Figure 11, Results displayed on Bounding Boxes are not readable. A separate text box may be included, else there's no meaning of these figure, where input image is visible and the predicted output , the confidence score is not readable.
9. Section 3.3, Line number 468: In Figure 12, Justify why the Scene-1 and Scene-2 images for Hough circle transform and PMD-YOLO is different from the Scene-1 and Scene-2 images shown in Figure 11.
20. Reference paper listing can be taken care as per the journal format.
Author Response
Thank you for your comments. Our detailed reply is in the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
In this manuscript, In order to verify the effectiveness and superiority of the PMD-YOLO proposed in this 24 paper, the PMD-YOLO is used for experimental research on the constructed dataset of pointer 25 meter, and the target detection algorithms such as Faster region convolutional neural network 26 (RCNN), YOLOv4, YOLOv4-Tiny, YOLOv5-s are compared under the same conditions. The experimental results show that the mean average precision of the PMD-YOLO is 97.82%, which is 28 significantly higher than the above algorithms.
Accepted as it is..
Author Response
Thank you for your comments.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
.Authors revised the paper carefully.