Research and Explainable Analysis of a Real-Time Passion Fruit Detection Model Based on FSOne-YOLOv7

Round 1

Reviewer 1 Report

This manuscript presents a compelling study that is relevant to the field of agricultural automation, specifically focusing on the intelligent harvesting process of passion fruit through real-time object detection. The authors have identified a key problem area, that being the challenges presented by the diverse appearance characteristics of passion fruit in various growth environments, and have proposed a solution through the FSOne-YOLOv7 model.

The FSOne-YOLOv7 model represents an innovative approach to improving object detection by integrating ShuffleOne as the backbone network and slim-Neck as the neck network into an enhanced version of YOLOv7. This combination of networks aids in feature extraction and fusion.

From the performance scope, the experimental results indicate a notable improvement over the original YOLOv7, as well as other established models like Faster-RCNN and SSD. The increase in precision and mean average precision (mAP), coupled with the reduction in parameter count and improved real-time detection speed, is promising.

Additionally, the manuscript is articulated with commendable clarity. The incorporation of the Gradient-weighted Class Activation Mapping methodology significantly contributes to attaining explainability, a component that is frequently absent in deep learning models. This inclusion is instrumental in fostering an enhanced comprehension and reliability of the model.

The authors astutely pinpoint and accentuate the aptitude of the model for deployment in contexts characterized by constrained memory and computational resources, a common occurrence in agricultural settings. Moreover, the paper posits that the model holds the potential to function as a benchmark in the technical domain, extending its applicability beyond passion fruit to encompass other analogous fruits. This represents a noteworthy contribution to the sphere of real-time object detection within the agricultural sector.

Some areas of interest that could be further explored for the sake of comprehensiveness, without however affecting the specific work at hand, would include an examination of how the FSOne-YOLOv7 model adapts or performs under varying environmental conditions, which is crucial in agriculture. Additionally, insights into how this model could be integrated into harvesting machinery or drones, and an exploration of the practical challenges associated with such integration, could be highly valuable. The paper compares the FSOne-YOLOv7 model to other models quantitatively; however, a qualitative analysis of why it outperforms others might offer additional depth. Moreover, it would be beneficial to investigate the scalability of this model for more extensive agricultural settings or for detecting a variety of fruits simultaneously. Finally, given the seasonal nature of agricultural applications, an evaluation of the model’s performance across different harvesting seasons could also provide important insights.

In summary, the work is accepted, with the consideration that it could benefit from an exploration of several dimensions: the adaptation and performance of the FSOne-YOLOv7 model under various environmental conditions which are integral to agriculture; the integration of the model into harvesting machinery or drones and an investigation into the practical challenges involved; a more qualitative comparative analysis to understand why it might outperform other models; an examination of the scalability potential for larger agricultural settings or the simultaneous detection of diverse fruits; and an evaluation of how the model performs across different harvesting seasons due to the seasonal nature of agricultural applications.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Review: In this manuscript (agronomy-2479598), the authors present a Real-Time Detection Model of Passion Fruit Based on FSOne-YOLOv7. I consider this study to be very interesting and it contributes a significant advancement in artificial vision aimed at real-time fruit detection that can be extrapolated to other fruits. The text is well written and substantiated. I am only going to make some formal suggestions:

(1): Lines 83, 118, 130, and 156: The link to the reference has been lost: “Error! Reference source not found”

(2): Figures 1 and tables 1 and 8 are not cited in the text.

(3): There is a change in font size between lines 461 and 487.

(4): Since this work can be the basis for future work, I suggest that in the "Data Availability Statement" section, a direct link to the material used in this work is included.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This is a good application paper of the Yolov7 to passion fruit detection. One thing needs to be revised is that your title does not represent the contents of your research work properly. Please revise the title more appropriately. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

The author presents in this paper improved YOLOv7 network for time detection model of passion fruit. The manuscript is generally well-designed and the results are adequate from a scientific point of view. However, the paper can be accepted after the following modifications.

1.The authors are invited to check reference. Moreover, check many text fomat (ex. between line 461 and 487)

2.In Materials section, we would like to see a global diagram showing the proposed work. This helps to simplify and improve the structure of the text.

3.We suggest adding the link of the dataset for giving an opportunity to the scientific community to access the data easily.

4. What about the cross-validation of improved Yolov7 for the evaluation method?

5.The authors are invited to justify why they are used yolov7 not for example the recent version yolov8 or other versions.

These are my observations and I hope they will help improve this research paper.

Regards

Author Response

Please see the attachment.

Author Response File: Author Response.docx