Real-Time Lightweight Detection of Lychee Diseases with Enhanced YOLOv7 and Edge Computing

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents an algorithm for detecting lychee diseases using edge computing and deep learning. The topic itself, covering lychees, is already quite interesting since there are very few works on the specific crop. Technically, the work also shows good effort on improving the model and making it fit for edge computing application. However, the main question is whether there is a huge need for boosting the inference speed of the model? In the end, disease detection has to be as accurate as possible and not as fast as possible. In reality, achieving real-time detection might not be exactly useful unless a real-time monitoring device (such as a wireless sensor network) is employed. All in all, it is a good effort but the objective might be an issue. Here are some items that need some attention:

Related work: This section was too verbose. Better reduce the text.

Figure 1: I believe this is unnecessary. Please just describe this via text.

3.1: Better if there is also a matching image here that shows how the images were acquired. 

Table 3: Based on this table, it appears that YOLOv7+M was the best model in terms of performance. In disease detection, normally the most accurate model is more recommended.  Due to the results in this table, it is more recommended to use YOLOv7+M instead even with the slightly poorer FPS. 

Figure 12: Did the authors also test other cameras for testing? Why was the specific camera used in this work?

Figure 13: It will also be interesting if the authors can provide a supplementary video of the detection results. In that way, readers can see any potential jittering of the detection results (which hopefully was gotten rid of by having a robust model). 

Conclusion: The mosaic data augmentation strategy is normally not a big novelty anymore as it is a common strategy. I believe the authors should focus more on how the real-time detection can be really used in practice and why it should be that way. 

Comments on the Quality of English Language

There are still quite many typographical error and slightly awkward expressions. Best if an English editor can also run through the text. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Your manuscript offers a significant contribution to the field of lychee disease detection through the implementation of the YOLOv7-MGPC-based algorithm, a topic with considerable real-world implications. However, the paper exhibits several types of errors, including redundancy, grammatical, and verb tense inconsistencies, and occasional unclear sentence structure. For instance, the verb tense in 'we collect datasets of lychee diseases' should be consistent with the rest of the abstract 'we collected datasets of lychee diseases', which employs the past tense. Also, the phrase 'which firmly demonstrate the effectiveness' should be corrected for subject-verb agreement to 'which firmly demonstrates the effectiveness' and so on. 

As well as I suggest conducting a comparative analysis of your YOLOv7 model against other leading object detection algorithms, aiming for a more comprehensive performance assessment. Also, considering emerging evidence showcasing YOLOv8's enhanced detection capabilities, providing a rationale for retaining YOLOv7 would add valuable depth to your study. Further, the inclusion of details on hyperparameter tuning would yield a fuller understanding of your model's optimization process. Finally, consistent with your emphasis on computational efficiency for edge deployments, I propose that you explore Feature Pyramid Networks (FPN) or DropBlock, both renowned for their resource efficiency.

Comments on the Quality of English Language

 

The quality of english lanage of paper needs to be improved. The paper presents several typo and gramatical errors. I strongly recommend conducting a meticulous review of the manuscript to correct grammatical errors and improve the paper's overall coherence and readability.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf