Multiscale Maize Tassel Identification Based on Improved RetinaNet Model and UAV Images
Round 1
Reviewer 1 Report
In order to realize recognition and counting of maize-tassel, the authors proposed a multi-scale RetinaNet recognition method combined with improved RetinaNet model and UAV images. The authors also tested the feasibility and effectiveness of the proposed method under different image resolution, brightness, plant variety and planting density.The experimental design is reasonable and the proposed method has obtained ideal experimental results. But the authors should further summarize the paper's main contributions. Added optimization of FPNS and added CBAM attention mechanisms have been used in many networks.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report
1. In section 2.2, the division of testing and training sets generally maintains random selection, while in the experiment, the selection is based on spatial regions. What is the purpose?
2. The expression of "zhengdang958" in the abstract is inconsistent with that in line 129. It is recommended that the text be checked.
3. In Figure 7, it is recommended to layout according to remote sensing requirements and verify the entire text.
4. In line 492, although 'male ears' means the same as' tassels', try to maintain consistency in addressing them before and after.
5. Revise references according to journal requirements.
6. In the discussion section of Article 4.1, model recognition was conducted on images with different resolutions. Among them, images with 250 * 313 pixels were basically unable to recognize any maize tassels, and the reasons were analyzed.
7. How to determine different brightness levels in Section 4.2? Are you using brightness to simulate actual lighting?
8. The article utilizes the improved RetinaNet model to detect maize tassels, which has obvious advantages in detecting weak targets. In fact, there are other challenges in weak target detection in agricultural remote sensing monitoring. In the future, I hope your research will continue to deepen and solve more similar problems in agricultural remote sensing.
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Line 43-69: paragraph is too long and should be divided to facilitate comprehension. It may be interesting to reinforce that tassels identification is more important for seeds production fields.
Lines 125-134: varieties have different growth periods, but images were obtained at the same date. Was there any impact of that on results?
Lines 135-147: describes a flight at an altitude of 10m, but there seems to be done another flight at an altitude of 5m to evaluate the effect of images resolution Lines 407-411). It should not be described somewhere?
Lines 293-321: How the actual number of tassels was estimated to calculate the evaluation metrics? Number of tassels was accounted visually from the images or in the field?
Lines 509-515: it would be interesting to inform the recommended plant density for each maize variety.
Lines 518-520: Are there any estimates of leaf area index? It would help to explain results and support discussion about plant density.
Lines 136-147: The training/validation and test sets were assigned based on the field experimental design? Plots used for test seems to be on the “border” of the groups of plots. Does it have any implications for the results?
Figure 1: Explain that A1 to A5 are related to maize varieties
Author Response
Please see the attachment.
Author Response File: Author Response.docx