Proposed Smart Monitoring System for the Detection of Bee Swarming
Round 1
Reviewer 1 Report
> several pre-trained networks of the authors’ proposed CNN Swarm-engine are also tested to detect swarming events
To a practical beekeeper, the above statement in the Abstract suggests that the device was actually tested in the field. This is misleading.
The first citation that I checked was the one for the following claim:
>Similarly, high humidity levels can lead to swarming events inside the beehive, excessive colony honey consumption, and the production of propolis and wax by the bees as a countermeasure [14,19].
I reviewed citations 14 and 19 (one was a self citation); neither one mentioned excessive honey consumption, nor the production of propolis or wax as countermeasures, so the citations were not supportive of the claim, and one was a simple self citation.Re: English translation, the word "swarm" appears to have been misused to indicate "cluster."
Author Response
Dear reviewer,
On behalf of my colleagues, I submit to you our responses to your comments. I hope that we have fulfilled your expectations. We once again thank you for your effort.
Comments and Responses list.
C1: To a practical beekeeper, the above statement in the Abstract suggests that the device was actually tested in the field. This is misleading.
R1: Appropriate figures of the tested prototype devices tested in the field are added in Appendix C. The illustrated devices have been tested in our laboratory's testing apiary station (Ligopsa, Ioannina, Greece). Nevertheless, "tested" has been modified to: validated to detect clustering events that may lead to swarming
C2: The first citation that I checked was the one for the following claim:
>Similarly, high humidity levels can lead to swarming events inside the beehive, excessive colony honey consumption, and the production of propolis and wax by the bees as a countermeasure [14,19]. : I reviewed citations 14 and 19 (one was a self citation); neither one mentioned excessive honey consumption, nor the production of propolis or wax as countermeasures, so the citations were not supportive of the claim, and one was a simple self citation.
R2: You are correct. Appropriate amendments have been made. Citations [14,19] have been replaced with proper sources mentioning how high humidity levels may lead to swarming or CCD (book references of traditional apiary practices as well as beekeepers' observations). Paragraphs at lines 30 and 43 have been re-written. Reference 19 has been placed at the line 30 paragraph (an RF temperature system with GPS capabilities), while reference 14 has been placed at the line 54 paragraph (as provided by the analysis at [24])
C3: Re: English translation, the word "swarm" appears to have been misused to indicate "cluster."
R3: Appropriate paragraph has been added in the Introduction elaborating swarming and how swarming correlates with bee clustering (line 29).
Reviewer 2 Report
This manuscript is dealing with bee colony swarming and application of machine vision technologies for its detection. Paper is actual as research of bee colonies is still very actual. Reviewer have several comments and suggestions how to improve the quality of this manuscript.
Introduction lacks the reference/s to the swarming definition (rows: 22-24). As well in the introduction authors should mention why it is important to detect the swarming event as there are some articles stating that this process is natural and should not be managed. What are the benefits for the beekeeper when swarming is detected?
In addition to the description of camera module, real image of mounted camera inside the hive should be valuable.
Figure 1 and FIgure 2 should be improved. Black text on the dark blue arrows is hard to read, arrows are of different types, visualisation of modules are not successfull (grey sqaures). Please consider to change the images.
Describing the Lora distances (Row 178) authors states that "The coverage distance of the version 2 concentrator also varies since they can cover distances up to 12-18Km for LOS setups and 1-5Km for non-LOS ones" – Is it tested by the experiments or just theoretical information. If theory, then references should be provided.
Row 207: Authors should demonstate the Bee-RMS mobile phone application and web dashboard, some screenshots how it looks like.
Rows 242-246: Different classes are defined (limited number, small number etc) it would be good to define the exact number of bees for each defined class.
Figure 3 must be improved. Bilateteral Filtering – it seems should be Bilateral filtering. Arrow from STEP 1 to Motion detected is missing as there is only back arrow from Motion Detected to STEP 1. Line from Motion Detected is not straight.
Rows 308-324: based on XML schema it seems that varroa mites and bee queen can also be detected, not only the ordinary bees. Reviewer missed within the section 3.2 description of dataset for Varroa and bee queen, there was only description on how the bee images are taken and trained. Should be some details on how the bee queen is detected as well on how the varroa mites are detected.
Tables 6,7,8 could be combined into one table, just put three numbers in each cell for each hardware configuration. Then it would be easier to compare the results of the experiments.
Regarding the accuracy of the models, authors used several metrics to evaluate the accuracy, but it was not clear if authors checked the model outcome with the manual observations of the images. For example model says that there are 10 bees on the image and authors should verify it by the human counting.
One critical issue that should be mentioned by the reviewer is that the manuscript title is about bee swarming detection, but after reading the manuscript it was not clear at all how the swarming event is related to the number of bees on the collected images. Was the swarming event practically observed or not and how it is connected to the number of observed bees on images. As for now reviewer is a little bit sceptical about such relation.
Additional minor comments:
All acronyms should be explained the first time they appears in the text, like LOS, BW, MFCC and others.
Row 253: LabelImg should not be mentioned once again as other tool, as already stated that it is used by the authors.
Row 292: “Figure 2 analytically illustrates” – it seems there should be Figure 3 not 2.
Row 333: Letter “I” is missing in the beginning of the sentence.
Author Response
Dear reviewer,
On behalf of my colleagues, I submit to you our responses to your comments. I hope that we have fulfilled your expectations. We once again thank you for your effort.
Comments and Responses list.
C1: Introduction lacks the reference/s to the swarming definition (rows: 22-24). As well in the introduction authors should mention why it is important to detect the swarming event as there are some articles stating that this process is natural and should not be managed. What are the benefits for the beekeeper when swarming is detected?
R1: An appropriate paragraph has been added in the Introduction stating what swarming is and why it is vital to detect bee clusterings at the beehive's end frame or the beehive space between the brood box and the lid; such clusterings are strong indications of swarming. The beekeepers' actions towards swarming are also briefly mentioned (added paragraph in the Introduction – third paragraph)
C2: In addition to the description of camera module, real image of mounted camera inside the hive should be valuable.
R2: Camera module photo of version 1 and version 2 devices that have been used for collecting the images and testing has been added in Appendix (C).
C3: Figure 1 and FIgure 2 should be improved. Black text on the dark blue arrows is hard to read, arrows are of different types, visualisation of modules are not successfull (grey sqaures). Please consider to change the images.
R3: Figures 1 and 2 have been improved by removing the black text from the dark blue arrows. Arrows have been set to similar types.
C4: Describing the Lora distances (Row 178) authors states that "The coverage distance of the version 2 concentrator also varies since they can cover distances up to 12-18Km for LOS setups and 1-5Km for non-LOS ones" – Is it tested by the experiments or just theoretical information. If theory, then references should be provided.
R4: Appropriate citation has been placed at row 178
C5: Row 207: Authors should demonstate the Bee-RMS mobile phone application and web dashboard, some screenshots how it looks like.
R5: Bee RMS mobile phone application and web bee counting interface have been added in Appendix (B)
C6: Rows 242-246: Different classes are defined (limited number, small number etc) it would be good to define the exact number of bees for each defined class.
R6: The per class number of bees detected for our experimentation has been set at rows 242-246
C7: Figure 3 must be improved. Bilateteral Filtering – it seems should be Bilateral filtering. Arrow from STEP 1 to Motion detected is missing as there is only back arrow from Motion Detected to STEP 1. Line from Motion Detected is not straight.
R7: Figure 3 has been corrected and improved.
C8: Rows 308-324: based on XML schema it seems that varroa mites and bee queen can also be detected, not only the ordinary bees. Reviewer missed within the section 3.2 description of dataset for Varroa and bee queen, there was only description on how the bee images are taken and trained. Should be some details on how the bee queen is detected as well on how the varroa mites are detected.
R8: Varroa mite and bee queen detection are under validation for the time being. Varroa detection is performed manually using color masks on the detected bees and queen detection class is still under training (the existing train is of low accuracy due to the small queen dataset obtained so far). Relevant details have been added below the XML schema in a new paragraph.
C9: Tables 6,7,8 could be combined into one table, just put three numbers in each cell for each hardware configuration. Then it would be easier to compare the results of the experiments.
R9: Tables 6,7,8 have been combined into one Table 6.
C10: Regarding the accuracy of the models, authors used several metrics to evaluate the accuracy, but it was not clear if authors checked the model outcome with the manual observations of the images. For example model says that there are 10 bees on the image and authors should verify it by the human counting.
R10: The authors used the MDA metric (Equation 1) to do what you have described. Their findings are shown in Table 7 (second column). Appropriate elaboration of the metric and its use has been put in the paragraph above Equation 1.
C11: One critical issue that should be mentioned by the reviewer is that the manuscript title is about bee swarming detection, but after reading the manuscript it was not clear at all how the swarming event is related to the number of bees on the collected images. Was the swarming event practically observed or not and how it is connected to the number of observed bees on images. As for now reviewer is a little bit sceptical about such relation.
R11: Swarming is related to the number of bees concentrated at the outer frame or in the lid. When the bee population presents such behavior, it is either stressed due mainly to: 1 ) the existence of a second (younger) queen that is ready to take part of the population and leave, 2) lack of food in the area and the colony is ready to migrate, 3) The queen is lost, and the colony is ready to leave since it is unable to create another queen because of royal jelly shortage (produced by the queen only) or 4) overpopulation limits have been reached and new frames need to be added to the hive (or even floor). Additionally, diseases such as varroa mite cause such behavior (which is the most severe disease in the EU apiaries over the last decade, as mentioned apiarists. Appropriate paragraphs have been added in the Introduction (line 25 and line 29).
These actions can be closely related to the bee counting of targeted beehive areas internally (not at the entrance as shown in the bibliography). The areas of interest (of bees' clustering – are the lid and the end frame).
The above statements have also been added to the Introduction to strengthen our claim of bee counting, as you have correctly indicated.
Additional minor comments:
C1: All acronyms should be explained the first time they appears in the text, like LOS, BW, MFCC and others.
R1: All acronyms have been explained the first time of appearance.
C2: Row 253: LabelImg should not be mentioned once again as other tool, as already stated that it is used by the authors.
R3: duplicate LabelImg has been removed
C3: Row 292: "Figure 2 analytically illustrates" – it seems there should be Figure 3 not 2.
R3: Fixed reference
C4: Row 333: Letter "I" is missing in the beginning of the sentence.
R4: Fixed missing letter
Round 2
Reviewer 2 Report
Thank you for taking into account my suggestions and comments. Now manuscript is improved well, but i still have one important question and one minor question.
Was the system practically used to detect swarming event to validate the alghoritm and outcome?
And one more technical question, if the camera is placed inside the hive how it is protected from vax. As i know bees are trying to cover everything by vax in the hive.
Author Response
Dear reviewer,
Thank you again for arising such targeted questions and considerations.
Regarding our proposed system's practical validation, we have added a new experimental scenario section IV (System validation towards swarming). Our validation test included successful validation for the case 1 scenario and preliminary validation for the case 2 scenario that needs further testing.
Regarding your technical question, there is nothing someone can do to confront bees' propolis or wax deployment over the camera lenses. We have destroyed quite a few cameras during this dataset acquisition. and we come up with some useful tips:
T1. Maintain a safe distance of two or three frames between the camera module and the bee frames. If this is not possible a second floor can be installed with the addition of frames for the queen to move and lay her eggs and placement of the camera on the opposite side of the box.
T2. The camera module should be installed on a separate frame facing above or in the middle maintaining the T1 distance. This shall allow the easy removal/placement of the camera.
T3. The use of 160-degree wide-angle lenses reduces the propolis deposition and makes it easier for the camera lens to clean
T4. periodic lens cleaning (once every two weeks or if high humidity levels are detected) will also keep the camera lens clean. This can also be detected if blur frames are received.
Best regards
Sotirios Kontogiannis
