Using Internet of Things (IoT), Near-Infrared Spectroscopy (NIRS), and Hyperspectral Imaging (HSI) to Enhance Monitoring and Detection of Grain Pests in Storage and Handling Operators

Round 1

Reviewer 1 Report

Please see the attached file.

Comments for author File: Comments.pdf

Author Response

Dear Sir or Madam,

First, we would like to thank you for your carefully and constructive comments.

We have taken due note of your comment that the specific contribution of our work should be more clearly highlighted. Wherever possible, we have added clarifications to the text mentioned in the following points.

• Comment 1: « It is usually a good idea to present paper’s contribution on the first section. Please consider moving the last part of Conclusion section into Introduction. »

To clarify the objectives of our work presented in this article, a paragraph has been added at the end of the introduction, which reflects the final points of the conclusions you mention (L 98 – 104)

• Comment 2: “Authors proposed an IoT trap for monitoring the grain pests. However, a clear explanation about the main components and how it works is missing. Figure 1 is unclear. How does the tube connect to the trap? Where is the camera installed? How is data collected? What is the purposed of the tube and why insect goes inside it? »

We understand that the trap design is not clear enough, especially for readers unfamiliar with traps used in grain storage. To improve the reader's understanding, we have:

• added a paragraph briefly describing how a probe trap works (L 68-71),
• added a bibliographical reference describing probe traps [17]
• changed the photograph in figure 1a for a captioned diagram to clearly position the various elements of the connected trap.

• Comment 3: “Authors has stated that the camera just take pictures of the top part of storage bins. So what is the purpose of the tube?”

The tube is the part of the device used to trap insects. It is inserted into the grain, and the insects fall into the tube as they move, becoming trapped in the reservoir. This is a common trapping technique in stored grain. For greater clarity, this has been recalled in the introduction and a bibliographical reference has been added [17].

We assume that there could have been a misunderstanding as to where the photo was taken: it was taken above the trap tank and not above the storage bin. Where necessary, we have made this clear in the text by replacing "tank" with "trap tank".

• Comment 4: “Line 113, why limit to 50 individuals?”

This limit is set empirically, with the aim of limiting manual insect counting times, considering that counting errors become too high when the number of insects exceeds this threshold. Moreover, the trap tank has a limited diameter of 5.5 cm, making it difficult for the camera and the algorithm to detect and count too many captures in the reservoir. A link between the manual counts and the automatic counts was compromised when the number of captures was too high. Therefore we did not consider it worthwhile to determine the precise number of captures in this case and set the threshold of 50 individuals captured for the manual count.

• Comment 5: “Line 116, How total adult beetles are identified?”

Insects are automatically counted on the image using an image recognition algorithm. This point has been clarified in L 124 for greater clarity. Similarly, to avoid confusion between detection and identification, the term "identified" has been replaced by "detected" when referring to automatic counting on the photo, as the species present are not identified.

• Comment 6: “Section 2.2 is unclear. What is the link with the IoT trap? How sample are prepared? Are those sample artificially made? What is the improvement of the proposed approach over the state of the art? This section looks like a simple related work exploration.”

Three methods are explored in this article. The first concerns monitoring via IoT, described in §2.1, the other two concern detection via NIRS or HSI, described in §2.2. The reminder of our objectives in the introductory section (see comment 1) clarifies this point.

The samples are indeed artificially contaminated with insects. This point has been clarified (L 137) for greater clarity.

Compared with previous work published on the subject (detection and quantification of insects present in a grain sample with NIRS or HIS technologies), the main contribution of our work lies in the ability of the proposed calibrations to detect and quantify low-density infestations in big sample. This is made clear in the discussion section (L 286-288).

In addition, as mentioned (L 79-83), the ability of NIRS to detect insects has been demonstrated in previous studies. However, the analyses were carried out on small quantities of grain analysed individually. This study involves larger quantities of grain (from 1kg to 5kg), which could solve the problem of sampling error.

For HSI, this study is particularly innovative because, as mentioned (L 90-97), few studies dealt with insect detection in cereals. No quantification studies were found.

• Comment 7: “Is NIRS/HIS used only as an offline tool or can it be integrated into handling operation when grains are being move from tanks?”

NIRS and HSI technologies have been selected for their ability to analyse large quantities of grain, possibly on-line on moving grain, as mentioned in Discussion section (L 296-300). For greater clarity, this has been explicitly mentioned also in the introduction (L 73). However, the belt speeds used in our tests are lower than those used in silos, so the transfer of these technologies to sites will require validation of calibration quality at higher speeds. This is mentioned in the Discussion section (L 316-319).

• Comment 8: “No information/discussion is presented regarding machine learning algorithm and training”

Explanations regarding PLS regression analysis used to treat NIRS data were added to the original manuscript (L 168-171) and two methodological references (numbered 40 and 41) have been added for greater clarity.

Information concerning HSI data treatment were added to the manuscript (L 175-181). The SVM-RFE used to select the discriminative wavelengths was also explained.

• Comment 9: “The paragraph starting from line 196 confusing. There are a lot of consideration from authors that need to be referenced or supported by the data.”

All comments are supported by the data presented in Tables 1 and 2. We recognize, however, that the presentation of the results is a source of confusion, particularly in the interpretation of the contingency table (Table 2) and the sensitivity/specificity calculations. We have therefore revised the wording of this paragraph to make it easier to read (L 218-225) and we've added the totals for each row and column to Table 2 to make it easier to understand.

• Comment 10: “Figure 3 is not clear. What information does it contains? Please expand the discussion about it.”

This figure shows the results of the linear regression between the number of insects counted in the trap tank (reference) and the number of insects automatically counted on the photograph (predictor). An addition in the text clarifies this point (L 228-230).

• Comment 11: “Line 222, so what is the improvements coming from using the trap? Authors concludes that the trap should be used as detector. Please consider expanding the discussion about the importance of an early detection of pest and how this could contribute to a better understanding of infestation. After how many pests the storage tank is considered compromised?”

The action thresholds for initiating curative insect control recommended by some authors as part of a binomial approach to trapping interpretation have been added to the discussion to provide further information (L 278).

• Comment 12: “Please avoid breaking table over multiple pages”

The correction has been made

• Comment 13: “Table 3 caption should be in the same page of the table”

The correction has been made

• Comment 14: “Citation on line 85 contains a typo”

The correction has been made

• Comment 15: “To make the point stronger, authors should clearly highlight what are the advantages of the proposed tools. How early detection works? Can some techniques be integrated in handling lines during storage operation? What is considered as a bad infestation and how it can be avoided by using the proposed tools.”

Introduction and discussion have been amended where appropriate.

 

Reviewer 2 Report

1.     Which species had been identified and counted in the trapping campaign? And how to count them automatically?

Author Response

Thank you for your careful reading of our manuscripts. Below is the answer to your question: 

• Which species had been identified and counted in the trapping campaign? And how to count them automatically?

The list of species identified in the trap tanks is shown in the caption of Table 1. Species are only identified during the manual count carried out when the trap is emptied. On the other hand, on the photographs transmitted by the trap, the insects are recognized (using a previously trained image recognition algorithm) but the species are not identified. Clarifications have been made in the Methods section to avoid confusion (L 123–132).

Reviewer 3 Report

This paper investigated the monitoring and detection of grain insects throughout the storage and handling chain by using IoT, NIRS, and HIS. However interesting the subject is, the paper lacks in many aspects. The following issues need to be addressed:

 1. In abstract section, the authors should present research objective, method, and results clearly. Don’t use proper names like ARVALIS, Javelot, PSS 2121, and SPECIM FX17. IoT technology didn’t appear in the abstract section, why? It didn’t correspond to the title.

2 In introduction section, the description is not convincing. The authors should express the purpose of the paper accurately. In addition, the authors just listed some references and did not analyze them. Comparing with other references, what is your research highlights?

3 In the materials and methods section, IPM requires early detection of insects to react before infestation spread or to isolate infested lots. However, in connected monitoring part, the IoTRAP probe can detect and quantify live adult insects. How to realize the early detection of insects?

 4. In results section, how to verify the effectiveness and efficiency of the proposed method?

Author Response

Thank you for your careful reading of our manuscript. Below you'll find responses to your comments.

• Point 1: In abstract section, the authors should present research objective, method, and results clearly. Don’t use proper names like ARVALIS, Javelot, PSS 2121, and SPECIM FX17. IoT technology didn’t appear in the abstract section, why? It didn’t correspond to the title.

References to ARVALIS, Javelot and the name of the trap (IoTRAP) have been removed from the summary. It has also been clarified that the trap is connected to the internet, to make the link with IoT mentioned in the title. References to spectrometer and hyperspectral camera models and their suppliers have been removed from the summary, as they appear in the "materials and methods" section.

• Point 2: In introduction section, the description is not convincing. The authors should express the purpose of the paper accurately. In addition, the authors just listed some references and did not analyze them. Comparing with other references, what is your research highlights?

To clarify the text, a summary of the objectives of the work has been added at the end of the Introduction section (L 98-104). The contribution of the work presented in comparison with the few references on the use of HSI in cereal analysis was also mentioned (L 92-96).

• Point 3: In the materials and methods section, IPM requires early detection of insects to react before infestation spread or to isolate infested lots. However, in connected monitoring part, the IoTRAP probe can detect and quantify live adult insects. How to realize the early detection of insects?

This limitation is inherent in the probe trap technique itself (internet connected or not) which can only capture circulating forms of grain beetle pests. This does not call into question early detection when the presence in a batch of grain of adult and larval forms of a species is concomitant (the oviposition of these species is not staggered or sequential). Numerous bibliographical references, the most important of which are mentioned in the list of references (44, 45, 10, L250), have demonstrated the value of probe traps for early detection of infestations in stored grain.

• Point 4: In results section, how to verify the effectiveness and efficiency of the proposed method?

We assume that the method mentioned is NIRS or IHS analysis. As stated in the Methods section, we did not have enough infested samples to constitute a validation set (L 171). Errors were assessed by cross-validation in the case of NIRS (SECV), or more simply by comparing predictions with the number of insects introduced into the sample for HIS (SEP). This is, of course, a limitation of our work, of which we are aware, as mentioned in the Discussion section (L 312-314). We need to set up validation sets before releasing these calibrations.

Round 2

Reviewer 1 Report

All my comments have been addressed by the authors. I suggest to accept the paper in the current form.

Reviewer 3 Report

agree