Using hyperspectral imaging and machine learning to identify food contaminated compostable and recyclable plastics

Review
General Impression:
The manuscript presents promising machine learning-based solutions to address challenges in identifying and sorting food-contaminated plastics. Overall, the manuscript is well-written and organized; however, there are areas where improvements could be made.

Comments for the introduction section:
In the introduction section, the manuscript does not clearly articulate the research gap being addressed.
The stated gap, "the identification of compostable and recyclable plastics with varying types and levels of food contamination" (lines 67–68), appears similar to those reported in your previous studies (Taneepanichskul et al., 2023; Taneepanichskul et al., 2024). It would be helpful to elaborate on how the current study builds upon and distinguishes itself from these earlier works.

Furthermore, the impact of this manuscript is not clearly articulated. In your previous publications you mentioned that the accuracy of the system decreased when detecting plastics that were dark, thin, small, or had high levels of contamination. Though the models in your current performed better for dark and small plastics, the novelty and impact of the manuscript is not clearly stated.

Conclude the introduction with a brief overview of the subsequent sections to provide readers with a roadmap of the manuscript's structure.

Comments for the methodology section:
While the methodology is well written, it would benefit from a brief discussion of advantages and limitations of the algorithms and the HSI system you use here, supported by relevant literature.

The speed of the conveyor belt during experiments is not described. While you reference earlier works for experimental setup details, neither Taneepanichskul et al. (2023) nor Taneepanichskul et al. (2024) provides this information. Conveyor belt speed is a critical factor influencing model efficiency and real-world applicability. Low speeds may facilitate higher identification accuracy, but this does not necessarily translate to practical systems where high throughput is essential. Including details about the conveyor belt speed and its implications would enhance the practical relevance of your findings.

Comments for the results section:
Between lines 303 and 304, include a brief overview to frame the reader for the results subsections.


Additional Comments:
Line 66: Question marks are not needed.
Line 74: Please place a comma after packaging.

Line 120 to 122: Is there a specific reason why your training dataset was limited to pristine plastics and plastics with low levels of contamination (25%) while using supervised algorithms?

Line 332 to 333: Is there a specific reason why the calibration dataset did not include plastics with over 25% of contamination? I would assume that you’d want your training model to represent real world data, and that includes highly contaminated plastics.

Line 501: Can you comment on why PLS-DA has a higher accuracy at high level of contamination compared to medium level of contamination?

Note:
This review refers to round 1 of peer review.

Review
Thank you for the opportunity to review your manuscript. Your work provides a comprehensive evaluation of the performance of hyperspectral imaging (HSI) combined with near-infrared (NIR) technology for detecting food-contaminated compostable and recyclable plastics. Below are suggestions to further enhance the quality and clarity of your paper:

1) Please ensure consistency in the correct use of terms like “near-infrared (NIR)” instead of "Near Infrared (NIR)" and “hyperspectral imaging (HSI)” throughout the manuscript.

2) In the last paragraph, “In this paper we present work developing new??? Chemometric … contamination” is unclear. Please revise for clarity.

3) The introduction could do with a small section on the importance of integrating near-infrared (NIR) spectrum information in detecting plastic waste, which is particularly important for enhancing accuracy in (real-world) scenarios where traditional RGB-based methods struggle, such as in distinguishing plastic materials with varying transparency, color, or weathered conditions. Some of the recent work on this includes:

i) Tamin, O., Moung, E.G., Dargham, J.A., Yahya, F. and Omatu, S., 2023. A review of hyperspectral imaging-based plastic waste detection state-of-the-arts. Int. J. Electr. Comput. Eng.(IJECE), 13, pp.3407-3419.

ii) Tamin, O., Moung, E.G., Dargham, J.A., Yahya, F., Farzamnia, A., Sia, F., Naim, N.F.M. and Angeline, L., 2023. On-Shore Plastic Waste Detection with YOLOv5 and RGB-Near-Infrared Fusion: A State-of-the-Art Solution for Accurate and Efficient Environmental Monitoring. Big Data and Cognitive Computing, 7(2), p.103.

iii) Tamin, O., Moung, E.G., Dargham, J.A., Yahya, F., Omatu, S. and Angeline, L., 2022, November. Machine learning for plastic waste detection: State-of-the-art, challenges, and solutions. In 2022 International Conference on Communications, Information, Electronic and Energy Systems (CIEES) (pp. 1-6). IEEE.

iv) Tamin, O., Moung, E.G., Dargham, J.A., Yahya, F., Omatu, S. and Angeline, L., 2022, September. A comparison of RGB and RGNIR color spaces for plastic waste detection using the YOLOv5 architecture. In 2022 IEEE International Conference on Artificial Intelligence in Engineering and Technology (IICAIET) (pp. 1-6). IEEE.


4) In subsection 2.1, you mention using HEINZ tomato ketchup for both training and validation, but only HEINZ mayonnaise for validation. Please explain why mayonnaise data wasn’t included in the training.

5) In subsection 2.2, consider adding a table to clarify the distribution of training, validation, and testing datasets for clarity.

6) In subsection 2.4, you may also consider evaluating model performance using additional metrics like precision, recall, and mean average precision (mAP) for a more comprehensive analysis.

7) Please fix the caption in Table 3 to use “The performance…” instead of “the performance…”.

8) I suggest splitting Section 4 Discussion and Conclusion into two separate sections. In the Conclusion, be sure to address both the study’s limitations and potential future research directions.

9) Subsection 4.3, “Application of HSI in anaerobic digestion, in-vessel composting, and recycling plant for detecting food-contaminated compostable plastics,” seems irrelevant. Please clarify its relevance to your study or consider removing it if it does not directly support the main research focus.

I look forward for the revised manuscript.

Note:
This review refers to round 1 of peer review.