Contamination Detection Using a Deep Convolutional Neural Network with Safe Machine—Environment Interaction
, Muhammad Adnan Khalil 1,2, Fabrizia Auletta 1,2, Mariangela Filosa 1,2,3, Domenico Camboni 1,2, Arianna Menciassi 1,2,3 and Calogero Maria Oddo 1,2,3,*Round 1
Reviewer 1 Report
Please see the attached file report.
Comments for author File: 
Comments.pdf
Author Response
Thank you very much for revision, the reponse letter is attached.
Author Response File: Author Response.pdf
Reviewer 2 Report
Summary/Contributions: This study proposes a human-robot collaboration paradigm for food and medical packaging contamination detection using an upgraded deep convolutional neural network (CNN). The study emphasizes operational Quality Assurance Department (QAD) detection of uncleaned shipments and the limits of manual screening. CNN generates a dataset and applies augmentation algorithms for nine classes of pollutants on glass surfaces in the proposed system. A mechatronic platform featuring a camera for contamination detection and a time-of-flight sensor for safe machine-environment interaction conducts the experiment. The experiment showed that the suggested system can identify contaminations with 99.74% Mean Average Precision (mAP).
Comments/Suggestions:
1. The paper mentions that no single contamination detection methodology can be considered as an optimal solution, and that various methods must be combined depending on the application. It would be beneficial to provide more specific examples of how different contamination detection methods could be combined to achieve optimal results.
2. The paper describes the modification of the standard SSD-based YOLO architecture for better contamination detection outcomes. It would be helpful to provide more details on the specific modifications made and how they contributed to the improved performance.
3. The paper states that the proposed scheme involves the use of proximity sensors for instantaneous recognition of various products. It would be useful to provide more information on how these proximity sensors work and how they are integrated into the overall system.
4. The paper mentions the creation of a dataset for experimental analysis. It would be beneficial to provide more information on how this dataset was created and how it was used to evaluate the performance of the proposed method.
5. It would be interesting to explore the potential limitations of the proposed method, particularly in terms of its applicability to different types of products or packages, and how these limitations could be addressed in future research.
6. The authors are invited to include some recent references, especially some references related to Deep Convolutional Neural Networks.
7. For instance the authors may include the following interesting references (and others):
a. https://www.mdpi.com/2073-431X/12/8/151
b. https://www.taylorfrancis.com/chapters/edit/10.1201/9781003393030-10/learning-modeling-technique-convolution-neural-networks-online-education-fahad-alahmari-arshi-naim-hamed-alqahtani
Can be improved
Author Response
Thank you very much for revision, the reponse letter is attached.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
The authors considerd my comments and suggestions.
Can be improved.
