The Surface Defects Detection of Citrus on Trees Based on a Support Vector Machine
Round 1
Reviewer 1 Report
This paper presents a citrus surface defect detection scheme that combines machine learning with image processing in a nighttime environment. This method has good accuracy and real-time performance.
There are a lot of issues that must be resolved before publication can be considered. If the following problems are well solved, it's believed that this paper is more perfect.
1. It is suggested to improve the resolution of the picture, such as Figure 7, 8 and 9;
2. The description of Figure 11 in Section 4.2 is not clear enough, so it is recommended to describe it in detail.
3. If there are two targets in a green rectangle in Figure 14 in Section 5.2 of the article, it is suggested to indicate whether this situation belongs to the correct detection state.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 2 Report
The manuscript investigates the use of ML and image processing to identify and detect defects in mature citrus fruit at night time. The results show that the algorithm had good accuracy and real-time performance in recognition and defect detection in citrus in a natural environment at night. The manuscript is within the theme of the journal and interesting for potential readers, and thus it could be considered for publication if the following concerns can be solved properly:
- In general, the manuscript focuses on the experimental setup and the classification accuracy. However, to be in line with the scope of the Agronomy journal and attract wider readers, the significance/impact and practical applications of this study should be discussed, perhaps in the introduction and discussion section.
- leave a space between texts and citations.
- Is the dot before or after the citation [21]?
- p.6: remove the comma before “For the H component …”.
- in the heading 3.3, is it “kernel”?
- p.7: this sentence “The toolkit had … kernel functions” should be re-written.
- p.7: “the accuracy of models with different kernel functions was 99.67%”, do you mean different kernel functions produced the same accuracy? Or RBF generated the best?
- p.7: section 3.3, please explain a bit clearer the role of the SVM kernel function and what the predicted accuracy of 99.67% is for, and how it will be linked to the next steps.
- section 5.1: the average running time of the algorithm is reported based on what computing platform (e.g., computer configuration, programming languages, software)?
- section 6: How the numbers of the accuracy and average running time reported here are different from those in the results section.
- Would the leaf cover be an issue?
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 3 Report
Manuscript “The Surface Defects Detection of Citrus on Trees Based on a Support Vector Machine”
This manuscript will be interesting for scientists, who are interested in studying the problem of the detection and classification of citrus defects under UV light. Also, these scientific results will have practical application. In order to improve the manuscript, I suggest the following corrections:
1. The paper states that as the largest citrus producing country, China accounts for more than 20% of the world citrus production (p. 1). The sources of this information referred by the authors are dated 2014 and 2016. It is necessary to provide references to current sources of statistical information.
2. The paper states that Nidhi et al. [14] used the fuzzy rulebased classification system (FRBCS) to predict the maturity of unripe tomatoes (p. 2). But, in the references, the paper with other authors is listed under number 14.
[14] Goel, N.; Sehgal, P. Fuzzy classification of pre-harvest tomatoes for ripeness estimation - An approach based on automatic rule learning using decision tree. Applied Soft Computing, 2015, 36, 45-56.
3. In the paper, the accuracy is indicated as a percentage, so Equation (1) should be written as follows
???????? = ??????·100/(?????? + ???? + ?????)
The paper states that Formula (1) was used to calculate the accuracy of citrus region detection: 407 correct, 1 wrong and 21 missing detection rectangles were finally counted. Therefore, the detection accuracy of citrus regions in the 39 citrus UV images was 95.10%.
Let's substitute the value of citrus region detection in the Equation (1)
???????? = 407·100/(407 + 21 + 1) = 94,87%.
Thus, the detection accuracy of citrus regions does not match the one stated in the paper.
4. In the paper, the accuracy, precision and recall are indicated as a percentage, so Equations (2), (3), (4) must be written as follows
???????? = (?? + ??)·100/(?? + ?? + ?? + ??)
Precision = TP·100/(TP+FP)
?????? = ??·100/(?? + ??)
5. In Table 1, a false positive (FP) result is marked FR.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Thank you for addressing my comments. I suggest a minor revision to include the response 11 into the Discussion and conclusion section.
Author Response
Please see the attachment
Author Response File: Author Response.docx