A Maturity Estimation of Bell Pepper (Capsicum annuum L.) by Artificial Vision System for Quality Control

Round 1

Reviewer 1 Report

The paper deals with quite interesting topic that could be potentially quite useful in practice. Although the proposed approach  is more or less standard, the proposed paper also has some original ideas, therefore I think that could be published in Applied Science Journal. This is really applied science but for real application the whole procedure will need more improvements, for example image acquisition methodology is not suitable for real practical use (five images per fruit in different position), but anyway that is technical problem. The paper present the idea not the final solution.

Before publishing, I suggest revision, because I have some remarks concerning paper content and its presentation, particularly in experiment description:

• There is some disagreement how many different samples are used in experiment. (Figure 1 and text in 2.1 Samples). On Figure there are 7 Class 2 samples i text nine. Sum of all samples mentioned in 2.1 is not 50.
• Figure 5. is not well explained. Authors mention RGB channels and Brix grades, but here only Red and Green Channel were used.
• Why images are moved from RGB to HSV colour space in Section 2.5 (I understand but it has to be explained)? Why particularly these thresholds were used (text and Table 1)? Why 400 pixels are used for region elimination?
• Section 2.6.1 – There are two Table 2. It is not clear why particularly this ANN models were used for maturity class estimation and Brix content estimation. Why linear output neurons for Brix? Is that connected with Figure 1? Figures 7 and 8 must be harmonized with labels (for example in Fig. 7 Output is description of final layer (down) and Class in the middle as output and on Fig.8. is opposite
• Section 2.6.2 Fuzzy Logic (it is wrongly marked as 2.5.2) – There is a difference in support set for various colours. Why? I suppose that support sets are pixels but it will be useful to emphasise that. Figure 14 is not entirely visible (right side). Why Takagi-Sugeno inference is used and not Mamdani? In Table 4 instead of Model is written Modelo.
• Section 3 is little bit confused. In previous sections we have all together 18 different models (10 of ANN and 8 of FL) and here in Table6 and Table 7 same Model 1,2,3,4 and 5 are mentioned. I suppose in Table 7 has to be Model 11 to 15.
• Section 4 – Discussion has to be improved and better explained. For some samples there is a great difference between samples and prediction (Figure 15). maybe better result analysis for Classes could be included. But this is connected with my final remark:
• From my point of view sample set of 50 bell peppers is not a big sampling set. It is ok for methodology demonstration, but for research that will result in real practical application more samples will be needed, so analysis of results mentioned in previous remark could be more relevant
• Finally I suggest that authors make publicly aviable a database of images used in this experiment, together with measured results Brix and classification results (ground truth), so that other researchers could test their method or even improve it. This is today standard.

Author Response

Dear Reviewer. We have listened to each of your observations. We hope the article is clearer with the modifications made.

Author Response File: Author Response.pdf

Reviewer 2 Report

Review of "A Maturity Estimation of Bell Pepper (Capsicum 2 annuum L.) by Artificial Vision System for Quality 3 Control "

 

The authors of this article propose an artificial vision system that automatically describes ripeness levels of the bell pepper, using both Fuzzy Logic (FL) and Artificial Neural Networks for classification phase. The proposed method is divided into four stages: images acquisition, segmentation of acquired image sample, regions of interest segmentation and classification, which outs the maturity stage of pepper. The introduction is followed by a section in which the methods used to carry out the different phases are precisely illustrated. In section three we can then find the results divided into various tables. in the end, we find the final discussions where is illustrated that the models with FL achieved a maximum precision of 88% in identifying the four stages of maturity corresponding to the shades of green, yellow, orange and red, while the Models with ANN have 100% precision to identify samples of green and red colour.

Based on this article we can make the following observations.

 

English must be absolutely checked. In fact, there are many grammatical errors such as nouns in the singular and verbs without the "s" ( as we can see for example inline 46: "Partially ripened fruit,..., have lower..), and also the acronym of words (such as computational vision is CV and not VC, or it is said Fuzzy Logic (FL) and not Logic Fuzzy).

 

After the introduction, it is necessary to add a section in which the state of the art and the related works are presented.

The tables must be better explained:  they are a little bit confusing.
It is also important to introduce graphs in which the accuracy trends of different models are linearly compared and are made according to the number of epochs.
Furthermore, as regards the epochs, it is seen that the nets have been towed for a maximum of 44 epochs and a minimum of 4. There is therefore too much difference of epochs between one model and another.

The figures in section 2.5 must be made smaller: it is not necessary to have such large figures, in fact, they risk being grainy.

The classification phase is done using Matlab: I suggest using a different environment, such as Python, to get more satisfying results in less time.

 

In general, however, the results obtained are good, so I suggest accepting this article with major revision.

 

Author Response

Dear Reviewer. We have listened to each of your observations. We hope the article is clearer with the modifications made.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The paper was strongly improved and it is suitable for publications. 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.