A Lightweight Deep Learning Semantic Segmentation Model for Optical-Image-Based Post-Harvest Fruit Ripeness Analysis of Sugar Apples (Annona squamosa)

Round 1

Reviewer 1 Report

Dear Authors,

I have thoroughly reviewed your paper on "ECD-DeepLabv3+: Post-Harvest Fruit Ripeness Segmentation Algorithm for Sugar Apple (Annona squamosa) In Optical Images." 

However, there are some suggestions:

1. In the title, it would be neccesary to put what is ECD.

2. In line 145, section 2.3 Data Augmention: Explain better the method used in this work and the number of images created.

3. In line 365, section Evaluation indicators: Report also metrics like precision, accuracy, recall to make a better comparision of performance.

4. In line 421, section 3.1 Experiment details:  You can not conclude that your model is not overfitting just by analyzing loss curves. It it nessary to analyze also the accuray curve. Put the accuracy curve to make a better conclusion. Also, I think overfitting ocurrs after epoch 40. With stop criteria do you consider to stop training the model?

5.  In line 467, section Qualitative analysis: To make inference, you do it with images you used to train the model or different? Specify that, becuase it is not correct to use images you used to train the model.

6. Line 607, section Conclusions: I suggest to make a better conclusion of all the work made.

Addressing these issues and suggestions will significantly enhance the overall quality of your manuscript.

No comments

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article explores the application of semantic segmentation techniques for detecting post-harvest sugar apple ripeness. The authors introduce an advanced segmentation model, ECD-DeepLabv3+, which simplifies model complexity while improving performance metrics. They create a dataset of optical images for validation and compare their model to other well-known models like U-Net, HRNet, PSPNet, and DeepLabv3+. Through quantitative and qualitative analysis, they demonstrate the effectiveness of their approach in accurately detecting different ripeness levels of sugar apples. Additionally, the study covers the use of data augmentation techniques to enhance model generalization and robustness.

Even if the application field is interesting, the proposed method is evaluated only on a self-made dataset, which is too small to be meaningful. Showing the performance achieved on different datasets could greatly improve the completeness of the work.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper, the authors describe “ECD-DeepLabv3+: Post-Harvest Fruit Ripeness Segmentation Algorithm for Sugar Apple (Annona squamosa) In Optical Images”. It can become an interesting paper for Agriculture after minor revision. Followings are my comments.

(1)   This manuscript illustrates a total of 1,600 images, 1,000 of which are sugar apples. Are the others kiwis and pineapples?

(2)   In addition, how many images are labeled for other and background, respectively?

(3)   Furthermore, after data augmentation, how many images are there of sugar apples, kiwis, and pineapples? Authors have to explain them in detail.

(4) As the journal is printed in black and white, please make the different markers for the different results in Figure 10.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors addressed all my concerns.