A Deep Learning-Based Model for Tree Species Identification Using Pollen Grain Images

Round 1

Reviewer 1 Report

In this paper, a deep learning based pollen grain image classification method is proposed. AlexNet and GoogleNet is leveraged to classify pollen grain of different trees. However, following questions should be considered:

1. Why pollen grains is used to identify tree species rather than leaves? As the images of leaves are easier to be acquired. 

2. Why AlexNet and GoogleNet are seclected as many other more powerful networks are proposed such as ResNet, DenseNet, etc. 

3. There exists a huge mistake in this paper that the author use same data for training and testing, which means the experimental reuslt can not effectively evaluate the performance of  proposed method. 

4. The best performance in each table can be folded to make the result more readable. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear Author

          1. kindly remove the numeric values from the abstract section.

          2. More explanation about the simulation should be added.

          3. more references should be included

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper discusses deep Learning-based model for tree species identification using pollen grain images. The authors have compared their work with existing state-of-the-art techniques and present significant results in comparison. The work performed is highly appreciated towards the power of deep learning framework and algorithms concerned. The manuscript is well written in a scientific guide manner and a lot of useful experiments have been performed for target approaches and their comparison and analysis.

However, the author(s) are advised to make the following minor changes to be considered for publication.

 

1.      In the abstract it would be more attractive to add some future research directions.

2.      Line 53: update as "using supervised machine learning methods" for reading fluency

3.      Line 185: Matthews Correlation Coefficient (MCC) has already been abbreviated in line 18.

4.      Line 235-237: The authors mentioned that GoogLeNet outperforms others. It would be worth mentioning that why the GoogLeNet achieves higher classification accuracy... this should be justified in discussion section.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Authors made an attempt to propose a DL model Tree Species Identification using Pollen Grain Images, however, the major flaws are reported as follows.

1. Innovation and Novelty of the work is very minimal

2. Authors utilized the existing pre-trained model such as AlexNet and GoogleNet to identify the species.

3. Many recent research work is not cited in the manuscript.

4. Methodology part is very weak and not explained properly

5. How did you deal the imbalanced dataset? In normally, for the imbalanced dataset, unweighted accuracy and unweighted F1 score will be the more suitable metrics for the evaluation. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The author has answered all my questions.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Authors revised the paper as per the reviewer comments. However, in the conclusion section, they have mentioned that few other pre-trained models, like ResNet and DenseNet, hyperparameter fine tuning is difficult. Justification is required.

Also the time complexity of the proposed model can be analysed.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf