Explainable Deep Learning Study for Leaf Disease Classification
Round 1
Reviewer 1 Report
The paper focuses on the interpretability of deep learning models in different agricultural classification tasks based on the fruit leaves dataset and it explores (based on dataset arranged to three experiments with different categories) whether the classification model is more inclined to extract the appearance characteristics of leaves, or the texture characteristics of leaf lesions during the feature extraction process.
The fruit leaf images are taken from Plant Village dataset and expanded with other different kinds of fruit leaf images, which were downloaded from Google, AI Challenge 2018. Therefore, the used data set is a hierarchical multi-label classification data set, and the overall flow of the paper`s experiments is given in Figure 1. The first-level label is the type of fruits. The type of fruits is judged by identifying the shape of the leaves. There are 11 types of fruits in total. The secondary label is the type of pests and diseases. The types of pests and diseases. The types of pests and diseases. corresponding to each fruit are different. There are 34 categories of fruit leaf data sets in total; each category contains about 1,000 pictures; the total number of pictures in the data set is 34,000.
Three classification experiments were carried out in this study. Each classification experiment used three frameworks based on VGG, GoogLeNet, ResNet, and two extended models of ResNet34-CBAM and ResNet50-CBAM (ResNet-attention models) for visual display. The attention-based ResNet framework was constructed by introducing CBAM into the ResNet network framework, having the structure given in Figure 3.
Experiment I - a multi-classification experiment of fruit type and combination of diseases and insect pests. The results show that the ResNet34-CBAM model recognizes the shape features of the leaves and the texture features of the diseased spots at the same time to achieve a better classification effect (Table 3 and Figure 4).
Experiment II - a binary-classification experiment of fruit disease or not. GradCAM method shows the best results. The GradCAM method can help tounderstand the model prediction mechanism (Table 4 and Figure 5).
Experiment III - a multi-classification based on fruit types classification. The GradCAM method offers the best results (Table 5 and Figure 6).
Therefore, the results show that the ResNet model has the highest accuracy rate in the three experiments.
Strengths
- The used research methodology
Weakness
- The work needs to be restructured. Discussions must be introduced before the conclusions and the conclusions must include the entire content of the paper presented.
Author Response
Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Explainable deep learning study for leaf disease classification”.
We have made corresponding amendments to the comments of reviewer 1. The the conclusion part is adjusted to the end of the article and highlighted in green. Please see the attachment.
Thank you for your comments and guidance. The quality of the article has been improved through modification. We hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Acceptable
Author Response
Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “Explainable deep learning study for leaf disease classification”.
We have made corresponding amendments to the comments of reviewers. The the conclusion part is adjusted to the end of the article and highlighted in green. Please see the attachment.
Thank you for your comments and guidance. The quality of the article has been improved through modification. We hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.
Author Response File: Author Response.docx
