Page 180 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 6 Plant leaf disease classification based on feature selection 169

Figure 6.13 Training and Validation result of AlexNet.

unwanted information. This helps with the prediction accuracy
as shown in Figs. 6.12 and 6.13. However, AlexNet provides
slightly better confusion matrix results.
We can see from Fig. 6.14 and Table 6.3 that training accuracy
of ResNet-18 model is sufficient; however, the validation result is
not very good. This can be explained by either more degrees of
freedom in parameters or there is a possibility of overfitting
(Tables 6.4e6.8).

4.2 Models with transfer learning
We also perform transfer learning to fine-tune this data set.
Initially, the model was trained by the PlantVillage data set, which
consists of about 56,000 leaf images of 19 crops with 38 types of
disease. This gives an advantage that a huge amount of data is
used to train this model, which can learn features efficiently. After
having the pretrained model, we continue to train it on our data
set. In this part, AlexNet, ResNet-18, and ResNet-50 are consid-
ered (Figs. 6.15e6.19).