Page 197 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

186 Chapter 6 Plant leaf disease classification based on feature selection

Table 6.10 Confusion matrix of MLP.

Class C1 C2 C3 C4
C1 34 0 0 1
C2 7 13 0 0
C3 2 0 18 2
C4 3 0 1 4
MLP, multilayer perceptron.

Table 6.11 Performance comparison of CNN models.

Without transfer learning With transfer learning
Training Validation Training Validation
Models accuracy accuracy accuracy accuracy
AlexNet 74.3% 66% 85.6% 78.8%
VGG16 78% 76% 84.5% 77.6%
ResNet-18 84% 68% 90.6% 80%
ResNet-50 63.5% 67.1% 85.4% 84.7%
CNN, convolutional neural network.

of images before processing by the deep neural network. Trans-
ferlearningfromother similarfeaturesdataset is also per-
formed to train the deep residual neural network, which gives
advantages in the learning process. The proposed method
achieves a best accuracy of 84.7%, which is higher than that of
other pretrained models. Also, we proposed a simpler MLP
framework but has competitive results. In future work, this pre-
sented method will be improved by increasing the number of
images in data set and optimizing parameters of the deep neural
network model.