Page 194 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 6 Plant leaf disease classification based on feature selection 183

Figure 6.24 Binary feature selection process.

The fitness function is defined as:

Maximize E
where E is the accuracy of the model with regard to the test set.
The model considered is a multilayer perceptron with the
following architecture: (Fig. 6.25)
• Input layer: number of features
• Hidden layer 1: 30 neurons with ReLU activation function
• Hidden layer 2: 20 neurons with ReLU activation function
• Hidden layer 3: 10 neurons with ReLU activation function
• Output layer: 4 neurons with Softmax activation function, cor-
responding to 4 classes
ReLU function is used in the hidden layers due to its special
characteristics. It does not activate all the neurons at the same
time, only when the input is positive. Due to this, ReLU is more
computationally efficient in comparison with sigmoid or tanh
function. The softmax function is chosen for the output layer
due to the multiclass classification problem. The loss function
is categorical cross-entropy.

Figure 6.25 Proposed MLP architecture. MLP, multilayer perceptron.