Page 77 - IJEEE-2023-Vol19-ISSUE-1
P. 77
Abdulla & Marhoon | 73
• Training and Testing Phase the accuracy and loss function curve during the training
process.
After dividing the dataset, used 12,771 images in the
training and validation process, where two models were used Table II
in extracting features from the images. The first model ACCURACY FOR TRAINING AND VALIDATION MODELS.
proposed for CNN consists of three main layers: the first is
the input layer, the second is the hidden layers (feature Proposed method Training Validation
extraction), and the third is the classification layer (output). accuracy accuracy
Fig.7 shows the structure of the proposed model used in the MobileNet_v2 99.89%
training dataset. Proposed method 98.15% 96.7%
The second model is pre-trained, MobileNet_v2, which used 97.41%
the transfer learning method of the model to train the dataset
by taking the network weights. And fine-tune the input layer Tested the two trained models on the test dataset of 2267
with a dataset for tomato plants and the output layer with an images and used metrics confusion matrix, precision, recall,
output layer for seven classes of plant diseases. f1-score, and accuracy in measuring the validity of the results
As for the parameters used in training the two networks, we obtained. Table III shows the values obtained during the test
used an optimizer of type Adam, categorical_crossentropy of the two models.
loss function, batch size=64, epochs=50, and accuracy
metric. Table II shows the training accuracy and validation
values obtained from training the two models. Fig.8 shows
Fig. 7: The structure of the proposed model.