Research Description
Pomegranate fruit offers numerous health benefits. On the other hand, various diseases infect
them. Deep Learning (DL) techniques have been applied to classify these diseases. Eleven
Pretrained Deep Convolution Neural Networks (PDCNN) have been used. The object of this
study is to determine which PDCNN has achieved the highest accuracy in classifying various
pomegranate diseases. Five pomegranate diseases (Anthracnose, Cercospora, Bacterial
Blight, Alternaria, and health) have been classified. A dataset including 4919 fruit images has
been used. The dataset is imbalanced; Hence, data-augmentation technique has been applied
to balance it. the updated balanced dataset has been used which includes 7070 images (1414
for each category).
The trained PDCNNs were really tested by 180 images. Eight PDCNNs have achieved
accuracy of 100% for healthy fruit. DenseNet201 has achieved the highest accuracy of 97.78%
in Cercospora diseases. EfficientNetB0 has achieved an accuracy of 100% in Bacterial
Blights. InceptionResNetV2 achieved the best accuracy of 97.78% in Anthracnose. AlexNet
has achieved 96.11% in Alternaria. In comparison to healthy or infected, the eight PDCNNs
have achieved an accuracy of 100%
