Abstract:
Dragon fruit is one of the highly valued economic and nutraceutical fruits produced in the
tropical regions. However, the cultivation of dragon fruit is often affected by several stem
diseases that if not controlled on time will reduce the output. Many existing disease
diagnosis techniques involve direct visit by expert practitioners which is slow, tedious and
often carries the risks of human errors. As such, the use of deep learning approaches is a
viable solution towards enabling the automatic and efficient identification of dragon fruit
stem diseases. This research focuses on the use of deep learning methods on the
classification of stem diseases of the dragon fruit in order to improve production in the
agricultural sector and thus minimize the losses. Given the set of labeled images of stems
of dragon fruit, several deep learning models were compared: VGG16, VGG19, ResNet50,
InceptionV3 and a CNN. The accuracy, precision, recall, and F1-score were used as
measurements to evaluate the models. Out of all these, InceptionV3 yielded the highest
accuracy score of 92. 36% while VGG16 achieved a score of 89. 41%. The experimental
results show that these new generations of neural networks are well equipped to identify
the stem diseases and classify them, thus playing a crucial role in disease.
