Design of Deep Learning Architecture for Classification of Orchid Diseases

Orchid disease is a huge obstacle in growing orchids. In general, visual inspection methods have been provided but they generally waste a lot of analyzing time and can cause a mistake in the identification of the diseases. Therefore, this research focuses on constructing a system to inspect and classify orchid diseases by using deep learning. In this paper, we designed a model based on deep learning architecture and compared it with pre-trained models: ResNet-50, VGG-16, and VGG-19. Pictures of orchid diseases, used as datasets for adjusting parameters and training the model, were collected from an orchid farm in Nonthaburi province and diagnosed by 3 orchid experts. The Keras library was employed for deep learning models. As a result, our proposed model was less complicated and required less time to process, compared with the pre-trained models. Although our model performed high accuracy of approximately 96.2 %, and the pre-trained model the highest accuracy of approximately 97.4 %, our model had 27 times fewer parameters than the pre-trained models, which had the advantage of reducing process time.