An advanced deep learning approach for detecting dragon fruit disease

Abstract

This area of cultivation is however faced with daunting challenges as the plants are prone to diseases that affect its health and production. To this end, we put forward an improved deep learning model for diagnosing dragon fruit disease based on CNN approach that better discriminates different categories of dragon fruit health conditions. Using a dataset comprising 2438 augmented images collected from a nearby dragon fruit nursery, annotated with four target attributes—Fresh, Brown Spot, Stem Rot, and Turning Yellow our study explores the effectiveness of five different CNN architectures: These include CNN, Xception, VGG16, DenseNet201, InceptionV3. Experimental analysis shows that DenseNet201 outperforms other networks, the best accuracy reaching 97. 23 % of all cases, whereas CNN and Xception networks detected it with accuracy of 92%. 62% and 92. 52%, respectively. However, it can be seen that VGG16 and InceptionV3 were able to produce slightly lower accuracy of about 86. 89% and 91. 80%, respectively. These findings also testify to the benefit of deep learning models especially DenseNet201 in diagnosing the diseases of dragon fruit. When applied, the proposed approach presents a viable strategy for the early detection of diseases in plants and efficient crop management to enhance productivity in the agricultural systems. More studies could focus on the ensemble and scale and transferability of deep learning approach for disease diagnostics to different agricultural environments making its utility superior and versatile in diagnosis of diseases in agriculture.