Abstract:
Dragon Fruits Stem Disease Classification is a critical task in agriculture to ensure the health
and productivity of dragon fruit crops. Utilizing deep learning techniques, this study aims to
accurately classify various stem diseases affecting dragon fruits, including Fresh, Maroon
Spot, Sun Burn, and Wilting. We collected and labeled a diverse dataset of dragon fruit stem
images from different farms and applied image processing and augmentation techniques to
enhance the dataset. Several Convolutional Neural Network (CNN) architectures were
implemented and evaluated, including Xception, VGG19, a custom CNN, InceptionV3, and
MobileNetV2. The results demonstrated that MobileNetV2 achieved the highest accuracy of
96.17%, followed by Xception with 92.17%, and InceptionV3 with 91.50%. The custom CNN
and VGG19 achieved accuracies of 85.75% and 84.83%, respectively. The high accuracy of
MobileNetV2 highlights its potential for real-world deployment in resource-constrained
environments. Our methodology provides a robust framework for dragon fruit stem disease
classification, ensuring precise disease identification and supporting agricultural management
practices.
