Grade and Disease Detection of Dragon Fruit based on Modified VGGNet with Identity Mapping

Abstract

In the agricultural sector, dragon fruit is one of the most frequently harvested fruits throughout the year, regardless of the season. To ensure optimal quality during the harvest period, fruit monitoring should begin from the immature stage. However, dragon fruit is highly susceptible to various diseases, particularly as it approaches ripeness. Therefore, there is a growing need for an automated system that can classify the quality grade of dragon fruit. In addition, such a system should also be able to detect potential diseases affecting fruit. In this study, we propose a modified VGGNet architecture integrated with Identity Mapping. The original VGGNet serves as the backbone. Identity Mapping is then incorporated into each block to improve training efficiency and computational performance. At the final stage, the fully connected and classification layers are fine-tuned to enhance overall accuracy. The proposed method is evaluated using a dragon fruit dataset collected from Bondowoso. Experimental results demonstrate that the proposed approach achieves superior performance, with 99.9% accuracy, recall, and precision. It also outperforms baseline models, including VGG16, AlexNet, and EfficientNet variants B0–B3 and B5–B6.