AI-Driven Hybrid Approach to Optimizing Dragonfruit Quality Assessment for Sustainable Agriculture

Dragonfruits are a widely cultivated fruit from the genus Hylocereus, used in various cosmetics like face masks, moisturizers, and serums for its hydrating, antioxidant, and skinnourishing properties that hold significant economic importance, while global dragonfruit industry valued around $1.3 billion in 2020, driven primarily by cosmetics and skin care products. However, post-harvest losses of dragon fruit due to defects can be around 20% to 30%, with some areas seeing losses of about 25% of the total harvest. The integration of AI, deep learning, and image processing is advancing the early and precise detection of dragon fruit quality, facilitating better sorting and grading that will minimize crop losses and enhance the overall efficiency and profitability of dragon fruit production. This study proposed a hybrid model approach utilizing xception and Efficientnetv2-s to classify Fresh and defective dragonfruit. The model was trained, tested and validated on a dataset having 1,652 images, which were pre-processed using techniques such as image inversion, augmentation, and outlier handling, achieving an accuracy of 99.20% along with a precision, recall, and F1-score of 0.9909, 0.9777 and 0.9843 respectively. Explainable AI techniques, such as Grad-CAM and LIME, were employed to better understand the decision-making processes of the model. This research could significantly impact society by advancing grading classification in agriculture, promoting more sustainable farming practices, and improving food security.