Dragon Fruit Maturity Detection and Quality Grading with Scalable Deep Neural Networks

The paper addresses the critical need for an automated system to accurately determine the ideal harvest time for dragon fruit. It effectively differentiates between fresh and defective fruit, as well as mature and immature fruit, to maintain high-quality standards. The study presents an advanced approach for detecting the maturity and freshness of dragon fruit using pre-trained deep learning models, including VGG16 (a lightweight model), MobileNetV3 (a moderate-weight model), and EfficientNetBO (a heavy-weight model). Each model utilizes three different activation functions: Exponential Linear Unit (ELU), Rectified Linear Unit (ReL U), and Scaled Exponential Linear Unit (SELU). The features extracted by these models, using the best activation function, were then employed to train five different machine learning classifiers: Support Vector Machine (SVM), Decision Tree, Random Forest Classifier, AdaBoost, and CatBoost. Experimental results indicate that for maturity detection, the combination of EfficientN etBO with the SVM Classifier yields the most favorable results. For quality grading, the optimal outcomes are achieved with the MobileNetV3-SVM combination. This hybrid approach achieves a precision and recall of 99%, along with 99% accuracy for both maturity and freshness detection, thus facilitating the assessment of dragon fruit.