Kelor: a mobile application for moringa powder quality detection using CNN ensemble transfer learning (case study: moringa powder)

Studies on powder quality detection using image processing have been widely conducted, particularly in the pharmaceutical field; however, the quality detection of moringa powder remains largely unexplored. Moringa powder, commonly used in products such as medicine, tea, and masks, is highly sensitive to sun and air exposure, which can significantly alter its quality. Therefore, reliable detection of moringa powder fit for consumption is crucial to ensure product safety and quality. This study investigates the application of a Convolution Neural Networks (CNN)-based transfer learning ensemble method for moringa powder quality detection, detailing aspects like datasets, pre-trained models, transfer learning concepts, ensemble techniques, and performance evaluation. In this research, ResNeSt101 and DenseNet121 were employed as pre-trained models within the transfer learning process. The evaluation results demonstrated that the ensemble transfer learning approach performed exceptionally well, achieving an AUC score of 0.868, with accuracies of 0.836 for ResNeSt101 and 0.838 for DenseNet121. These results highlight the effectiveness of this approach in detecting moringa powder suitable for consumption, underscoring the potential of the transfer learning ensemble method as a reliable solution for moringa powder quality detection.