<![CDATA[In real-time dragon fruit testing, classification systems based on D-L, M-L, ANN, SVM, K-NN, CNN, and TL algorithms were used to achieve high accuracy, detect dissimilar images, and enhance the precision of dragon fruit sorting by applying sensor technology, which was previously low. This is owing to the tiny architecture and dataset. To enhance the classification accuracy of the dragon fruit sorting control system, a CNN-TL classification system with Adam Optimizer is presented. Testing 100 epochs with the CNN method and Adam optimizer provided an accuracy of 98.63%. Additionally, testing the same epoch with the TL technique and the Adam optimizer provided an accuracy of 99.11-100%. The CNN-TL approach combined with the Adam optimizer achieved an accuracy rate of 99.55% at iteration 5 and 100% at iteration 100. The accuracy values of CNN, TL, and CNN-TL with the Adam optimizer in the classification of ripe, unripe, and rotten dragon fruit were 33.3%, 85.2%, and 95.3%, respectively. The goal of establishing a sorting control system using the CNN-TL algorithm is to enhance the quality of dragon fruit production in Banyuwangi Regency. During and after the harvest season, production quality is assessed using classification results and automatic sorting.]]>
Copyrights © 2025
