Classification System of Crystal Guava (Psidium Guajava) Using Convolutional Neural Network And Rectrified Linear Unit Method Based on Android

Abstract

These instructions Abstract - However, determining the ripeness of fruit is frequently done by hand, which presents problems with consistency and efficiency. In order to improve the sorting of crystal guava fruit maturity, this study suggests combining machine learning technology with the creation of digital image-based apps. Fruit ripeness is classified using a convolutional neural network (CNN), a deep learning model, based on the color of its skin. It is anticipated that the method will increase productivity and offer superior precision while sorting crystal guava fruit. The System Development Life Cycle (SDLC) with a Waterfall approach is the methodology employed. The system design formed from the deep learning model resulted in excellent performance in classifying images of crystal guava fruit by utilizing model training from the base models ResNet50V2, DenseNet121, NASNetMobile, and MobileNetV2 with a combination of training using K-fold cross-validation with a 5-fold configuration. The best-trained model achieved an average highest accuracy of 99.92% in model training using MobileNetV2 with the lowest average loss value of 0.0088. The system application was developed using mobile Android, leveraging the Flutter framework and Dart programming language. The research results demonstrate a comparison of testing on crystal guava and local guava fruits against ripeness classification parameters