Implementation of a CNN-trained model for coffee type detection in an Android app with photo input of beans, fruits, and leaves

Abstract

Coffee is the most consumed type of drink in the world. Each type of coffee has different physical characteristics from leaves, fruits to seeds. Now technology is needed in the world of agriculture in making decisions. To determine the type of coffee with fission characteristics, there are still many people who do not understand in distinguishing the physical characteristics of coffee plants. In this case, an application was developed using the RAD method by utilizing the flutter framework and the Convolutional Neural Network model that has been trained. The pre-train model used is NasNet Mobile with a dataset of 900 photos and 100 epochs with early-stopping utilization and heti at epoch 55 with an accuracy of 90.67%. In this study, implementing existing models into Android applications using the Flutter framework. With the implementation process carried out by the application can help the detection process using an android device. The implementation results get good test results with a score of 0.97. This application can help the process of identifying the type of coffee and minimize errors in identifying directly.

Keywords

• cnn
• flutter
• kopi
• android
• implementasi