Segmentasi dan Penilaian Mutu Biji Kopi Robusta Secara Real-Time Menggunakan YOLO11n-Seg Pada Perangkat Android

Abstract

The quality assessment of Robusta coffee beans in Indonesia is generally performed manually based on the Indonesian National Standard (SNI) 01-2907-2008. This approach is time-consuming, highly dependent on the evaluator's experience, and susceptible to inconsistent assessment results. This study aims to develop an Android-based application utilizing the YOLO11n-seg instance segmentation model for real-time Robusta coffee bean quality assessment. A primary dataset was collected by capturing images of Robusta coffee beans and annotating them into 21 classes consisting of one normal bean class and twenty defect classes according to SNI 01-2907-2008. The dataset underwent preprocessing, including image cropping, resizing, and data augmentation, before being used to train the model through transfer learning. The best-performing model was subsequently converted into TensorFlow Lite format and deployed in an Android application developed using Kotlin with the Model–View–ViewModel (MVVM) architecture. Experimental results demonstrated that the proposed YOLO11n-seg model achieved a Precision of 90.18%, Recall of 90.10%, mAP@0.5 of 95.59%, mAP@0.5:0.95 of 91.89%, Mean Intersection over Union of 70.91%, and Mean Dice of 79.22%. Compared with YOLO11s-seg, YOLO11n-seg exhibited lower computational complexity, a smaller model size, and faster inference time, making it more suitable for deployment on resource-constrained Android devices. The developed application is capable of performing real-time segmentation, defect identification, and quality grading of Robusta coffee beans, demonstrating its potential as a faster, more objective, and more consistent alternative to conventional manual inspection.