Zidan Hilmi, Muhammad Rafi (2026) Rancang Bangun Sistem Alat Handheld Klasifikasi Biji Jagung Menggunakan Metode Image Processing Dengan Metode YOLOv8-PSO. Diploma thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 2042221125-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
2042221125-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy |
Abstract
Klasifikasi biji jagung secara manual cenderung memakan waktu dan menghasilkan penilaian yang tidak selalu konsisten. Penelitian ini merancang alat handheld berbasis pengolahan citra untuk mengklasifikasikan biji jagung ke dalam dua kelas, yaitu Good dan Bad, menggunakan model YOLOv8m yang dioptimalkan dengan Particle Swarm Optimization (PSO) pada hyperparameter. Sistem dilengkapi dengan kamera, pencahayaan LED, tray grid dengan Region of Interest (ROI), serta Human Machine Interface (HMI) untuk menampilkan hasil secara real-time. Dataset yang digunakan terdiri atas 1.702 citra. Pengujian dilakukan pada beberapa skenario tray dengan variasi warna tray dan komposisi sampel menggunakan metrik precision, recall, F1-Score, mean Average Precision (mAP), dan Confusion Matrix. Hasil penelitian menunjukkan bahwa YOLOv8m+PSO memberikan kinerja yang lebih baik dibandingkan baseline YOLOv8m, dengan F1-Score sebesar 0,96, akurasi sebesar 0,93, dan waktu pemrosesan selama 7 detik, sedangkan model baseline memperoleh F1-Score sebesar 0,92, akurasi sebesar 0,86, dan waktu pemrosesan selama 10 detik. Peningkatan performa juga ditunjukkan oleh meningkatnya nilai True Positive (TP) dan menurunnya nilai False Negative (FN).
=====================================================================================================================================
Manual classification of corn kernels is time-consuming and often produces inconsistent results. This study developed a handheld image-processing device to classify corn kernels into two categories, namely Good and Bad, using the YOLOv8m model optimized with Particle Swarm Optimization (PSO) for hyperparameter tuning. The system is equipped with a camera, LED lighting, a grid tray with a Region of Interest (ROI), and a Human Machine Interface (HMI) to display classification results in real time. The dataset consisted of 1,702 images. Performance evaluation was conducted under several tray scenarios with variations in tray color and sample composition using precision, recall, F1-Score, mean Average Precision (mAP), and Confusion Matrix metrics. The results demonstrate that YOLOv8m+PSO outperformed the baseline YOLOv8m, achieving an F1-Score of 0.96, an accuracy of 0.93, and a processing time of 7 seconds, whereas the baseline model achieved an F1-Score of 0.92, an accuracy of 0.86, and a processing time of 10 seconds. Performance improvements were also reflected by an increase in True Positive (TP) and a decrease in False Negative (FN) values.
Actions (login required)
 |
View Item |