ALDOTAMA, ALVINO (2026) PERBANDINGAN MOBILENET V2 DENGAN YOLO HEAD DAN FPN UNTUK DETEKSI SAMPAH ORGANIK-ANORGANIK. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text (Halaman Cover)
COVER.pdf Download (492kB) | Preview |
|
![[img]](https://repository.mercubuana.ac.id/style/images/fileicons/text.png) |
Text (Bab 1)
BAB 1.pdf Restricted to Registered users only Download (133kB) |
|
|
Text (Bab 2)
BAB 2.pdf Restricted to Registered users only Download (314kB) |
|
|
Text (Bab 3)
BAB 3.pdf Restricted to Registered users only Download (240kB) |
|
|
Text (Bab 4)
BAB 4.pdf Restricted to Registered users only Download (728kB) |
|
|
Text (Bab 5)
BAB 5.pdf Restricted to Registered users only Download (116kB) |
|
|
Text (Daftar Pustaka)
DAFTAR PUSTAKA.pdf Restricted to Registered users only Download (160kB) |
|
|
Text (Lampiran)
LAMPIRAN.pdf Restricted to Registered users only Download (477kB) |
Abstract
Pengelolaan sampah yang efektif memerlukan pemilahan yang akurat antara sampah organik dan anorganik. Penelitian ini bertujuan untuk mengevaluasi dan membandingkan performa dua arsitektur deep learning berbasis MobileNetV2 untuk deteksi sampah: MobileNetV2 dengan YOLO Head dan MobileNetV2 dengan Feature Pyramid Network (FPN). Dataset yang digunakan terdiri dari 2,080 gambar sampah dengan proporsi 70% training, 20% validation, dan 10% testing. Kedua model dilatih menggunakan konfigurasi hyperparameter yang identik selama 50 epochs dengan data augmentation yang komprehensif meliputi HSV adjustment, geometric transformation, mosaic, dan mixup augmentation. Hasil eksperimen menunjukkan bahwa kedua model mencapai performa sangat tinggi dengan [email protected] di atas 99%. Model MobileNetV2+YOLO mencapai mAP 99.4%, precision 98.2%, dan recall 99.5% dengan ukuran model 5.96 MB, sementara MobileNetV2+FPN mencapai mAP 99.5%, precision 99.6%, dan recall sempurna 100% dengan ukuran model 21.48 MB. Analisis confusion matrix menunjukkan FPN memiliki keunggulan dengan zero false positives dan total error 60% lebih rendah. Evaluasi trade-off mengungkapkan bahwa YOLO menawarkan efisiensi superior (3.6x lebih kecil, inference lebih cepat) yang ideal untuk edge devices dan real-time processing, sedangkan FPN memberikan akurasi maksimum yang cocok untuk mission-critical applications. Penelitian ini memberikan panduan praktis dalam memilih arsitektur deteksi sampah yang sesuai dengan requirements dan constraints deployment scenario spesifik. Effective waste management requires accurate sorting between organic and inorganic waste. This research aims to evaluate and compare the performance of two MobileNetV2-based deep learning architectures for waste detection: MobileNetV2 with YOLO Head and MobileNetV2 with Feature Pyramid Network (FPN). The dataset consists of 2,080 waste images with a 70% training, 20% validation, and 10% testing split. Both models were trained using identical hyperparameter configurations for 50 epochs with comprehensive data augmentation including HSV adjustment, geometric transformation, mosaic, and mixup augmentation. Experimental results demonstrate that both models achieve excellent performance with [email protected] above 99%. The MobileNetV2+YOLO model achieves 99.4% mAP, 98.2% precision, and 99.5% recall with a model size of 5.96 MB, while MobileNetV2+FPN achieves 99.5% mAP, 99.6% precision, and perfect 100% recall with a model size of 21.48 MB. Confusion matrix analysis shows that FPN has advantages with zero false positives and 60% lower total errors. Trade-off evaluation reveals that YOLO offers superior efficiency (3.6x smaller, faster inference) ideal for edge devices and real-time processing, while FPN provides maximum accuracy suitable for mission-critical applications. This research provides practical guidance in selecting waste detection architectures appropriate to specific deployment scenario requirements and constraints.
Actions (login required)
 |
View Item |
