Otomasi Estimasi Nilai Ekonomi Sampah Anorganik Menggunakan Instance Segmentation (Mask R-Cnn) Secara Real-Time Dan Integrasi Harga Pasar

Fadhilah, Fazrul Ahmad (2026) Otomasi Estimasi Nilai Ekonomi Sampah Anorganik Menggunakan Instance Segmentation (Mask R-Cnn) Secara Real-Time Dan Integrasi Harga Pasar. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027221025-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Pengelolaan sampah anorganik di bank sampah masih banyak dilakukan secara manual, terutama pada proses identifikasi jenis sampah, penentuan bobot, dan perhitungan nilai ekonominya. Proses tersebut membutuhkan ketelitian pengguna serta berpotensi menimbulkan kesalahan, khususnya pada sampah dengan karakteristik visual yang mirip. Oleh karena itu, penelitian ini mengembangkan sistem otomasi estimasi nilai ekonomi sampah anorganik berbasis citra digital menggunakan instance segmentation Mask R-CNN dan integrasi harga pasar. Sistem yang dikembangkan mampu mendeteksi enam kelas sampah, yaitu PET bening bersih, PET biru bersih, PET kotor, kardus bagus, kardus jelek, dan HVS. Model utama menggunakan Mask R-CNN dengan backbone ResNet-50 FPN. Hasil segmentasi berupa kelas objek, bounding box, confidence score, dan mask digunakan untuk mengekstraksi fitur visual objek. Marker ArUco digunakan sebagai referensi skala untuk mengonversi ukuran piksel ke sentimeter. Estimasi massa dilakukan menggunakan pendekatan cangkang tipis untuk botol PET dan pendekatan dua tampilan untuk HVS serta kardus. Massa estimasi kemudian dikalikan dengan harga per kilogram untuk menghasilkan estimasi nilai ekonomi. Hasil pengujian menunjukkan bahwa ResNet-50 FPN memberikan performa yang seimbang dengan precision 0,4984, recall 0,5667, F1-score 0,5304, AP50 0,5643, mAP50-95 0,4619, mean mask IoU 0,9051, dan mean Dice 0,9478. Validasi estimasi massa menghasilkan MAE 76,35 gram dan MAPE 55,66%. Error terkecil diperoleh pada kardus bagus sebesar 1,73% dan kardus jelek sebesar 3,46%, sedangkan error terbesar terdapat pada HVS. Hasil ini menunjukkan bahwa sistem mampu melakukan deteksi, segmentasi, estimasi massa, dan estimasi nilai ekonomi secara otomatis, tetapi masih memerlukan peningkatan pada dataset, kalibrasi parameter estimasi, dan mekanisme pengambilan gambar.
===============================================================================================================================
Inorganic waste management in waste banks is still mostly performed manually, particularly in identifying waste types, determining weight, and calculating economic value. This process requires user accuracy and may lead to errors, especially for waste objects with similar visual characteristics. Therefore, this research develops an automated system for estimating the economic value of inorganic waste based on digital images using Mask R-CNN instance segmentation and market price integration. The developed system detects six waste classes: clean clear PET, clean blue PET, dirty PET, good cardboard, poor cardboard, and HVS paper. The main model uses Mask R-CNN with a ResNet-50 FPN backbone. The segmentation outputs, including object class, bounding box, confidence score, and mask, are used to extract visual features. An ArUco marker is used as a scale reference to convert pixel measurements into centimeters. Mass estimation is performed using a thin-shell approach for PET bottles and a two-view approach for HVS paper and cardboard. The estimated mass is then multiplied by the price per kilogram to obtain the estimated economic value. The experimental results show that ResNet-50 FPN provides the most balanced performance, with precision of 0.4984, recall of 0.5667, F1-score of 0.5304, AP50 of 0.5643, mAP50-95 of 0.4619, mean mask IoU of 0.9051, and mean Dice of 0.9478. The mass estimation validation produces an MAE of 76.35 grams and a MAPE of 55.66%. The lowest errors are obtained on good cardboard at 1.73% and poor cardboard at 3.46%, while the highest error occurs on HVS paper. These results indicate that the system can automatically perform detection, segmentation, mass estimation, and economic value estimation, although improvements are still needed in the dataset, estimation parameter calibration, and image acquisition mechanism.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Bank Sampah, Instance Segmentation, Mask R-CNN, ArUco, Estimasi Bobot, Nilai Ekonomi Sampah. Waste Bank, Instance Segmentation, Mask R-CNN, ArUco, Weight Estimation, Waste Economic Value.
Subjects:	T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Fazrul Ahmad Fadhilah
Date Deposited:	30 Jul 2026 04:23
Last Modified:	30 Jul 2026 04:23
URI:	http://repository.its.ac.id/id/eprint/140311

Actions (login required)

View Item