Klasifikasi Tingkat Kematangan Buah Kelapa Sawit Menggunakan Metode Semi-Supervised Learning

Adi, Azarel Grahandito (2026) Klasifikasi Tingkat Kematangan Buah Kelapa Sawit Menggunakan Metode Semi-Supervised Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221126-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB)

Abstract

Industri kelapa sawit merupakan pilar vital bagi perekonomian Indonesia, namun kualitasnya sangat bergantung pada akurasi pemanenan buah pada tingkat kematangan yang optimal. Metode penilaian visual yang umum digunakan saat ini bersifat subjektif dan tidak efisien. Meskipun pendekatan supervised learning dengan visi komputer menawarkan akurasi tinggi, ketergantungannya pada data berlabel dalam jumlah masif menjadi penghalang utama untuk implementasi skala industri karena proses pelabelan yang mahal dan memakan waktu. Penelitian ini mengusulkan pendekatan Semi-Supervised Learning (SSL) yang mengintegrasikan sifat ordinal melalui fungsi loss ordinal untuk meningkatkan akurasi model citra terhadap dataset kematangan buah sawit. Pengintegrasian fungsi loss CORN ke dalam algoritma SSL SoftMatch direalisasikan melalui mekanisme Ordinal Confidence (OC), yang mengadaptasi pembobotan berdasarkan konsistensi keordinalan. Pada skenario 25% data berlabel, metode SoftMatch dengan CORN dan OC mencapai akurasi sebesar 0,898, F1-Score 0,899, dan MAE 0,146, yang secara signifikan lebih baik dibandingkan penggunaan Cross-Entropy. Pada skenario dengan jumlah data berlabel yang lebih terbatas, yaitu 1% dan 10% sampel, pendekatan ini juga menunjukkan peningkatan akurasi dan F1-Score sebesar 2–5% serta penurunan MAE hingga 8-15%. Penelitian ini memiliki potensi yang kuat untuk diterapkan pada permasalahan klasifikasi ordinal di domain lain.
==============================================================================================================================
The palm oil industry is a vital pillar of the Indonesian economy, yet its quality is critically dependent on the accurate harvesting of fruit at optimal maturity levels. Current common visual assessment methods are subjective and inefficient. Although supervised learning approaches with computer vision offer high accuracy, their reliance on massively labeled datasets presents a major barrier to industrial-scale implementation due to the costly and time-consuming labeling process. This study proposes a Semi-Supervised Learning (SSL) approach that integrates ordinal properties through an ordinal loss function to improve image-based model accuracy. The integration of the CORN loss function into the SSL SoftMatch algorithm is realized through the Ordinal Confidence (OC) mechanism, which adapts sample weighting based on confidence through ordinal consistency. In the setting with 25% labeled samples, the SoftMatch method combined with CORN and OC achieves an accuracy of 0.898, an F1-score of 0.899, and an MAE of 0.146, significantly surpassing the performance obtained using Cross-Entropy loss. Under more limited labeled-data conditions, namely with 1% and 10% labeled samples, the proposed approach also yields consistent improvements, with accuracy and F1-score increases of approximately 2-5% and MAE reductions of up to 8-15%. These findings indicate strong potential for application to other ordinal classification domains.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Kelapa Sawit, Klasifikasi Citra, Semi Supervised Learning, Ordinal Classification, Loss Function
Subjects:	Q Science > QA Mathematics > QA76.76.A65 Application software. Enterprise application integration (Computer systems)
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Mr. Tondo Indra Nyata
Date Deposited:	13 Oct 2026 05:29
Last Modified:	13 Oct 2026 05:29
URI:	http://repository.its.ac.id/id/eprint/145520

Actions (login required)

View Item