Identifikasi Kontaminasi Produk Agri-Food pada Citra Hiperspektral Menggunakan Pendekatan Deteksi Anomali Berbasis Gated Transformer

Alamsyah, Surya Fadli (2026) Identifikasi Kontaminasi Produk Agri-Food pada Citra Hiperspektral Menggunakan Pendekatan Deteksi Anomali Berbasis Gated Transformer. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221059-Undergraduate-Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Produk oat rentan terhadap masuknya kontaminan dari panen, pemrosesan, hingga pengemasan sehingga mengancam keamanan pangan. Kontaminan seperti plastik, logam, kaca, dan residu tanaman sering kali memiliki karakteristik visual yang menyerupai produk sehingga sulit dibedakan menggunakan citra RGB. Oleh karena itu, diperlukan Hyperspectral Imaging (HSI) yang merekam ratusan band spektral sehingga mampu membedakan karakteristik material kontaminan dari produk. Namun, kontaminasi bersifat jarang dan bervariasi sehingga data kontaminan berlabel sulit diperoleh. Dalam penelitian ini, kontaminan dipandang sebagai anomali karena memiliki spectral signature yang berbeda dari produk normal dan hanya menempati sebagian kecil area citra. Oleh karena itu, penelitian ini menerapkan pendekatan deep learning secara unsupervised. Penelitian ini mengembangkan model deteksi kontaminasi berbasis deteksi anomali pada HSI menggunakan arsitektur Gated Transformer dengan mekanisme dual branch dan Content Matching Method (CMM) yang memisahkan fitur data normal dan fitur anomali secara adaptif. Evaluasi dilakukan pada 41 data uji dari dataset HSI-AgriFoodAnomaly. Eksplorasi konfigurasi model mencakup variasi encoder depth, ukuran patch, jumlah komponen Principal Component Analysis (PCA), dan fungsi loss. Model terbaik mencapai rata-rata AUC sebesar 0,846 serta memberikan performa terbaik dibandingkan enam metode pembanding, yaitu DMS2F, SGLNet, OTAD, Auto-AD, BockNet, dan U-Net. Hasil penelitian menunjukkan bahwa arsitektur Gated Transformer mampu mendeteksi kontaminasi pada produk oat. Kata kunci: Citra Hiperspektral, Deteksi Anomali, Gated Transformer, Unsupervised, Keamanan Pangan.
======================================================================================================================================
Oat products are susceptible to contamination during harvesting, processing, and packaging, posing a threat to food safety. Contaminants such as plastic, metal, glass, and plant residues often exhibit visual characteristics similar to those of oat products, making them difficult to distinguish using RGB images. Therefore, Hyperspectral Imaging (HSI), which captures hundreds of spectral bands, is employed to differentiate the material characteristics of contaminants from those of the products. However, contamination events are rare and highly diverse, making labeled contamination data difficult to obtain. In this study, contaminants are treated as anomalies because they exhibit spectral signatures different from those of normal products while occupying only a small portion of the image. Therefore, an unsupervised deep learning approach is adopted. This study develops an anomaly detection-based contamination detection model for HSI using a Gated Transformer architecture with a dual-branch mechanism and a Content Matching Method (CMM) to adaptively separate normal and anomalous feature representations. The proposed model was evaluated on 41 test samples from the HSI-AgriFoodAnomaly dataset. Configuration exploration included variations in encoder depth, patch size, the number of Principal Component Analysis (PCA) components, and loss functions. The best-performing model achieved an average Area Under the Receiver Operating Characteristic Curve (AUC) of 0.846 and outperformed six baseline methods, namely DMS2F, SGLNet, OTAD, Auto-AD, BockNet, and U-Net. The results demonstrate that the Gated Transformer architecture is capable of detecting contamination in oat products.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Citra Hiperspektral, Deteksi Anomali, Gated Transformer, Unsupervised, Keamanan Pangan, Hyperspectral Imaging, Anomaly Detection, Gated Transformer, Unsupervised, Food Safety
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques. Image analysis--Data processing.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Surya Fadli Alamsyah
Date Deposited:	24 Jul 2026 07:25
Last Modified:	24 Jul 2026 07:25
URI:	http://repository.its.ac.id/id/eprint/137274

Actions (login required)

View Item