Model Caption Otomatis Untuk Deskripsi Kondisi Cuaca Dengan Pendekatan Vision Language Models Blip

Salasa, Sulaeman (2026) Model Caption Otomatis Untuk Deskripsi Kondisi Cuaca Dengan Pendekatan Vision Language Models Blip. Masters thesis, Institut Teknologi Sepuluh November.

Text 6025232009-Master_Thesis.pdf Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Pengamatan kondisi cuaca memegang peranan krusial dalam mitigasi bencana dan perencanaan aktivitas harian. Namun, metode konvensional yang mengandalkan pengamatan manual sering kali terkendala oleh subjektivitas pengamat dan keterbatasan sumber daya manusia dalam melakukan pemantauan secara kontinu. Untuk mengatasi permasalahan tersebut, penelitian ini mengembangkan sistem otomatisasi pengamatan cuaca berbasis Image Captioning dengan menerapkan arsitektur Vision Language Model, yaitu BLIP (Bootstrapping Language-Image Pre-training). Model ini dipilih karena keunggulannya dalam memahami konteks visual dan tekstual secara mendalam melalui mekanisme transfer learning. Penelitian ini bertujuan untuk menghasilkan deskripsi cuaca otomatis yang akurat dengan memanfaatkan model pre-trained yang dilatih ulang (fine-tuned) pada dataset citra cuaca spesifik. Evaluasi dilakukan secara ketat untuk menguji kemampuan adaptasi dan generalisasi model pada domain meteorologi. Berdasarkan hasil eksperimen, model BLIP yang dikembangkan berhasil mencapai nilai evaluasi terbaik dengan skor BLEU-4 sebesar 0.5267 dan skor METEOR sebesar 0.7094. Capaian ini mengindikasikan bahwa sistem terbukti efektif dalam menghasilkan deskripsi kondisi cuaca secara real-time dan konsisten. Hasil penelitian ini menyimpulkan bahwa implementasi model BLIP layak digunakan untuk meningkatkan kualitas data meteorologi dan mengurangi ketergantungan pada pengamatan manual.
===========================================================================================================================
Weather observation plays a crucial role in disaster mitigation and daily activity planning. However, conventional methods relying on manual observation are often constrained by observer subjectivity and limited human resources for continuous monitoring. To address these issues, this research develops an automated weather observation system based on Image Captioning by applying a state-of-the-art Vision Language Model architecture, namely BLIP (Bootstrapping Language-Image Pre-training). This model was selected for its superiority in understanding visual and textual contexts deeply through transfer learning mechanisms. This study aims to generate accurate automated weather descriptions by utilizing a pre-trained model that is fine-tuned on a specific weather image dataset. Rigorous evaluation was conducted to test the model's adaptability and generalization capabilities within the meteorological domain. Based on experimental results, the developed BLIP model achieved the best evaluation values with a BLEU-4 score of 0.5267 and a METEOR score of 0.7094. These achievements indicate that the system is proven effective in generating real-time and consistent weather condition descriptions. The results of this research conclude that the implementation of the BLIP model is feasible for improving meteorological data quality and reducing dependency on manual observation.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Kata Kunci: Image Captioning, Cuaca, Deep Learning, BLIP, Transfer Learning, Fine-tuning. Keywords: Image Captioning, Weather, Deep Learning, BLIP, Transfer Learning, Fine-tuning.
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 > 55101-(S2) Master Thesis
Depositing User:	Sulaeman Salasa
Date Deposited:	02 Feb 2026 06:49
Last Modified:	02 Feb 2026 06:49
URI:	http://repository.its.ac.id/id/eprint/131584

Actions (login required)

View Item