Penentuan Kematangan Buah Melon Golden Langkawi Berbasis Neuro-Symbolic AI Melalui Integrasi Deep Learning dan Web Semantik

Umar, Ubaidillah (2025) Penentuan Kematangan Buah Melon Golden Langkawi Berbasis Neuro-Symbolic AI Melalui Integrasi Deep Learning dan Web Semantik. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 7022 221 013-Doctoral_Thesis.pdf] Text
7022 221 013-Doctoral_Thesis.pdf - Accepted Version
Restricted to Repository staff only

Download (5MB) | Request a copy

Abstract

Pertanian melon menghadapi tantangan dalam menentukan waktu panen yang optimal dan mengelola praktik budidaya secara efisien, terutama bagi petani pemula. Kesulitan ini diperburuk oleh cuaca yang tidak menentu dan keterbatasan teknologi berbasis data, sehingga berpotensi menyebabkan keputusan panen yang kurang tepat. Oleh karena itu, pengembangan sistem yang dapat mendeteksi kematangan melon secara akurat dan memberikan rekomendasi berbasis data sangat penting untuk meningkatkan hasil pertanian. Penelitian ini bertujuan untuk mengembangkan sistem berbasis Web Semantik yang mengintegrasikan YOLOv8 dengan mekanisme perhatian dan ontologi semantik untuk memberikan rekomendasi real-time kepada petani melon. Optimasi YOLOv8 memungkinkan deteksi yang lebih akurat pada kondisi lapang yang beragam, dengan evaluasi kuantitatif menunjukkan precision 0.98, recall 0.96, dan mAP@50 0.98 dalam mendeteksi buah matang dan belum matang. Waktu inferensi rata-rata 0.9 ms/gambar menunjukkan kemampuan real-time. Pendekatan ini merepresentasikan Neuro-Symbolic AI, yang memadukan persepsi visual berbasis YOLOv8 (neural) dengan penalaran simbolik menggunakan ontologi semantik dan W3C Time Ontology (timestamp dipetakan ke time:Instant, fase budidaya ke time:Interval). Informasi visual dari YOLOv8 dikonversi menjadi pengetahuan terstruktur dalam ontologi, lalu digabungkan dengan data lingkungan dari sensor (suhu udara, kelembapan udara, pH tanah, dan soil moisture) serta data cuaca eksternal. Dengan aturan penalaran berbasis SPARQL dan representasi temporal menggunakan time:Instant dan time:Interval, sistem dapat memberikan rekomendasi waktu panen yang mempertimbangkan riwayat kondisi lingkungan dan tingkat kematangan aktual di setiap petak tanaman. Berdasarkan hasil pengujian, sistem yang dikembangkan terbukti mampu memberikan rekomendasi waktu panen secara dinamis dengan menggabungkan deteksi kematangan visual dan kondisi lingkungan terkini. Meskipun hasil yang diperoleh sudah memadai, pengembangan lebih lanjut diperlukan untuk meningkatkan akurasi deteksi pada kondisi pencahayaan ekstrem dan occlusion, serta mendorong penerapan sistem ini pada skala besar dalam konteks smart agriculture.
====================================================================================================================================
Melon farming faces challenges in determining the optimal harvest time and efficiently managing cultivation practices, especially for beginner farmers. This difficulty is exacerbated by unpredictable weather, hard-to-detect pests, and the limitations of data-driven technologies. Therefore, developing a system that can accurately detect melon ripeness and provide data-driven recommendations is crucial to improving agricultural productivity. This study aims to develop a Semantic Web-based system that integrates YOLOv8 with an attention mechanism and semantic ontologies to provide real-time recommendations to melon farmers. Optimization of YOLOv8 allows for more accurate detection under various field conditions, with quantitative evaluation showing precision of 0.98, recall of 0.96, and mAP@50 of 0.98 in detecting ripe and unripe melons. The average inference time of 0.9 ms/image indicates real-time capabilities. This approach represents Neuro-Symbolic AI, combining YOLOv8-based visual perception (neural) with symbolic reasoning using semantic ontologies and W3C Time Ontology (timestamp mapped to time:Instant, growing phases to time:Interval). The visual information from YOLOv8 is converted into structured knowledge within the ontology, then combined with environmental data (temperature, humidity, soil pH, and external weather data). Using SPARQL-based reasoning and temporal representation with time:Instant and time:Interval, the system can provide harvest recommendations that consider the historical environmental conditions and the current ripeness level of melons in each plot. Based on the experimental results, the developed system has proven capable of dynamically providing harvest recommendations by combining visual ripeness detection and real-time environmental conditions. Although the results are satisfactory, further development is needed to enhance detection accuracy, especially under extreme lighting conditions and occlusion, and to encourage the system's application at a larger scale within the context of smart agriculture.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: Neuro-Symbolic AI; Web Semantik; MelonONT; W3C Time Ontology; YOLOv8; smart agriculture Neuro-Symbolic AI; Semantic Web; MelonONT; W3C Time Ontology; YOLOv8; smart agriculture
Subjects: T Technology > T Technology (General) > T58.62 Decision support systems
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis
Depositing User: Ubaidillah Umar
Date Deposited: 04 Dec 2025 05:58
Last Modified: 04 Dec 2025 05:58
URI: http://repository.its.ac.id/id/eprint/128868

Actions (login required)

View Item View Item