Sistem Klasifikasi Tingkat Kematangan Buah Kelapa Sawit Menggunakan Sinyal Reflektansi Radar Dan Citra Kamera Berbasis Machine Learning

Resta, Firdausa Sonna Anggara (2026) Sistem Klasifikasi Tingkat Kematangan Buah Kelapa Sawit Menggunakan Sinyal Reflektansi Radar Dan Citra Kamera Berbasis Machine Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 6022232018-Master_Thesis.pdf Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Penentuan tingkat kematangan tandan buah segar (TBS) kelapa sawit berpengaruh langsung terhadap rendemen dan mutu minyak, namun proses grading manual di lapangan cenderung subjektif karena dipengaruhi pencahayaan, pengalaman grader, serta variasi fisik tandan. Tesis ini mengusulkan sistem klasifikasi kematangan TBS berbasis multimodal dengan mengintegrasikan reflektansi radar gelombang mikro 2,4 GHz, citra RGB, dan berat tandan. Radar diimplementasikan menggunakan SDR ADALM-Pluto pada skema continuous wave (CW); sinyal pantulan direkam sebagai sampel baseband dan diringkas menjadi fitur daya pantulan rata-rata (dB). Modalitas visual diproses menggunakan YOLOv8-Nano untuk menghasilkan probabilitas kelas, sedangkan modalitas radar–berat membentuk vektor fitur yang mencakup daya pantulan (dB), berat (kg), dan rasio power-to-weight (dB/kg) yang diklasifikasikan menggunakan XGBoost. Keluaran probabilitas kedua modalitas kemudian digabungkan melalui skema late fusion (mean, weighted, dan stacking) untuk prediksi empat kelas kematangan (unripe, underripe, ripe, dan overripe). Evaluasi pada data uji menunjukkan bahwa model visual mencapai akurasi 0.843 dan F1-score 0.854, sedangkan model radar–berat mencapai akurasi dan F1-score 0.899. Integrasi multimodal meningkatkan kinerja secara signifikan, dengan metode stacking fusion memberikan akurasi dan F1-score makro tertinggi masing-masing 0.949 dan 0.956. Hasil ini menunjukkan bahwa kombinasi radar-vision-weight berpotensi menjadi pendekatan non-destruktif yang lebih konsisten untuk mendukung grading TBS di kebun maupun pabrik.
==================================================================================================================================
Ripeness grading of oil palm fresh fruit bunches (FFB/TBS) directly affects oil yield and quality, yet manual field assessment is often subjective due to lighting variations, grader experience, and bunch-to-bunch physical differences. This thesis proposes a multimodal ripeness classification system that integrates 2.4-GHz microwave radar reflectance, RGB imaging, and bunch weight. Radar measurements are implemented using an ADALM-Pluto SDR in a continuous-wave (CW) configuration; the backscattered signal is recorded as complex baseband samples and summarized into an average backscatter power feature (dB). The visual modality is processed using YOLOv8-Nano to produce class probabilities, while the radar–weight modality forms a feature vector comprising backscatter power (dB), weight (kg), and a power-to-weight ratio (dB/kg), which is classified using XGBoost. The probability outputs from both modalities are then combined via late-fusion schemes (mean, weighted, and stacking) to predict four ripeness classes (unripe, underripe, ripe, and overripe). Evaluation on the test set shows that the visual model achieves an accuracy of 0.843 and an F1-score of 0.854, whereas the radar–weight model achieves an accuracy and F1-score of 0.899. Multimodal integration further improves performance, with stacking fusion yielding the highest accuracy and macro-averaged F1-score of 0.949 and 0.956, respectively. These results indicate that the proposed radar–vision–weight combination is a promising non-destructive approach for more consistent FFB ripeness grading in plantation and mill operations.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	Kelapa Sawit, Tandan Buah Segar, Kematangan, Radar CW 2,4 GHz, YOLOv8, XGBoost, Multimodal, Late Fusion. Oil Palm, Fresh Fruit Bunch, Ripeness, 2.4 GHz CW radar, YOLOv8, XGBoost, Multimodal sensing, Late Fusion.
Subjects:	T Technology > T Technology (General) > T57.8 Nonlinear programming. Support vector machine. Wavelets. Hidden Markov models. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7872 Electromagnetic Devices T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7878 Electronic instruments
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Firdausa Sonna Anggara Resta
Date Deposited:	21 Jan 2026 07:32
Last Modified:	21 Jan 2026 07:32
URI:	http://repository.its.ac.id/id/eprint/129978

Actions (login required)

View Item