Al Hadad, Afan Ghafar (2026) Sistem Deteksi Jamur Ganoderma Boninense Pada Kelapa Sawit Berbasis Microwave Sensing Dan Machine Learning. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6022232025-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6022232025-Master_Thesis.pdf Restricted to Repository staff only Download (9MB) | Request a copy |
Abstract
Penyakit Basal Stem Rot (BSR) akibat infeksi jamur Ganoderma boninense merupakan ancaman utama bagi produktivitas kelapa sawit karena sering terlambat terdeteksi dan dapat menurunkan hasil secara signifikan. Penelitian ini mengembangkan dan mengevaluasi sistem deteksi Ganoderma boninense berbasis microwave sensing dan machine learning melalui dua pendekatan, yakni pengukuran respon kandungan air pada batang secara in vivo menggunakan radar Continuous Wave (CW) 24 GHz serta pengukuran parameter dielektrik jaringan daun, pelepah, dan akar secara ex vivo menggunakan Vector Network Analyzer (VNA) yang terhubung dengan Dielectric Assessment Kit (DAK) dan open-ended coaxial probe. Pada metode radar, sinyal pantulan dari dua ketinggian batang diolah menjadi deret waktu mean power dan Mean Power Frequency (MPF), kemudian disegmentasi ke dalam interval 5, 10, 15, 20, dan 25 menit dan diringkas menjadi fitur statistik dari batang bawah dan batang atas. Model klasifikasi berbasis XGBoost yang dilatih pada fitur gabungan ini mampu membedakan kelapa sawit sehat, infeksi ringan, dan infeksi berat dengan kinerja tinggi pada seluruh durasi segmentasi, dengan akurasi masing-masing sebesar 91,3%, 89,6%, 92,6%, 87,3%, dan 91,1% untuk segmentasi 5, 10, 15, 20, dan 25 menit. Pada metode VNA–DAK, kurva frekuensi konstanta dielektrik (εᵣ′), faktor rugi dielektrik (εᵣ″), loss tangent (tan δ), dan konduktivitas (σ) pada rentang 4 MHz hingga 3 GHz diringkas menjadi 36 fitur statistik per sampel dan digunakan untuk membangun model klasifikasi per-bagian daun, pelepah, dan akar berbasis XGBoost. Model dielektrik yang dihasilkan mencapai akurasi 90,3% untuk daun, 91,7% untuk pelepah, serta 93,1% untuk akar, yang mengindikasikan bahwa respon radar pada batang maupun sifat dielektrik jaringan daun, pelepah, dan akar mengandung informasi yang kuat untuk membedakan tingkat keparahan infeksi Ganoderma boninense dan memperkuat potensi teknologi microwave sensing berbasis machine learning sebagai dasar pengembangan sistem pemantauan infeksi Ganoderma boninense pada kelapa sawit.
==================================================================================================================================
Basal Stem Rot (BSR) caused by the fungus Ganoderma boninense is a major threat to oil palm productivity because it is often detected too late and can severely reduce yield. This study develops and evaluates a microwave sensing and machine learning based detection system for Ganoderma boninense using two complementary approaches. The first approach performs in vivo monitoring of trunk water-related responses using a 24 GHz continuous wave radar installed at two heights on the stem. The second approach performs ex vivo measurement of dielectric properties of leaf, frond, and root tissues using a vector network analyzer connected to a dielectric assessment kit and an open-ended coaxial probe to obtain the dielectric constant (εᵣ′), dielectric loss factor (εᵣ″), loss tangent (tan δ), and conductivity (σ). For the radar method, backscattered signals from the two trunk heights are processed into time series of mean power and Mean Power Frequency, then segmented into 5, 10, 15, 20, and 25 minute intervals and summarized into statistical features from the lower and upper trunk. XGBoost classifiers trained on these features discriminate healthy, mildly infected, and severely infected trees with high performance across all segment lengths, achieving accuracies of 91.3%, 89.6%, 92.6%, 87.3%, and 91.1% for 5, 10, 15, 20, and 25 minute segments. For the VNA–DAK method, frequency responses of εᵣ′, εᵣ″, tan δ, and σ over 4 MHz to 3 GHz are summarized into 36 statistical features per sample and used to train part-specific XGBoost classifiers for leaf, frond, and root. The resulting dielectric models reach accuracies of about 90.3% for the leaf model, 91.7% for the frond model, and 93.1% for the root model. These findings indicate that both radar responses from the trunk and dielectric properties of leaf, frond, and root tissues carry strong information for discriminating disease severity. Overall, the results highlight the potential of microwave sensing combined with machine learning as a basis for developing objective monitoring systems for basal stem rot in oil palm plantations.
Actions (login required)
 |
View Item |