Prediksi Cepat Kandungan Kimia Buah Kelapa Sawit Secara Non-Destruktif Menggunakan Portable NIR Spectrometer

Abstract

Metode rapid dan non-destruktif merupakan kebutuhan esensial untuk memprediksi kadar kimia buah kelapa sawit, yang hingga saat ini menjadi tantangan bagi petani maupun pelaku agroindustri untuk mencapai operasional yang efisien dan praktis di lapangan. Data kimia awal di lapangan mengenai kandungan minyak, asam lemak bebas (ALB), dan air dalam buah kelapa sawit sangat penting untuk menjamin kualitas dan menentukan nilai komersial di pasaran. Metode pengukuran konvensional bersifat kompleks, merusak, hanya menargetkan satu bahan kimia, membutuhkan keahlian khusus, mahal, dan memakan waktu. Spektroskopi Near-Infrared (NIR) menjawab sebagian permasalahan tersebut melalui prinsip non-destruktif dan efisiensinya yang tinggi. Namun, dimensi besar dan harga spektrometer yang mahal menjadi kendala dalam operasional dan hilirisasinya. Integrasi antara spektrometer NIR portabel, kemometrika, dan machine learning diusulkan sebagai metode prediksi yang inovatif, efisien, dan akurat. Dengan demikian penelitian ini bertujuan untuk mengembangkan metode prediksi kadar kimia buah kelapa sawit menggunakan portable NIR spectrometer dengan kalibrasi Partial Least Squares Regression (PLSR) dan hybrid Partial Least Squares-Artificial Neural Network (PLS-ANN). Sejumlah 408 sampel buah kelapa sawit (Elaeis guineensis Jacq). varietas Tenera dikumpulkan dari Perkebunan Kelapa Sawit Cikabayan, IPB University, Indonesia, yang mewakili 10 tingkat kematangan (3–6 maa). Pengukuran spektra sampel dilakukan menggunakan spektrometer near-infrared (NIR) portabel dengan panjang gelombang 740–1070 nm untuk merekam karakteristik spektra reflektansi, yang kemudian ditransformasikan menjadi spektra absorbansi. Pretreatment spektra yang dilakukan yaitu first derivative Savitzky-Golay (D1 SG), kombinasi D1 SG + smoothing Savitzky-Golay (smoothing SG), Detrending, D1 SG + Detrending, dan Standard Normal Variate (SNV) + D1 SG. Hasil penelitian menunjukkan bahwa PLS-ANN memiliki kemampuan prediktif lebih baik dari PLSR. Metode prediksi kadar minyak terbaik yaitu kombinasi PLS-ANN dan pretreatment D1 SG + smoothing SG (R²kal = 0,98 dan RPDval = 6,83). Metode prediksi kadar ALB terbaik yaitu kombinasi PLS-ANN dengan pretreatment D1 SG + smoothing SG (R²kal = 0,81 dan RPDval = 2,34), sedangkan metode prediksi kadar air terbaik yaitu kombinasi PLS-ANN dan pretreatment detrending (R²kal = 1,00 dan RPDval = 12,52). Hasil menunjukkan bahwa metode pengolahan spektra dan kalibrasi yang dikembangkan tergolong akurat untuk penentuan kandungan minyak dan air buah sawit secara nondestruktif sehingga memiliki potensi yang baik untuk diintegrasikan dengan perangkat spektroskopi untuk keperluan pengukuran secara praktis baik di tingkat petani maupun industri sawit.

Rapid and non-destructive methods are essential for predicting the chemical content of oil palm fruit, a challenge that has hindered farmers and agro-industrial players from achieving efficient and practical field operations. Initial field chemical data regarding the water, oil, and free fatty acid (FFA) content in oil palm fruit is crucial for ensuring quality and determining commercial value in the market. Conventional measurement methods are complex, destructive, target only one chemical, require specialized expertise, are expensive, and are time-consuming. Near-Infrared (NIR) spectroscopy addresses some of these limitations through its non-destructive principles and high efficiency. However, the large dimensions and high price of the spectrometer are obstacles to its operation and downstreaming. The integration of a portable NIR spectrometer, chemometrics, and machine learning is proposed as an innovative, efficient, and accurate prediction method. This study aims to develop a prediction method for the chemical content of oil palm fruits, using a portable NIR spectrometer with calibration of Partial Least Square Regression (PLSR) and hybrid Partial Least Square-Artificial Neural Network (PLS-ANN) A total of 408 samples of oil palm (Elaeis guineensis Jacq.) fruit of the Tenera variety were collected from the Cikabayan Oil Palm Plantation, IPB University, Indonesia, representing 10 maturity levels (3–6 maa). Sample spectra were measured using a portable near-infrared (NIR) spectrometer with a wavelength of 740–1070 nm to record the characteristics of the reflectance spectra, which were then transformed into absorbance spectra. The spectra pretreatments performed were the first derivative Savitzky-Golay (D1 SG), a combination of D1 SG + smoothing Savitzky-Golay (smoothing SG), Detrending, D1 SG + Detrending, and Standard Normal Variate (SNV) + D1 SG. The results showed that PLS-ANN had better predictive ability than PLSR. The best oil content prediction method was a combination of PLS-ANN and D1 SG pretreatment + SG smoothing (R²cal = 0.98 and RPDval = 6.83). The best FFA prediction method was a combination of PLS-ANN with D1 SG pretreatment + SG smoothing (R²cal = 0.81 and RPDval = 2.34), while the best water content prediction method was a combination of PLS-ANN and detrending pretreatment (R²cal = 1.00 and RPDval = 12.52). The results showed that the developed spectral processing and calibration methods were quite accurate for non-destructive determination of oil and water content of palm fruit, so that they have good potential to be integrated with spectroscopic devices for practical measurement purposes both at the farmer and palm oil industry levels.