Comparison of sample preparation methods for higher heating values in various sugarcane varieties using near-infrared spectroscopy

This study developed an efficient system for measuring the energy characteristics of energy canes in breeding programs using near-infrared spectroscopy with the aim of significantly improving the accuracy of selecting high-performing sugarcane clones. A key parameter for evaluating energy potential is the heating value, which is typically determined using bomb calorimetry. However, this traditional method is time-consuming because it requires pre-drying of samples and is limited by small sample sizes, potentially leading to inaccuracies and the need for repeated measurements, thereby increasing the overall cost. The study was conducted using two sample preparation conditions: a shredded cane without juice extraction and a dried shredded cane with juice extraction. Two Near-Infrared Spectrometers were evaluated: a laboratory-based FT-NIR spectrometer and portable Micro-NIR device. Spectral data were pre-processed using seven techniques to minimize noise, and four variable selection algorithms–Variable Importance in Projection, Successive Projection Algorithm, Genetic Algorithm, and correlation-based selection via Partial Least Squares Regression–were employed to improve modelling accuracy. In parallel, four machine learning models–AdaBoost, Gradient Boosting, K-Nearest Neighbors, and Random Forest–were applied to the same dataset for Higher heating value prediction. The results revealed that models developed using shredded canes without juice extraction outperformed those based on juice-extracted samples. FT-NIR achieved the highest predictive accuracy, whereas the portable micro-NIR device provided sufficient performance for field-level pre-screening. The integration of NIR spectroscopy with machine learning presents a practical and scalable approach for accelerating the evaluation of sugarcane clones, offering significant advantages in terms of time efficiency, cost reduction, and sample throughput.