Maturity Assessment-Based Dry Matter Content in Mon Thong Durian Pulp Using Machine Learning and Near-Infrared Hyperspectral Image Analysis

Durian is a high-value tropical fruit cultivated widely across Southeast Asia, with Mon Thong being the most commercially favored variety in Thailand. Accurate assessment of durian maturity is essential to ensure optimal quality and prevent premature harvesting, which negatively affects consumer satisfaction and export standards. This study proposed a method for predicting the dry matter content (DMC), a key indicator of durian maturity, using near-infrared hyperspectral imaging (NIR-HSI) combined with machine learning and neural network algorithms. In total, 300 Mon Thong durian fruits were analyzed across six maturity stages. Spectral data were preprocessed using the standard normal variate (SNV), multiplicative scatter correction (MSC), and derivative transformations, followed by feature extraction via principal component analysis (PCA). Four predictive models were developed: partial least squares regression (PLSR), least squares support vector machine (LS-SVM), multi-layer perceptron (MLP), and one-dimensional convolutional neural network (1D-CNN). Model performance was evaluated using the coefficient of determination (R²), root mean square error (RMSE), mean bias error (MBE), ratio of performance to deviation (RPD), and range error ratio (RER) metrics. Based on the results, LS-SVM and 1D-CNN outperformed the other models, with 1D-CNN achieving the highest validation accuracy (R² = 0.97, RMSE = 1.49%, MBE = 0.06, RPD = 5.9, RER = 24.9) when combined with PCA and SNV preprocessing. While PLSR offered simplicity and interpretability, the neural network models demonstrated superior predictive performance and generalization. The findings highlighted the potential of integrating NIR-HSI with advanced learning algorithms for reliable durian maturity assessment, supporting improved quality control and market readiness.