Machine learning assisted-hyperspectral imaging for in-situ evaluation of compost maturity

Accurate and rapid assessment of compost maturity is essential for ensuring the quality and safe utilization of compost products in organic waste recycling systems. However, compost maturity is inherently multi-dimensional, and models trained on single indicators with inconsistent thresholds often exhibit limited robustness and generalizability. In addition, existing hyperspectral imaging (HSI) studies frequently rely on full-spectrum modeling or single-indicator supervision, constraining their practical applicability. To address these limitations, this study developed an integrated HSI-machine learning framework guided by a unified maturity representation derived from multiple chemical and biological indicators. Following principal component analysis-based spectral reduction, four machine learning models were established for maturity classification and regression. Random Forest and Extreme Gradient Boosting demonstrated superior performance, achieving recall rates of 96.30% and 98.72% for classification and R² values of 0.917 and 0.886 for maturity regression, respectively. Replacing full-spectrum inputs with ten characteristic bands effectively reduced spectral redundancy and computational cost while decreasing the mean relative error to 2.42% and maintaining model stability. Overall, the proposed framework provides a rapid, non-destructive, and characteristic-band-based strategy for compost maturity monitoring, supporting intelligent quality control in sustainable waste management systems.