Machine Learning-Based Nondestructive Determination of Apple Brix From Near-Infrared Spectra

Brix is a key indicator used to evaluate the quality of apples. To address the problem of low accuracy in nondestructive apple Brix determination, a model combining spectral analysis and neural networks was developed. Specifically, the improved genetic algorithm (GA)-optimized PLS-BP model-comprising partial least squares (PLS) for dimensionality reduction, a back-propagation (BP) neural network for Brix prediction, and an improved GA for optimizing the initial weights and thresholds of the BP network-was proposed to determine apple Brix. The final test results show that the improved GA-optimized PLS-BP model achieves a root mean square error of prediction (RMSEP) of 0.2534 and a coefficient of determination (R²) of 0.9320 on the test set, demonstrating superior predictive performance. In terms of computational efficiency, the inference time of the proposed model is only 1.2 ms, which meets the requirements of practical applications and enables fast, high-precision apple Brix determination, thereby providing effective technical support for apple quality control.