Research on solubility prediction method for milk powder quality assessment: based on morphological characteristics

The solubility of instant whole milk powder (IWMP) is a key indicator of reconstitution performance, mainly influenced by particle size and microscopic morphology. Traditional testing methods are manual, slow, and unsuitable for large-scale analysis. This study develops a prediction framework combining image processing and regression models. Milk powder from various brands was sieved and imaged with optical microscopy. Nine shape descriptors were extracted using a MATLAB program, and kernel density estimation (KDE) was applied to generate new features. These features, together with measured solubility values, were used to train machine learning models. Comparative analysis showed that models using enhanced features achieved high accuracy (R² = 0.95, MAE = 0.029), providing a novel image-driven approach for modeling physical properties such as milk powder solubility.