Non-destructive analysis of Ganoderma lucidum composition using hyperspectral imaging and machine learning

Ganoderma lucidum is a widely used medicinal fungus whose quality is influenced by various factors, making traditional chemical detection methods complex and economically challenging. This study addresses the need for fast, noninvasive testing methods by combining hyperspectral imaging with machine learning to predict polysaccharide and ergosterol levels in Ganoderma lucidum cap and powder.

Hyperspectral images in the visible near-infrared (385-1009 nm) and short-wave infrared (899-1695 nm) ranges were collected, with ergosterol measured by high-performance liquid chromatography and polysaccharides assessed via the phenol-sulfuric acid method. Three machine learning models-a feedforward neural network, an extreme learning machine, and a decision tree-were tested.

Notably, the extreme learning machine model, optimized by a genetic algorithm with voting, provided superior predictions, achieving R ² values of 0.96 and 0.97 for polysaccharides and ergosterol, respectively.

This integration of hyperspectral imaging and machine learning offers a novel, nondestructive approach to assessing Ganoderma lucidum quality.