Linear Variable Filter Hyperspectral Imaging for Determination of Acidity and Hardness of Multiple Fruits

Acidity determines the maturity of fruits and is an important component of fruit taste. Hardness is a key indicator for judging ripening status, storage tolerance, and transportation quality. Therefore, detecting acidity and hardness is of great significance. This article uses a linear gradient filter type hyperspectral imager to obtain hyperspectral data of apples, pears, and kiwifruit. A total of 150 fruits (50 apples, 50 pears, and 50 kiwifruits) were used. The dataset was split into modeling and validation sets in a 1:1 ratio using the concentration gradient (SPXY) method. Six spectral indices (NI, RI, DI, AI, TBNI, TBRI) were used to construct detection models for acidity and hardness across multiple fruit types, suitable for combined multi-fruit datasets. Results showed that the Three-Band Normalized Index (TBNI) achieved the highest accuracy for acidity prediction, with R2C = 0.832, R2V = 0.783, MAE = 0.14, and MRE = 0.03. The Three-Band Ratio Index (TBRI) achieved the highest accuracy for hardness prediction, with R2C = 0.914, R2V = 0.898, MAE = 0.63, and MRE = 0.19. These findings provide technical support for rapid detection of fruit physiological indicators in combined multi-fruit scenarios.