Development of a three-dimensional near-infrared hyperspectral imaging technique for non-destructive visualization of soluble solids content in kiwifruit

Although recent advances in near-infrared hyperspectral imaging (NIR-HSI) have shown promise for non-invasive quality assessment of various agricultural products, challenges remain in correcting the effects caused by curvature and variability in sample shapes. This study has investigated a novel approach that combines a push-broom line-scanning NIR-HSI camera, sample rotator, and 3D laser profiler to simultaneously capture the spectral imaging and surface geometry data of kiwifruit. Subsequently, angle and height corrections were applied to the hyperspectral data using Lambert-based and calibration-based methods, respectively. Finally, partial least squares (PLS) regression analysis was employed to develop soluble solid content (SSC) calibration models for further mapping analysis. Overall, the coefficients of determination (R²) and the root mean squared error (RMSE) were 0.61 and 0.54 % for the calibration set, 0.52 and 0.52 % for the validation set, respectively. In contrast to previous NIR-HSI studies, although the enhancement of the average SSC prediction accuracy through PLS regression analysis was not truly achieved, the corrected models effectively mitigated the influence of geometric shape distortions. This adjustment enabled the non-destructive 3D visualization of the SSC distribution across the entire surface of the kiwifruit. In the mapping test, despite variations in sample sizes, the differences in SSC among the samples were clearly identifiable, which also underscores the importance of shape correction in spectral image data. These findings demonstrate that this method has the potential to revolutionize the quality evaluation of irregularly shaped agricultural products.