The compression test is the standard procedure to measure fruit firmness in sweet cherries. Nevertheless, this measurement is not always well correlated with perceived texture by buyers and consumers; therefore, the cherry industry needs a better way to grade fruit firmness. Hyperspectral information was correlated to fruit firmness as an alternative to compression values. ‘Skeena’ cherries, grown under commercial conditions in central Washington, were harvested in 2023 and stored for 30 days at 0-1oC. Compression measurements (at 20oC; FirmTech 2, BioWorks Inc) were carried out at harvest and 15 and 30 days into storage. Immediately after these measurements, hyperspectral images from the fruit (n=1030) were taken using a Vi-NIR camera (Headwall Photonics). The comparison between low (< 303 mm/g), medium (303-374 mm/g), and high firmness (>374 mm/g) groups did not yield any spectral differences. Despite this, iPLS wavelength selection showed bands > 800 nm suitable to model these compression groups. On the other hand, Neural Network, Random Forest, and PLS models were not able to predict compression values (regression) or firmness groups (classification). Furthermore, the regression models tested did not have coefficients of determination higher than 0.42 with root mean squared errors of 40 mm/g for compression values with the training dataset. Classification models achieved total accuracies of around 65-70 % and had problems distinguishing between low-medium and medium-high compression values. All models showed poor performance when tested with an independent data set. These results are in contrast to previous reports, which used a lower fruit number, reinforcing the challenge of tailoring a non-destructive technique to predict firmness through compression values in sweet cherries, a highly variable phenotypic characteristic.
