Japan is a major export market for the Korean cut rose flower industry. The longevity of cut roses significantly influences consumer purchasing decisions, prompting Japan to establish a quality guarantee system to ensure cut flower longevity. However, existing longevity guarantee methods rely heavily on subjective evaluations, overlooking critical factors such as senescence characteristics and disease infections. Hyperspectral imaging (HSI) technique is used for quality control of many fruits and can be performed at video rates, and could thus provide accurate data on aspects of cut flower quality. The You Only Look Once version 8 (YOLOv8) and Random Forest models for object detection and classification enable consistent quality assessment and swift longevity prediction. In this study, we developed a non-contact and rapid detection technique for the potential longevity of cut roses using deep learning techniques based on HSI data. Cut ‘Unforgettable’, ‘Egg Tart’, and ‘Catalina’ rose flowers were held in wet conditions during the exportation to Japan. HSI data within the visible near-infrared range 450-900 nm wavelengths were obtained for analysis of the disease infection and quality of cut roses. Image data of diseased cut roses were collected and corresponding data processing was carried out to build diseased cut roses and quality detection dataset. We developed the longevity prediction model based on scoring a grading standard on the flower quality and this model was then used to predict the longevity and evaluate quality changes of cut roses after exporting to Japan. The results showed that the longevity of exported cut roses was 8 d (‘Egg Tart’), 5.9 d (‘Catalina’), and 4.9 d (‘Unforgettable’). The longevity of cut roses was primarily terminated by gray mold disease (‘Unforgettable’ and ‘Catalina’), petal wilting and discoloration (‘Egg Tart’ and ‘Catalina’), and petal abscission (‘Catalina’). The predictive accuracy of the three rose flowers longevity prediction model was three rose flowers ‘Egg Tart’ (r2=0.80), ‘Unforgettable’ (r2=0.78), and ‘Catalina’ (r2=0.65). These results demonstrate that the combination of HSI and deep learning is a reliable method for evaluating the longevity of exported cut roses.
