Petal damage and bent flower detection method of rose cut flowers based on computer vision

Post-harvest quality grading plays a crucial role in further enhancing the market competitiveness of cut roses. One of the key tasks of post-harvest cut flower grading is defect detection. This study proposes a method for identifying rose cut flower damage based on deep learning technology. This method first sets up a cut flower image acquisition system and establishes a petal damage dataset and a bent head flower dataset. For petal damage, this study makes lightweight improvements to the YOLOv5 model and uses the improved YOLOv5 model to identify petal damage, achieving an APobj of 92.1 %. In comparison with other models, the improved YOLOv5 model also has excellent accuracy and fewer parameters. For bent flowers, this study makes lightweight improvements to the HRNet model and uses the improved HRNet to identify the position of the flower's center. The improved HRNet has a decrease in recognition accuracy, but the number of parameters is significantly reduced compared to the original model. After obtaining the position of the flower's center, it is judged whether it is a bent flower according to the best distance threshold obtained through the training set data. The average damage recognition accuracy of the final rose cut flowers is 97.9 %. In conclusion, the proposed method in this study can effectively identify petal damage and bend flower in cut roses, and it can also provide new ideas and technical means for the quality detection of cut roses.