Fig. 6

Comparison of different models. a Comparison of YOLOv5 and Faster R-CNN. The YOLOv5 model has a higher recognition speed than Faster R-CNN, and the Faster R-CNN model has a higher detection accuracy than YOLOv5. b Comparison of with or without FPN (Feature Pyramid Networks) The mAP@0.5:0.95 of the improved model increased by 0.002, R² \({R}^{2}\) of "close" class increased by 0.003, and R² of the "open" class and "all" the decreased slightly. c Comparison of with or without data augmentation. The improved model has a slight decline in the number of R² in the open category and an improvement in other evaluation indicators. d Comparison of with or without data Multi-Scale. The results showed that the mAP@0.5:0.95 of the model was improved by 0.003 after Multi-Scale training. \({R}^{2}\) R² in the "open" and "close" categories fell by 0.0092 and 0.0007, respectively. R² \({R}^{2}\) in the "all" category increased to 0.0086. "open" and "close" represent dehiscent and indehiscent anthers, respectively