Potato Bud Eyes Recognition Method Based on Improved Mask R-CNN
The accurate recognition of potato bud eyes stands as the prerequisite for the automatic cutting of potato seed potatoes.To improve the recognition effect of potato bud eyes and solve the long-existing problem of low efficiency of the manual cutting,a potato bud eye recognition method based on the improved Mask R-CNN was put forward.With the optimization of the RoIAlign operation by the deterministic rearrangement technology,the uncertainty of the interpolation operation was eliminated,and the uniqueness of the mapping of candidate regions of different sizes on the feature map was guaranteed,improving the accuracy and stability of target classification and segmentation.The test results indicated that the enhanced model had a recognition accuracy rate of 98.47%,a recall rate of 96.99%,a harmonic mean F1 of 97.72%,and an average recognition time of a single image of 0.135 seconds.Compared with the algorithm before improvement,the recognition accuracy,recall rate,and F1 value had increased by 6.46,12.01,and 9.36 percentage points respectively,and the average recognition time of a single image had increased by 0.004 seconds.The improved Mask R-CNN algorithm could better acclimate to the special shape of potato bud eyes and different environment factors and identify potato bud eyes,paving a solid ground for the research of potato seed potato smart cutting devices.
万方数据
维普
