Automatic Segmentation of Stomata in Leaf Microscopic Images Using Box-Supervised Instance Segmentation
Deep learning techniques have been used for stomatal segmentation tasks.However,labeling of training data is a mechanical and time-consuming manual process,especially when the data set is relatively large.Therefore,this study proposes a stomatal segmentation method based on the weakly supervised model Boxlevelset.The feature extraction network ResNet50 of the original model is replaced with ResNest50,and the CBAM module is introduced in the feature extraction process.Taking black poplar stomata as the research object,this method can effectively segment the stomata,with an F1 score of 79.89%.The proposed method reduces the time required to annotate training data while ensuring segmentation accuracy,thereby significantly reducing the workload required to train the stomatal segmentation network.
NETL
NSTL
万方数据
维普
