Accurate segmentation of flake particles is prerequisite for particle size analysis.To solve the problem of difficulty in segmenting highly adhesive particles using existing segmentation algorithms

an UNet semantic segmentation network which integrates edge features

EB-UNet

is proposed.Aiming at the problem of low edge segmentation precision

an edge feature fusion network is paralleled in the encoder part of the UNet

to enhance the edge features of particles.Aiming at the problem of insufficient model representation ability

residual modules are used to replace ordinary convolutional modules in the UNet to increase the depth of the network.The experimental

results show that compared to the traditional UNet

the improved network has a mean accuracy(

A

C

)improve to 0.9665

a mean intersection over union(

R

IoU

)improve to 0.9473

and an F1 score (

F

1-Score

)of 0.8694.The segmentation performance of adhesive flake particles is significantly improved.