Droplet Segmentation Method Based on Improved U-Net Network

Accurate segmentation of droplet images is an important part of high-precision contact angle measurement. For the inaccurate targets, incomplete contours, and solid-liquid-vapor intersections and boundary details that exist in the process of droplet segmentation In this paper, a neural network model suitable for droplet segmentation is proposed. The model is based on the U-Net network, and a 1×1 convolution layer is added to its input to summarize image features to avoid losing information from the initial image; And the Resnet18 structure is used as the feature learning encoder of U-Net, which enhances the expressive ability of the network and promotes the propagation of gradients. In the decoding process, the feature fusion technology of dense connection is introduced to improve the detailed information of the segmentation target and reduce the network. parameters. Finally, a batch normalization operation is added after each convolutional layer, which further optimizes the network performance. The experimen