为了准确测量纳米颗粒的尺寸,依据透射电子显微镜拍摄的纳米颗粒图像,提出了一种基于U-Net卷积神经网络的颗粒自动分割方法。将U-Net部分网络结构与批量归一化层相结合,减弱了网络对初始化的依赖,提升了训练速度。对纳米颗粒图像进行半隐式偏微分方程滤波以增强图像边缘信息,利用改进的U-Net网络训练纳米颗粒个体分割模型,得到了分割结果。研究结果表明,所提方法能准确分割出图像中的纳米颗粒,对边缘模糊和强度不均的纳米颗粒的分割效果提升显著。

In order to accurately measure the size of nanoparticles, an automatic particle segmentation method based on U-Net convolutional neural network is proposed according to the nanoparticle images captured by the transmission electron microscopy. Combined with the Batch Normalization (BN) layer, it reduces the dependence of networks on initialization and thus speeds up training. The nanoparticle image is filtered by the semi-implicit partial differential equation to enhance the image edge information. The improved U-Net network is used to train the nanoparticle individual segmentation model and the segmentation result is obtained. The research results show that the proposed method can accurately segment the nanoparticles in the image, and the segmentation effect is especially obvious for the nanoparticles with edge blurs and uneven intensities.