汉译英：Liu等人在UNet网络架构中加入了扩张卷积模块实现了语义特征的多尺度编码可以减少原始UNet网络训练中梯度消失的问题绵羊后腿图像分割的精度达到09820每个样本的检测时间为82ms

Liu et al. added expansion convolution modules to the UNet network architecture, achieving multi-scale encoding of semantic features and reducing the problem of gradient disappearance during original UNet network training. The accuracy of sheep hind leg image segmentation reached 0.9820, with a detection time of 82ms per sample.