基于深度学习的铸件CT图像分割算法

针对现有方法分割弱边缘铸件CT图像难度大、精度低、鲁棒性差的问题，提出一种融合残差模块与混合注意力机制的U型网络分割算法(AttRes-U-Nets)。该算法以U-Net网络为基础，首先构建深度残差网络ResNets作为算法的编码网络，解决传统U-Net网络特征提取能力不足的问题；然后，引入改进后的混合注意力机制，突出分割目标区域与通道的特征响应，提高网络灵敏度；最后，将Focal loss与Dice loss结合为一种新损失函数FD loss缓解样本不平衡带来的负面影响。使用120阀体数据集对算法性能进行验证，实验结果表明，本文算法对铸件分割的像素准确率(PA)和交互比(IoU)分别达到98.72%和97.40%,优于传统U-Net算法与其他主流语义分割算法，为弱边缘分割提供了新思路。

Casting CT image segmentation algorithm based on deep learning

The existing methods for segmenting CT images of castings with weak edges have problems of difficulty, low precision and poor robustness. To address these issues, this article proposes a U-shaped network segmentation algorithm that fuses residual module and mixed attention mechanism. Firstly, the algorithm is based on U-Net. The deep residual networks ( ResNets) is established as the backbone of the network to solve the inadequate feature extraction capability of the original U-Net. Then, the improved hybrid attention mechanism is introduced, and it characterize the target region and the channel to improve the network sensitivity. Finally, a new loss function (FD loss) combining Focal loss and Dice loss is used to mitigate the negative effects of sample imbalance. The performance of the algorithm is evaluated by using the 120 valve body dataset. The experimental results show that the pixel accuracy ( PA) and intersection over union (IoU) of the proposed algorithm for casting segmentation reach 98. 72% and 97. 40% , which are better than the those of the original U-Net and other mainstream semantic segmentation algorithms. This work provides a new idea for the weak edge segmentation problem.