Chromosome instance segmentation network based on regression correction

The invention provides a chromosome instance segmentation network based on regression correction for solving the problem that the segmentation precision of chromosome instances in a real chromosome data set is low, and high-precision segmentation of chromosomes can be realized. The network corrects the classification confidence through the output of the regression branch and the mask branch to obtain a higher association with the positioning precision and the segmentation precision. A non-maximum suppression algorithm based on masks is provided, and missing segmentation and wrong segmentation of chromosomes are effectively prevented. A K-IoU loss function is provided, and more effective and reasonable loss is provided for an error segmentation region. Experimental results show that the method can effectively improve the accuracy of chromosome instance segmentation. 本发明针对真实染色体数据集中染色体实例分割精度较低等问题提出了一种基于回归修正的染色体实例分割网络，可实现染色体的高精度分割。该网络通过回归分支和掩码分支的输出修正分类置信度，以获得与定位精度和分割精度更高的关联。提出了一种基于掩码的非极大抑制算法，有效防止了染色体的遗漏分割和错误分割。提出