基于YOLACT++的秀丽隐杆线虫显微图像分割方法

Segmentation Method of Caenorhabditis Elegans Microscopic ImageBased on YOLACT+ +

在分割网络提取特征的过程中,会存在低层特征缺失中缺失语义信息的现象,数据集中存在的一些噪声对分割给结果也会产生一定的干扰。为了提高分割网络对特征图的理解,解决在低层特征中缺失的语义信息,论文在YOLACT++的基础上引入了Attention机制,通过对数据的一些预处理,然后结合对损失函数的改进,对分割结果进行进一步的优化。由实验数据比较,论文的方法相比于传统的分割算法的MIOU提高了16%,相比于原始的YOLACT++的方法,该方法分割的MIOU提高了3.2%。

In the process of feature extraction from segmentation network,there will be the phenomenon of low-level feature missing and semantic information missing,and some noise in the data set will also produce some interference to the segmentation re-sults.In order to improve the understanding of feature graph in segmentation network and solve the semantic information missing in low-level features,this paper introduces the attention mechanism on the basis of YOLACT++.Through some data preprocessing,and then combined with the improvement of loss function,the segmentation result is further optimized.By comparing the experimen-tal data,the MIOU of this method is 16%higher than that of the traditional segmentation algorithm,and 3.2%higher than that of the original YOLACT++.