| 基于ResGRUU Net网络的图像分割算法 |
 点此下载全文 |
| 引用本文：丁璇.基于ResGRUU Net网络的图像分割算法[J].计算技术与自动化,2025,(2):128-133 |
| 摘要点击次数: 895 |
| 全文下载次数: 743 |
|
|
| 中文摘要:为提高密集图像分割训练效率，在深度残差网络、GRU、U Net等网络模型基础上，设计了一种改进的ResGRUU Net完成密集图像分割任务。首先，ResGRUU Net中卷积块由两个或三个ResGRU块组成，可通过多条连通通路从原始信息中提取出不同的特征，从而记忆和消化先前特征中包含的规律；其次为提高数据使用效率，采用随机切片和Mosaic切片方法对数据进行扩增；最后，分别在DRIVE、STARE和CHASE_DB1测试数据集上对训练模型进行评估。通过仿真分析，改进后的模型在测试数据集上的准确率分别提升了0.28%、0.17%和0.46%，表明改进后的模型相较于改进前的模型在精度和训练速度上有所提高。 |
| 中文关键词:图像分割 深度学习 残差网络 门控循环单元 数据增强 |
| |
| Image Segmentation Algorithm Based on ResGRUU Net |
|
|
| Abstract:In order to improve the training efficiency of dense image segmentation, this paper designs an improved ResGRUU Net based on the research of ResNet, GRU、U net and other network models. Firstly, the convolution block in resgruu net is composed of two or three ResGRU blocks, which can extract different features from the original information through multiple connected paths, so as to memorize and digest the rules contained in the previous features; secondly, in order to improve the efficiency of data utilization, this paper uses random slicing and Mosaic slicing methods to amplify the data; finally, the training model was evaluated in data sets of DRIVE, STARE and CHASE_DB1, respectively. Through the simulation analysis, the accuracy rate of the improved model on the test data set is increased by 0.28%, 0.17% and 0.46%, respectively, which shows that the improved model has higher accuracy and training time than the original model. |
| keywords:image segmentation deep learning residual network gated recursive unit data enhancement |
| 查看全文 查看/发表评论 下载pdf阅读器 |
|
|
|
位访问者,今日访问
