一种基于对抗网络和自注意力机制的无监督语义分割方法
By generating images using adversarial networks and extracting features using a self-attention mechanism, the problems of unstable pre-segmentation results and inaccurate global feature extraction in unsupervised semantic segmentation are solved, achieving more accurate and stable image segmentation.
CN115346045BActive Publication Date: 2026-07-17BEIJING UNIV OF TECH
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNIV OF TECH
- Filing Date
- 2022-07-11
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing unsupervised semantic segmentation methods are sensitive to and unstable in pre-segmentation results, and CNNs are inaccurate in global feature extraction, which affects the accuracy of image segmentation.
Method used
Image data is generated using an adversarial network and local and global features are extracted using a self-attention mechanism. The neural network is then optimized using a mutual information loss function to achieve unsupervised segmentation.
Benefits of technology
It improves the accuracy and stability of image segmentation and enhances the performance of unsupervised semantic segmentation.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115346045B_ABST
Abstract
一种基于对抗网络和自注意力机制的无监督语义分割算法涉及人工智能、自动驾驶领域，实现对自动驾驶图像的准确分割，其包括以下步骤：步骤1、获得自动驾驶图像数据；步骤2、利用对抗生成网络得到原始图像的生成图像；步骤3、利用超像素图像分割算法对原始图像进行预分割；步骤4、将图像输入局部特征提取网络获得图像的局部特征；步骤5、将图像输入全局特征提取网络获得图像的全局特征；步骤6、将局部特征和全局特征相加，并经过一层卷积层和Softmax函数层得到图像的初步分割结果；步骤7、计算预分割结果与原始图像分割结果之间的互信息以及预分割结果与生成图像分割结果之间的互信息；步骤8、采用梯度下降法对分割模型进行训练得到图像分割结果。
Need to check novelty before this filing date? Find Prior Art