计算机科学
分割
背景(考古学)
人工智能
光学(聚焦)
图像融合
图像分割
特征提取
图像(数学)
频道(广播)
数据挖掘
特征(语言学)
钥匙(锁)
计算机视觉
领域(数学)
模式识别(心理学)
比例(比率)
信息抽取
信息融合
上下文模型
图像压缩
相互信息
图层(电子)
信息模型
空间分析
压缩(物理)
空格(标点符号)
构造(python库)
人工神经网络
数据压缩
代表(政治)
融合
接头(建筑物)
图像配准
深度学习
领域(数学分析)
作者
Luyao Ren,Wenxin Yu,Zhiqiang Zhang,Chang Liu,Jun Gong
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-06-30
卷期号:: 1-6
标识
DOI:10.1109/icme59968.2025.11210170
摘要
In the field of medical image segmentation, most models only pursue performance and pay less attention to the parameters. It is a challenging problem to achieve the balance between performance and parameters. To solve this problem, we propose a context-aware information fusion network. The model uses Depth Enhancement Attention(DEA) combined with SE at the bottom layer to extract local information. Subsequently, it uses Gated Focus Residual(GFR) to transform and convert local information into global information. Then, the information is passed through Scale Dynamic Gated Attention(SDGA) to capture global information after dilated convolution. Finally, Feature Compression Fusion(FCF) are used to complete upsampling. The model also uses cross-jump connections to integrate low-level details and high-level semantic information across multiple scales, followed by further extraction through Channel Space Parallel Attention(CSPA).In addition, the experimental results on the BUSI and ISIC2018 datasets show that compared with state-of-the-art(SOTA), our parameters are reduced by 4%, mIoU is increased by%1.69 on average, and DCS is increased by%0.96 on average, achieving the best performance.
科研通智能强力驱动
Strongly Powered by AbleSci AI