基于注意力机制结合残差神经网络的胃癌图像预测方法  

Gastric Cancer Image Prediction Method Based on Attention Mechanism and Residual Neural Network

作　　者：郑子龙 曾安[1] ZHENG Zi-long;ZENG An(School of Computer Science,Guangdong University of Technology,Guangzhou 510006 China)

机构地区：[1]广东工业大学计算机学院,广东广州510006

出　　处：《自动化技术与应用》2025年第2期136-139,184,共5页Techniques of Automation and Applications

基　　金：广东省省级科技计划项目(2019A050510041)。

摘　　要：随着深度学习的发展,单图像超分辨率技术对医学图像的分析成为研究热点。传统方法对胃癌图像细节重建不充分,丢失了图像的细节信息。提出一种新型高效的注意力机制结合残差神经网络的胃癌图像预测方法。对编码器输出的特征进行细化,并根据改进的融合注意方法对空间注意模块和通道注意模块进行重组,为像素融合提供更好的输入特征,达到重建解码器的效果。最后,通过添加像素融合后的输入特征,从而加速网络高频残差的收敛,提高图像重建效果。在重建放大倍数为2、3、4倍的条件下,进行实验验证。实验结果表明,在较低的模型参数下,方法具有最佳的PSNR和SSIM值。With the development of deep learning,the analysis of medical images by single image super-resolution technology becomes a re-search hotspot.The traditional method is not enough to reconstruct the details of gastric cancer image,and the details of the im-age are lost.This paper presents a new and efficient method of gastric cancer image prediction based on attention mechanism and residual neural network.The output features of the encoder are refined,and the spatial attention module and channel attention module are reorganized according to the improved fusion attention method to provide better input features for pixel fusion,so as to achieve the effect of reconstructing the decoder.Finally,the input features after pixel fusion are added to accelerate the conver-gence of network high-frequency residuals and improve the image reconstruction effect.Experimental verification is carried out under the condition of reconstruction magnification of 2,3 and 4 times.The experimental results show that the method has the best PSNR and SSIM values at lower model parameters.

关 键 词：超分辨率重建 注意力模块 像素融合 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]