基于多孔卷积神经网络的图像空间结构信息细节表征  

Detailed Characterization of Image Spatial Structure Information Based on Porous Convolutional Neural Network

作　　者：徐叶军[1] XU Yejun(College of Biotechnology,Suzhou Industrial Park Institute of Service Outsourcing,Suzhou Jiangsu 215123,China)

机构地区：[1]苏州工业园区服务外包职业学院生物科技学院,江苏苏州215123

出　　处：《盐城工学院学报（自然科学版）》2024年第1期20-25,共6页Journal of Yancheng Institute of Technology:Natural Science Edition

摘　　要：针对传统图像空间结构信息表征方法存在细节表征模糊度较高、信息训练损失较高等问题,提出一种新的基于多孔卷积神经网络的图像空间结构信息细节表征方法。该方法通过图像空间结构信息细节相似性度量,并以图像的形状、颜色和纹理特征对图像空间结构信息细节进行编码,再去除图像冗余信息,利用多孔卷积神经网络对图像空间结构的深度信息进行融合,从而完成图像空间结构信息的细节表征。实验结果表明,基于多孔卷积神经网络的图像空间结构信息细节表征方法在模糊度、训练损失、图像相似性等方面都比传统的3种方法优越,能够清晰地表征图像空间结构信息。Aiming at the problems of high detail characterization ambiguity and high information training loss in traditional image spatial structure information characterization methods,a new image spatial structure information characterization method based on porous convolutional neural network is proposed.This method measures the similarity of image spatial structure information details,encodes the image spatial structure information details based on the shape,color and texture features of the image,then removes the redundant information of the image,and uses the porous convolutional neural network to fuse the depth information of the image spatial structure information,thus completing the detailed characterization of the image spatial structure information.The experimental results show that the detailed characterizationmethod based on porous convolutional neural network is superior to the traditional three methods in terms of ambiguity,training loss and image similarity,and can clearly characterize the spatial structure information of image.

关 键 词：多孔卷积神经网络 图像空间结构 细节表征 冗余信息 深度信息融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]