基于CNN的频域光学相干断层扫描图像多尺度特征融合方法  

A multi-scale feature fusion method for frequency-domain optical coherence tomography images based on CNN

作　　者：吕牡丹 LYU Mudan(Information Engineering College,Jiangxi University of Technology,Nanchang 330000,China)

机构地区：[1]江西科技学院信息工程学院,南昌330000

出　　处：《激光杂志》2024年第12期149-154,共6页Laser Journal

基　　金：江西省教育厅科学技术研究项目(No.GJJ2202608)。

摘　　要：在频域光学相干断层扫描图像处理过程中,主要采用递归神经网络实现特征融合,运算过程中存在着梯度消失情况,使得多尺度特征融合结果的MAP(平均精度)较低。因此,提出基于CNN(卷积神经网络)的频域光学相干断层扫描图像多尺度特征融合方法。以生成对抗网络为核心建立扫描图像去噪所需的网络架构,通过原始图像域生成不包含噪声信息的高质量扫描图像。运用离散小波变换原理,将去噪后处理后的图像分解为多个子图像,通过构造灰度梯度共生矩阵,提取多尺度图像特征向量。从图像局部对比度和全局对比度入手,计算出图像自适应调节系数,以此来实现图像细节特征增强处理。最后,依托于卷积神经网络构建特征融合模型,通过对增强特征的匹配分析和衔接处理,得到多尺度特征融合结果。实验结果表明:新研究方法应用后,频域光学相干断层扫描图像多尺度特征融合结果的MAP值高于0.8,证明了其可以实现不同尺度特征的有效融合。In the process of image processing in frequency-domain optical coherence tomography,recursive neural networks are mainly used to achieve feature fusion.During the operation process,there is a gradient vanishing situation,which leads to low MAP(average accuracy)of multi-scale feature fusion results.Therefore,a multi-scale feature fusion method for frequency domain optical coherence tomography images based on CNN(Convolutional Neural Network)is proposed.Establish a network architecture for denoising scanned images based on generative adversarial networks,and generate high-quality scanned images without noise information through the original image domain.Using the principle of discrete wavelet transform,the denoised image is decomposed into multiple sub images.By constructing a grayscale gradient co-occurrence matrix,multi-scale image feature vectors are extracted.Starting from the local and global contrast of the image,calculate the image adaptive adjustment coefficient to achieve image detail feature enhancement processing.Finally,a feature fusion model is constructed based on convolutional neural networks,and multi-scale feature fusion results are obtained through matching analysis and concatenation processing of enhanced features.The experimental results show that after the application of the new research method,the MAP value of the multi-scale feature fusion results of frequency domain optical coherence tomography images is higher than 0.8,proving that it can effectively fuse features of different scales.

关 键 词：CNN 频域光学相干断层扫描 特征提取 小波变换 自适应增强 特征融合 

分 类 号：TN391[电子电信—物理电子学]