【作者】 李冰； 李广庆； 吴少勇；

【Author】 Bing Li;Guangqing Li;Shaoyong Wu;School of Automation, Harbin University of Science and Technology;Heilongjiang Provincial Key Laboratory of Complex Intelligent System and Integration;

【机构】 哈尔滨理工大学自动化学院； 黑龙江省复杂智能系统与集成重点实验室；

【摘要】 深度学习在视网膜OCT图像分类领域中,存在原始图像不清晰、浅层网络提取的更多是一些边缘信息,提取特征能力弱,网络层数过深,在训练过程中容易导致梯度弥散等问题。针对这些问题,本文设计了一种通过融合不同尺度信息,进一步提升分类精度的网络,该网络融合卷积块注意力机制(Convolutional Block Attention Module,CBAM),使网络可以提取输入图像分布在通道和空间中的有效部分进行学习;融合多尺度残差(Multiscale Residual,MR)模块提升网络的捕获信息能力,对小目标病进行有效的特征提取。实验结果表明,本文所用方法对于视网膜OCT图像分类效果优于多数网络。更多还原

【Abstract】 In the field of retinal OCT image classification, depth learning has some problems,such as unclear original images, more edge information extracted by shallow networks, weak ability to extract features, too deep network layers, and gradient dispersion in the training process. To solve these problems, this paper designs a network that further improves the classification accuracy by fusing information of different scales. This network fuses the Convolutional Block Attention Module(CBAM) for the important information that the input information is distributed in the space and channel, so that the network can extract the effective part of the input image distributed in the channel and space for learning; The fusion of multi-scale residual(MR) module improves the network’s ability to capture information, and effectively produce the characteristics of small, objective diseases.Experimental results show that the method used in this paper is better than most networks for classifying the retinal OCT image.更多还原