首页|期刊导航|电子科技|基于混合卷积的糖尿病视网膜病变分类

基于混合卷积的糖尿病视网膜病变分类OA

Classification of Diabetic Retinopathy Based on Hybrid Convolution

中文摘要英文摘要

糖尿病视网膜病变是糖尿病患者常见的并发症之一,若不及时进行诊断和治疗,可能导致患者失明.为了更有效地捕捉图像细节以及提高特征提取能力和分类准确率,文中提出了一种基于混合卷积和注意力机制的糖尿病视网膜病变图像分类网络.基于轻量化网络DenseNet121优化所提网络,通过融合混合卷积和注意力模块提升分类精度和特征提取能力.经Inception模块处理图像,多尺度卷积用于提取不同尺度的特征信息,从而增强对视网膜图像细节的捕捉.将提取的特征输入至包含混合卷积的密集块中,优化了梯度流动性和特征传递效率.通过通道注意力和空间注意力机制动态调整特征,确保对关键病灶区域的聚焦,提取到的有效特征用于对糖尿病视网膜病变图像分类.实验结果表明,所提方法在公开数据集的分类准确率和灵敏度优于其他模型,证明了其在实际应用中具有潜在优势,有利于临床糖尿病视网膜病变的初步筛查.

Diabetic retinopathy is one of the common complications for diabetic patients.If not diagnosed and treated in time,it may lead to blindness in patients.In order to capture image details more effectively and improve the feature extraction ability and classification accuracy,a diabetic retinopathy image classification network based on a hybrid convolution and attention mechanism is proposed.The proposed network is optimized based on the light-weight network DenseNet121,and the classification accuracy and feature extraction ability are improved by fusing the hybrid convolution and attention modules.The image is processed by the inception module,and multi-scale convolution is used to extract feature information of different scales,thereby enhancing the capture of details in reti-nal images.The extracted features are input into a dense block containing mixed convolution,optimizing the gradi-ent fluidity and feature transfer efficiency.The features are dynamically adjusted through the channel attention and spatial attention mechanisms to ensure the focus on the key lesion areas,and the extracted effective features are used for the classification of diabetic retinopathy images.The experimental results show that the classification accu-racy and sensitivity of the proposed method on the public dataset are superior to those of other models,which proves that it has potential advantages in practical applications and is conducive to the preliminary screening of clinical dia-betic retinopathy.

张小宝;朱金荣;王颖;李朝越

扬州大学 物理科学与技术学院,江苏 扬州 225009扬州大学 物理科学与技术学院,江苏 扬州 225009扬州大学 物理科学与技术学院,江苏 扬州 225009扬州大学 物理科学与技术学院,江苏 扬州 225009

信息技术与安全科学

糖尿病视网膜病变混合卷积注意力机制图像分类DenseNet121轻量化网络

diabetic retinopathyhybrid convolutionattention mechanismimage classificationDenseNet121lightweight network

《电子科技》 2026 (10)

63-70,8

国家自然科学基金(62375234)江苏省研究生研究与实践创新计划(KYCX24_3714) National Natural Science Foundation of China(62375234)Jiangsu Graduate Innovation Program(KYCX24_3714)

10.16180/j.cnki.issn1007-7820.2026.10.007

评论