融合注意力机制与迁移学习的宫颈细胞图像多分类优化研究OA
Optimization research on multi-classification of cervical cell images integrating attention mechanism and transfer learning
针对相近类别宫颈细胞图像的形态特征相似性较高,数据集规模有限导致模型泛化能力不足的问题,本研究提出了一种改进的 Resnet34 模型,用于宫颈细胞图像多分类任务.首先,在 Resnet34 的残差块中融合注意力模块,并将激活函数ReLU 替换为 PReLU,提升模型对细节的捕捉能力和适应性;其次,采用迁移学习策略,提升模型的泛化性能.结果显示,改进后的模型准确率达 96.78%,相比原始 Resnet34 提升了 1.32%.本研究模型能有效捕捉相近类别宫颈细胞图像的细微差异特征,优化模型的识别性能与泛化能力,可为宫颈癌的早期自动化筛查提供可靠的技术支撑.
Aiming at the problems of high similarity of morphological features of similar categories of cervical cell images and insuf-ficient generalization ability of the model caused by the limited size of the dataset,we proposed an improved Resnet34 model for the multi-classification task of cervical cell images.Firstly,an attention module was introduced into the residual blocks of Resnet34,and the activation function ReLU was replaced with PReLU,thereby enhancing the model's ability to capture details and its adaptability.Secondly,the transfer learning strategy was adopted to enhance the generalization performance of the model.The result showed that the accuracy of the improved model reached 96.78%,was 1.32%higher than that of the original Resnet34.The research can effectively capture the subtle differences in the image morphology of cervical cells belonging to similar categories,optimize the recognition perform-ance and generalization ability of the model,can provide reliable technical support for the early automated screening of cervical cancer.
杨晨;胡晓鹏
西南交通大学 计算机与人工智能学院,成都 611756||万锦医工(沈阳)科技有限公司,沈阳 110000西南交通大学 计算机与人工智能学院,成都 611756
医药卫生
多分类宫颈细胞图像Resnet34注意力机制迁移学习
Multi-classificationCervical cell imagesResnet34Attention mechanismTransfer learning
《生物医学工程研究》 2026 (3)
197-203,7
评论