2025年03期 v.46 30-36页
王锦俊1 蔡延光1,2
（1.广东工业大学自动化学院，广东 广州 510006
2.广州理工学院人工智能学院，广东 广州 510540）
摘要：为了提高乳腺图像的识别准确率，提出一种改进ResNet34模型的乳腺图像识别方法。该方法以ResNet34模型为基础，引入平行注意力残差块（PARB）模块来增强乳腺图像通道间的关联性，进一步提取乳腺图像的重要特征信息，从而提高模型的识别准确率；利用科尔莫戈洛夫-阿诺尔德网络（KAN）替代传统的多层感知器（MLP），减少模型参数，提高模型的识别速度。实验结果表明，改进的ResNet34模型比ResNet34模型的准确率、精确率、召回率和F1-Score分别提升了4.0%、0.6%、8.0%和4.7%，表明该方法对乳腺图像具有更好的识别效果。
关键词：乳腺图像识别；ResNet34；平行注意力残差块；科尔莫戈洛夫-阿诺尔德网络
中图分类号：TP391.41; TP183 文献标志码：A 文章编号：1674-2605(2025)03-0005-07
DOI：10.12475/aie.20250305 开放获取
An Improved ResNet34 Model for Mammographic Image Recognition Method
WANG Jinjun1 CAI Yanguang1,2
(1.College of Automation, Guangdong University of Technology, Guangzhou 510006, China
2.School of Artificial Intelligence, Guangzhou Institute of Science and Technology, Guangzhou 510540, China)
Abstract: To enhance the recognition accuracy of mammographic images, an improved ResNet34 model for mammographic image recognition method is proposed. Building upon the ResNet34 model, this method introduces a parallel attention residual block (PARB) module to strengthen inter-channel correlations in mammographic images, further extracting critical feature information to improve recognition accuracy. Additionally, it replaces the traditional multilayer perceptron (MLP) with Kolmogorov-Arnold networks (KAN) to reduce model parameters and increase recognition speed. Experimental results demonstrate that the improved ResNet34 model achieves enhancements of 4.0%, 0.6%, 8.0%, and 4.7% in accuracy, precision, recall, and F1-Score respectively compared to the original ResNet34 model, indicating superior recognition performance for mammographic images.
Keywords: mammographic image recognition; ResNet34; parallel attention residual block; Kolmogorov-Arnold networks
