基于改进ResNet34网络的脑肿瘤分类方法研究
2025,33(2):184-191
摘要：传统的人工分辩脑肿瘤医学影像费时费力、浅层的卷积神经网络模型分类准确率较低,为了提高脑肿瘤图像分类的高效性和准确率,对深度残差网络进行了研究,提出了一种基于改进ResNet34网络的脑肿瘤分类模型；该模型以ResNet34残差网络作为骨干网络,结合多尺度特征提取思想,采用多尺度输入模块作为ResNet34网络的第一层,将inception v2模块作为残差下采样层；再通过通道注意力机制模块,从通道域的角度赋予图像不同通道不同的权重,得到更重要的特征信息；经过五折交叉实验后的结果表明,改进后的新网络模型的分类准确率约为98.82%,比ResNet34提升约1.1%,且模型参数数量仅为原模型的80%；这说明改进后的网络不仅提高了准确率,还减少了模型复杂度,达到了参数更少,准确率更高的检测效果。
关键词：图像处理;改进ResNet34;脑肿瘤分类;多尺度输入;注意力机制;inception v2
Research on brain tumor classification method based on improved ResNet34 network
Abstract：In order to improve the efficiency and accuracy of traditional brain tumor image classification, a brain tumor classification model based on an improved ResNet34 network is proposed. Firstly, the model uses ResNet34 residual network as the backbone network to alleviate the gradient vanishing and explosion problems caused by deep convolution. Secondly, the multi-scale feature extraction idea is integrated into the first layer and residual downsampling module of the ResNet34 network, and image information of different receptive fields is obtained through convolutional kernels of different sizes. Finally, add the channel attention mechanism module to assign different weights to different channels of the image from the perspective of channel domain, in order to obtain more important feature information. The experimental results of 5-fold cross-validation show that the classification accuracy of the improved new network model is about 98.82%, which is 1.1% higher than ResNet34, and the number of parameters is only 80% of the original model. This indicates that the improved network not only improves accuracy but also reduces model complexity, achieving a detection effect with fewer parameters and higher accuracy.
Key words：image processing; improved ResNet34; brain tumors classification; multiscale inputs; attention mechanism; inception v2
收稿日期：2023-12-11
基金项目：
