基于改进ResNet模型的图像分类方法

Image Classification Method Based on Improved ResNet Model

卷积神经网络是一种具有卷积结构的深层前馈神经网络模型,被广泛地应用于图像分类等重要领域。针对原始ResNet网络提取特征能力不足的问题,提出一种基于改进ResNet模型的图像分类方法。将现有的SE通道注意力机制,嵌入到原始ResNet网络中每个残差结构的末端,进行跨信道的信息交互,捕捉更显著的通道或像素信息。在CIFAR-100和CIFAR-10数据集上进行大量的实验表明,相对于原始的ResNet网络,Top-1 Error和Top-5 Error下降明显。

The convolutional neural network is a deep feed-forward neural network model with convolutional structure,which is widely used in important fields such as image classification.In view of the problem of insufficient feature extraction capability of the original ResNet network,an image classification method based on improved ResNet model is proposed.The existing SE channel attention mechanism is embedded at the end of each residual structure in the original ResNet network to perform cross-channel information interaction and capture more significant channel or pixel information.Extensive experiments performed on the CIFAR-100 and CIFAR-10 datasets show that the Top-1 Error and Top-5 Error decrease significantly relative to the original ResNet network.