利用图像超分辨率提升交通标志分类精度研究

Research on Improving the Accuracy of Traffic Sign Classification Using Image Super-Resolution

为了解决利用卷积神经网络进行交通标志分类时精度低的问题,通过将图像超分辨率网络与分类网络相连接,提出一种超分级联网络结构。首先使用改进的双重注意力机制超分辨率网络作为级联网络的子网络;然后训练图像分类网络,用于对超分辨率处理后的图像进行分类;最后利用分类准确率衡量超分辨率重构对图像分类任务的有效性。模拟和真实交通标志数据集验证结果表明,经过超分辨率处理的图像在分类模型中均取得了更高的分类准确率,证明了超分辨率技术对于交通标志图像分类准确率的提高具有促进作用。

To solve the problem of low accuracy when classifying traffic signs using Convolutional Neural Networks(CNN), this article proposes a cascade super-resolution network structure by connecting an image super-resolution network to a classification network. A modified dual-attention mechanism super-resolution network is first used as a sub-network of the cascade network, and then the image classification network is trained for classifying the super-resolution processed images, and finally the classification accuracy is used to measure the effectiveness of super-resolution reconstruction for the image classification task. The validation results of both simulation and real traffic sign datasets show that the superresolution processed images achieve higher classification accuracy in the classification model, which proves that the superresolution technology has a facilitating effect on the improvement of the classification accuracy of traffic sign images.