基于CBAM-InceptionV3迁移学习的食品图像分类

为提高食品图像自动识别分类的准确率,提出一种嵌入通道注意力机制和空间注意力机制的卷积块注意力模块(CBAM)的“开端”第三版(Inception V3)分类模型。将带有图像网络(ImageNet)预训练权重参数的InceptionV3模型拆分后,在每个Inception块后嵌入CBAM模块,再重新组装成新模型,共嵌入11个CBAM模块。将此模型用于经过填充和缩放到299×299像素的Food-101食品图像数据集进行迁移学习,最高准确率达到82.01%。与原始的InceptionV3模型相比,CBAM模块能够有效提升模型的特征提取和分类能力;同时迁移学习与从头开始训练相比也可以大幅提高准确率、缩短训练时间。与其它几类主流卷积神经网络模型进行对比实验,结果表明该模型具有较高的识别准确率,可为食品图像分类识别提供有力支撑。

Research on the Application of Wireless Network in Collecting Road Traffic Information

Due to the characteristics of variability and dispersion in traffic information, to get the reliable real-time traffic information has been a bottleneck in the development of intelligent transportation systems. However, with the development of wireless network technology and mobile Internet, the mobile phones are rapidly developed and more popular, so it is possible to get road traffic information by locating the mobile phones in vehicles. The system structure for the road traffic information collection is designed based on wireless network and mobile phones in vehicles, and the vehicle recognition and its information computation methods are given and discussed. Also the simulation is done for vehicle recognition and computation based on fuzzy cluster analysis method and the results are obtained and analyzed.