基于轻量型卷积神经网络的菜品图像识别

Dishes Image Recognition based on Lightweight Convolutional Neural Network

使用卷积神经网络分析研究识别菜品,能够帮助人们了解食物,根据不同的需求选择适合的菜品;同时也能被使用在自助餐厅结算系统中,提高结算效率。由于卷积神经网络有大量的卷积计算,大量参数致使卷积模型体积庞大,不利于将模型嵌入移动设备中,因此设计了一种轻量型卷积神经网络MobileNetV2-pro分类菜品。通过引入通道混洗、注意力机制提高网络的检测能力;利用随机擦除等图像预处理技术对菜品图像进行处理,提高系统的泛化能力。实验结果表明,该新结构网络能显著提高菜品分类准确率。

Convolutional neural network can be used to analyze and recognize dishes,helping people know about food and choose suitable dishes according to different needs.At the same time,it can also be used in cafeteria settlement system to improve settlement efficiency.A large number of convolution calculations and parameters in the convolutional neural network make the convolution model bulky,which is not conducive to embedding the model in a mobile device.This paper proposes to design a lightweight convolutional neural network MobileNetV2-pro to classify dishes.Channel shuffling and attention mechanism are introduced to improve the detection ability of the network.Image preprocessing techniques such as random erasure are used to process the image of dishes to improve the generalization ability of the system.Experimental results show that the new structure network can significantly improve the accuracy of dish classification.