杨 彪^*，张伊涵，李亚文

（商洛学院电子信息与电气工程学院，商洛市人工智能研究中心，商洛726000）

摘要：为了实现对不同动物肉快速、低成本识别，提出了一种基于深度学习和显微图像技术的肉类识别系统。利用微型手机显微镜和安卓手机，采集了猪肉、牛肉、羊肉、鸭肉、鸡肉等5种肉类的显微图像5052幅，然后按照1：10比例对其进行数据扩增，共得到55583张肉的显微图像。将其分为训练集、测试集和验证集。采用MobileNetV3深度学习算法建立了基于显微图像技术的肉类识别模型并将其部署到安卓平台。用验证集图像对模型的精度进行验证。结果表明，模型在验证集的准确率为99.98%，能够准确地识别5种肉类，为肉的品种识别提供一个快速、低成本的新技术手段。

关键词：深度学习；肉掺假；显微图像；安卓平台；MobileNetV3

中图分类号：TP183 文献标识码：A 文章编号：1674-506X（2025）04-0030-0007

A Meat Type Recognition System Based on MobileNetV3 and Microscopic Imaging Technology

YANG Biao^*，ZHANG Yihan，LI Yawen

(College of Electronic Information and Electrical Engineering，Shangluo University，Shangluo Artificial Intelligence Research Center，Shangluo 726000，China)

Abstract: To enable rapid and low-cost identification of different types of meat， a meat recognition system based on deep learning and microscopic imaging technology was developed. Using a smartphone- compatible miniature microscope and an Android phone，a total of 5 052 microscopic images were collected from 5 types of meat： pork， beef， mutton， duck， and chicken， then the images were expanded with data augmentation at a ratio of 1∶10，a total of 55583 microscopic images were obtained. These images were then divided into training，test， and validation sets. A meat recognition model based on microscopic imaging technology was constructed using the MobileNetV3 deep learning algorithm and integrated into the Android platform. The validation set images were used to verify the accuracy of the model. The result showed the model had an accuracy of 99.98% in validation set，could accurately distinguish among the 5 types of meat，offering a new technical solution for rapid and low-cost meat type recognition.

Keywords: deep learning；meat adulteration；microscopic imaging；Android platform；MobileNetV3

doi: 10.3969/j.issn.1674-506X.2025.04-005