基于ResNet50模糊称重装置双目识别系统的设计  

Design of Binocular Identification System Based on ResNet50 Fuzzy Weighing Device

作　　者：赵亚文 范剑红 陈金国[2] 涂志松 佘栋梁 徐志勇 ZHAO Yawen;FAN Jianhong;CHEN Jinguo;TU Zhisong;SHE Dongliang;XU Zhiyong(School of Engineering Industry,Putian University,Putian 351100,China;School of Mechanical and Information Engineering,Putian University,Putian 351100,China;Putian Chengxiang District Chengwei Food Co.,Ltd.,Putian 351100,China)

机构地区：[1]莆田学院新工科产业学院,福建莆田351100 [2]莆田学院机电与信息工程学院,福建莆田351100 [3]莆田市城厢区诚味食品有限公司,福建莆田351100

出　　处：《四川轻化工大学学报（自然科学版）》2024年第5期69-77,共9页Journal of Sichuan University of Science & Engineering(Natural Science Edition)

基　　金：福建省自然科学基金面上项目(2020J01918);莆田市科技计划项目(2021G2001ptxy06);福建省中青年教师教育科研项目(JAT220297)。

摘　　要：针对肉类食品行业生产效率低,以及采用人工称重分拣的问题,设计了一种基于ResNet50的模糊称重装置双目识别系统,旨在实现称重设备的文字实时识别,提高后期分拣率。首先,设计一款模糊称重装置,用于对食品进行有效的定量称重;其次,选用双目相机作为称重装置的识别硬件,并进行标定;再次,设定称重食盒采集实验,并对食盒图像进行预处理;然后,以ResNet50前5层的残差网络为核心,利用CTPN算法对食盒图像进行模型训练,用LSTM神经网络、NMS算法对待测试图像数据特征序列进行预测;最后,利用LabVIEW软件设计文字实时监测界面,进行系统监测。经测试,系统网络的Precision、Recall、F-measure值分别达到98.64%、97.60%、98.12%,系统文字识别率得到有效提高,且能够进行实时监测。To address the issues of low production efficiency and the reliance on manual weighing and sorting in the meat industry,a binocular recognition system based on a fuzzy weighing device and ResNet50 is designed.The aim is to achieve real-time text recognition on weighing equipment and improve the subsequent sorting rate.First,a fuzzy weighing device is designed for effective quantitative weighing of food products.Next,a binocular camera is selected as the recognition hardware for the weighing device and calibrated.Then,an experiment is set up to collect food box images,and the images are preprocessed.The core of the system uses the first five layers of the ResNet50 residual network,employing the CTPN algorithm for model training on food box images,and the LSTM neural network along with the NMS algorithm to predict the feature sequence of the test image data.Finally,a real-time text monitoring interface is designed using LabVIEW software for system monitoring.Testing shows that the system achieves a Precision of 98.64%,a Recall of 97.60%,and an F-measure of 98.12%,respectively.Consequently,the text recognition rate of the system has been significantly improved,enabling effective real-time monitoring.

关 键 词：称重装置 双目识别 食盒图像 ResNet50 文字识别率 

分 类 号：TH165[机械工程—机械制造及自动化] TP391.41[自动化与计算机技术—计算机应用技术]