一种基于树莓派的智能分类垃圾桶的设计
来源:用户上传
作者:金莉娟 刘栖昌 张许如 刘峻杰
摘 要:为解决使用普通垃圾桶r采用人工分拣垃圾造成垃圾分类不准确、效率低下的问题,设计了一款基于树莓派的智能分类垃圾桶。该垃圾桶的分类算法基于TensorFlow架构,采用全卷积网络(Fully Convolutional Networks, FCN)实现对垃圾图像特征的学习和识别,用于训练数据集以改进图像识别的准确率,并明确垃圾所属分类。实际应用时,利用传感器采集数据,利用摄像头识别物品,利用舵机带动投放口到正确的分类位置投放垃圾,并且语音播报当前的垃圾种类。本研究共收集了五类垃圾图像,每类图像训练34 组,每组150 次。实验结果表明,该智能垃圾桶的分类准确率可达到85%以上,具有较好的分类效果。
关键词:垃圾分类;树莓派;TensorFlow;FCN;深度学习
中图分类号:TP399 文献标识码:A
Design of a Intelligent Classification Garbage Can based on Raspberry Pi
JIN Lijuan, LIU Qichang, ZHANG Xuru, LIU Junjie
Abstract: Aiming at the problem of inaccurate garbage classification and low efficiency caused by manual classification of garbage when using ordinary garbage cans, this paper proposes to design an intelligent classification garbage can based on Raspberry Pi. Its classification algorithm is based on TensorFlow architecture, and FCN (Fully Convolutional Networks) is used to realize the learning and recognition of garbage image features, which is used to train the data set to improve the accuracy of image recognition and clarify the classification of garbage. In practical applications, sensors are used to collect data, cameras to identify items, steering gears to drive the discharge port to the correct classification position, and the current type of garbage is broadcast by voice. In this study, a total of 5 types of garbage images are collected. Each type of image is trained for 34 groups, 150 times in each group. The experimental results show that the classification accuracy of the intelligent garbage can reaches more than 85%, which has a good classification effect.
Keywords: garbage classification; Raspberry Pi; TensorFlow; FCN; deep learning
1 引言(Introduction)
随着社会的发展,传统的垃圾分类方式且仅限于手动开关垃圾桶已不能满足人们的生活需求[1]。为此,国内外开展了广泛研究,如美国的Transhbot和BigBelly、芬兰的Enevo,它们大多基于实时检测、自动报警、液晶显示屏进行设计,但是还未涉及垃圾自动分类[2];国内也进行了相关研究,主要分为太阳能垃圾桶与感应性垃圾桶,使得垃圾桶具有自动归类、报警和持续监测等功能,但还不能满足自动识别分类的需求[3]。
本文针对当前现状,设计了一个基于树莓派的智能分类垃圾桶。该垃圾分类系统能够自动、准确地判断出垃圾的类型,通过舵机带动投放口到正确位置,打开挡板,将垃圾投放到桶内,从而实现自动分类并可语音播报出垃圾的种类。此外,在垃圾桶满载时,会发出满载提醒。
2 系统结构(System structure)
垃圾分类系统主要由软件和硬件组成,软件的核心是分类算法,硬件的核心是树莓派。该系统的硬件组成如图1(a)所示,树莓派Pi4B开发板结构如图1(b)所示。
2.1 硬件结构
(1)选择树莓派Pi4B开发板作为核心硬件系统,该款产品较好地达到了系统的设计要求[4]。
(2)采用光电传感器作为检测垃圾容量的硬件系统。在垃圾容量达到75%以后启用喇叭工具进行满载提醒。光电传感器在一般情况下由三部分构成:发送器、接收器和检测电路。其基本原理是以光电效应为基础,将被测物体的变化量转换为光信号的变化,进而利用光电器件将非电信号转换成电信号。
(3)使用托盘暂存待投掷垃圾。使用托盘暂存垃圾时,该垃圾桶的识别分类系统也正在工作,待垃圾识别完成后,依据垃圾的识别结果完成分类。
nlc202206161137
转载注明来源:https://www.xzbu.com/1/view-15433764.htm