摘要
传统垃圾分类方法不再适用于物质丰富的时代。为提高垃圾分类效率、推动资源最大化利用、改善垃圾对环境的污染,开发垃圾分类应用软件十分必要。本文基于ResNext-101模型和端对端的语音合成系统(Tacotron),开发集目标识别与语音合成于一体的垃圾分类APP,并以50329张图片、基于《生活垃圾分类标志》共4个大类211个小类作为数据集进行训练和实验。运用ResNeXt-101算法构建基于目标识别的垃圾分类卷积神经网络模型,垃圾识别分类准确率高达80.14%。同时采用Tacotron神经网络模型和Griffin-Lim算法生成语音。该App设计用户操作友好,基本适合于现阶段垃圾分类的应用需求。基于图片的垃圾分类提醒不仅有利于提升日常生活中垃圾分类的准确度,而且在很大程度上提高了垃圾分类效率,具有科学的应用价值。
Traditional garbage classification methods are no longer applicable in the era of rich material life,the software is developed to promote efficiency in classification,the maximum utilization of resources and improve environmental protection.Based on the ResNext-101 model and the end-to-end speech synthesis system(Tacotron),this paper develops a garbage classification APP that integrates target recognition and speech synthesis.50329 pictures in 4 major classes and 21 subclasses are used as data set for experiments based on the"Domestic Garbage Classification Signs".Experimental results show that using the ResNeXt-101 algorithm to build a garbage classification convolutional neural network model based on target recognition,the accuracy of garbage recognition classification is as high as 80.14%,and the Tacotron neural network model and Griffin-Lim algorithm are used to generate speech.App design is user-friendly in actual use and meets the requirement of garbage classification at this stage.The picture-based garbage classification reminder not only helps to improve the accuracy of garbage classification in daily life but also improves the efficiency of garbage classification to a large extent,which has a high use-value.
作者
刘鉴建县
黄辛迪
王志辉
肖晓霞
LIU Jianjianxian;HUANG Xindi;WANG Zhihui;XIAO Xiaoxia(School of Informatics,Hunan University of Chinese Medicine,Changsha,China,410208)
出处
《福建电脑》
2021年第11期1-8,共8页
Journal of Fujian Computer
基金
湖南中医药大学计算机科学与技术学科开放基金(No.2018JK04)
湖南中医药大学校级科研基金(No.2019XJJJ029)
湖南中医药大学信息科学与工程学院学科开放基金学生创新性实验(No.7)
湖南中医药大学教学改革研究项目(No.2020-JG029)
2019年湖南省普通高等学校教学改革研究项目(湘教通[2019]291号-389)
2020年湖南省新工科研究与实践项目(湘教通[2020]90号-25)资助。
关键词
垃圾分类
神经网络
机器学习
语音合成
Garbage Classification
Neural Network
Machine Learning
Speech Synthesis
