基于注意力时间卷积网络的香蕉新鲜度识别与剩余货架期预测
Freshness recognition and remaining shelf life prediction of banana based on attention Temporal Convolutional Network
李鑫 1朱磊 1张媛 1杜艳平 1邢晓1
作者信息
- 1. 北京印刷学院机电工程学院,北京 102627
- 折叠
摘要
[目的]解决传统机器学习算法(BP、SVM)无法很好地挖掘时序数据特征导致模型识别和预测效果不佳的问题,最大限度减少新鲜水果在流通过程中鲜度的损失.[方法]以香蕉为研究对象,使用时间卷积网络(TCN)结合有效通道注意力网络(ECA-NET)建立香蕉新鲜度识别模型(ECA-TCN),并进行仿真测试.[结果]BP、SVM、TCN、ECA-TCN的识别准确率分别为84.89%,85.16%,97.83%,99.03%.[结论]试验方法对香蕉的新鲜度识别具有更好的效果.
Abstract
[Objective]To address the issue of traditional machine learning algorithms(BP,SVM)struggling to effectively extract features from time series data,which leads to subpar model recognition and prediction performance,and aim to minimize the freshness loss of fresh fruits during the distribution process.[Methods]Taking bananas as the research subject,established a banana freshness recognition model(ECA-TCN)by combining Time Convolutional Networks(TCN)with Efficient Channel Attention Networks(ECA-NET)and conduct simulation tests.[Results]The recognition accuracies for BP,SVM,TCN,and ECA-TCN were 84.89%,85.16%,97.83%,and 99.03%,respectively.[Conclusion]The experimental method demonstrates superior performance in recognizing the freshness of bananas.
关键词
香蕉/新鲜度/传感器阵列/时间卷积网络(TCN)/注意力机制/剩余货架期预测Key words
bananas/freshness/sensor arrays/TCN/attention mechanism/remaining shelf-life forecasting引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据