Abstract:
This study established long short-term memory (LSTM), convolution neural network long short-term memory (CNN_LSTM), and radial basis function neural network (RBFNN) based on optimized excitation-emission matrix (EEM) from fish eye fluid to predict freshness changes of rainbow trout under nonisothermal storage conditions. The method of residual analysis, core consistency diagnostics, and split-half analysis of parallel factor analysis was used to optimize EEM data, and two characteristic components were extracted. LSTM, CNN_LSTM, and RBFNN models based on characteristic components of EEM used to predict the freshness indices. The results demonstrated the relative errors of RBFNN models with an R2 above 0.96 and relative errors less than 10% for K-value, total viable counts, and volatile base nitrogen, which were better than those of LSTM and CNN_LSTM models. This study presents a novel approach for predicting the freshness of rainbow trout under nonisothermal storage conditions.
摘要:
这项研究建立了长期短期记忆(LSTM),卷积神经网络长短期记忆(CNN_LSTM),基于优化的鱼眼液激发发射矩阵(EEM)和径向基函数神经网络(RBFNN)预测非等温储存条件下虹鳟鱼新鲜度的变化。残差分析法,核心一致性诊断,并采用平行因子分析的分半分析来优化EEM数据,并提取了两个特征成分。LSTM,CNN_LSTM,和基于EEM特征分量的RBFNN模型用于预测新鲜度指数。结果表明,RBFNN模型的相对误差大于0.96,K值的相对误差小于10%,总可行计数,和挥发性碱氮,比LSTM和CNN_LSTM模型更好。本研究提出了一种在非等温储存条件下预测虹鳟鱼新鲜度的新方法。