ABSTRACT In this study, quality changes in water-holding capacity, weight loss, color, texture properties, and total sulfhydrylcontent of glazed frozen squids during frozen storage at −5, −10, −20, −30 and −40°C, were determined. In addition, backpropagation neural network (BP-NN) and long short-term memory neural network (LSTM-NN) model were established to predict storage time of glazed frozen squids, and then these two models were performed with a comparative study. The results showed that the influence on the quality of the squids during frozen storage at different temperatures had significant difference (P < .05), and at the lower storage temperatures, the declined rate of squids’ quality was slower, especially. However, changes in the total SH content of squid stored at −30 and −40°C, were not significant differences in the first 60 days. The squid frozen at −5°C for 80d, reached the end of the shelf life. Both BP-NN and LSTM-NN model ware reliable models for predicting the storage time of glazed frozen squid. Experimental results of the LSTM-NN model provided an improvement in the accuracy of prediction compared with those obtained by using the BP-NN model, in which the mean absolute percentage error (MAPE) was 5.01% that was lower than the results by BP-NN model (7.67%). However, the LSTM-NN model had some shortcomings in terms of training time compared with the BP-NN model. The use of LSTM-NN provides a technique to predict accurately the storage time of glazed frozen squid.

中文翻译：

---

基于神经网络的不同温度速冻鱿鱼冻藏时间预测

摘要 在本研究中，上釉冷冻鱿鱼在 -5、-10、-20、-30 和 -40°C 冷冻储存期间的持水能力、重量损失、颜色、质地特性和总巯基含量的质量变化分别是决定。此外，建立了反向传播神经网络（BP-NN）和长短期记忆神经网络（LSTM-NN）模型来预测上釉冷冻鱿鱼的储存时间，然后对这两种模型进行比较研究。结果表明，不同温度下冷冻贮藏对鱿鱼品质的影响存在显着差异（P < .05），尤其是在较低的贮藏温度下，鱿鱼品质下降速度较慢。然而，在-30°C和-40°C下储存的鱿鱼总SH含量的变化在前60天没有显着差异。鱿鱼在-5℃冷冻80d，达到保质期。BP-NN 和 LSTM-NN 模型均提供可靠的模型，用于预测上釉冷冻鱿鱼的储存时间。与使用 BP-NN 模型获得的结果相比，LSTM-NN 模型的实验结果提高了预测的准确性，其中平均绝对百分比误差 (MAPE) 为 5.01%，低于 BP-NN 模型的结果。神经网络模型（7.67%）。但是，LSTM-NN 模型与 BP-NN 模型相比，在训练时间方面存在一些不足。LSTM-NN 的使用提供了一种准确预测釉面冷冻鱿鱼的储存时间的技术。与使用 BP-NN 模型获得的结果相比，LSTM-NN 模型的实验结果提高了预测的准确性，其中平均绝对百分比误差 (MAPE) 为 5.01%，低于 BP-NN 模型的结果。神经网络模型（7.67%）。但是，LSTM-NN 模型与 BP-NN 模型相比，在训练时间方面存在一些不足。LSTM-NN 的使用提供了一种准确预测釉面冷冻鱿鱼的储存时间的技术。与使用 BP-NN 模型获得的结果相比，LSTM-NN 模型的实验结果提高了预测的准确性，其中平均绝对百分比误差 (MAPE) 为 5.01%，低于 BP-NN 模型的结果。神经网络模型（7.67%）。但是，LSTM-NN 模型与 BP-NN 模型相比，在训练时间方面存在一些不足。LSTM-NN 的使用提供了一种准确预测釉面冷冻鱿鱼的储存时间的技术。