|
| 基于BP-ANN与SHAP可解释性的鸡肉饼贮藏时间判别研究 |
| Study on Discrimination of Chicken Patties Storage Time Based on BP-ANN and SHAP Interpretability |
| 投稿时间：2026-04-29 修订日期：2026-07-07 |
| DOI： |
| 中文关键词: 反向传播人工神经网络模型 鸡肉饼 贮藏时间 沙普利可加性解释 |
| 英文关键词:back propagation artificial neural network model chicken patties storage time shapley additive explanations |
| 基金项目:国家自然科学基金青年科学基金项目（32001723）； 山东省重点研发计划（重大科技创新工程）项目（2024CXGC010913） |
|
|
|
|
| 摘要点击次数: 2 |
| 全文下载次数: 0 |
| 中文摘要: |
| 目的 建立一种兼顾准确性与可解释性的鸡肉饼贮藏时间智能判别方法。方法 以4 ℃贮藏的熟制鸡肉饼为研究对象, 采集其在0、1、3、5、7 d贮藏期内的水分特性、色差、质构特性、电子鼻（electronic nose, E-nose）响应信号及感官评分等多维品质数据；采用主成分分析（principal component analysis, PCA）对数据进行降维与可视化；在此基础上, 构建反向传播人工神经网络（back propagation artificial neural network, BP-ANN）分类模型对贮藏时间进行判别；并引入沙普利可加性解释（shapley additive explanations，SHAP）方法, 量化分析各项输入指标对模型预测结果的贡献度。结果 贮藏期间, 鸡肉饼水分含量显著下降(P<0.05), 而水分活度呈上升趋势；亮度值(L*)显著升高, 红度值(a*)由–0.09降至–0.25；质构特性整体显著降低, 其中不同贮藏天数肉饼的硬度存在显著差异(P<0.05), 咀嚼性下降61.7%。感官评分随贮藏时间延长总体下降, 但对相邻贮藏天数样品的区分度有限。所构建的BP-ANN模型对贮藏时间的整体判别精确度为0.985, 召回率为0.975, 平均绝对误差(mean absolute error, MAE)为0.246 d。SHAP分析表明, 红度值(a*)、电子鼻传感器信号S3及黏着性是模型判别的三个核心特征。结论 基于BP-ANN模型和SHAP法不仅能准确判别不同贮藏时间的鸡肉饼, 还可阐明其决策逻辑, 为肉制品货架期预测与品质分析提供了可靠且透明的技术手段。 |
| 英文摘要: |
| Objective This study aimed to develop an intelligent method for determining the storage time of chicken patties that balances prediction accuracy and model interpretability. Methods Cooked chicken patties stored at 4 °C were selected as the research object. Multi-dimensional quality data were collected across storage days 0, 1, 3, 5, and 7, including moisture-related properties, color difference, texture characteristics, electronic nose (E-nose) response signals, and sensory evaluation scores. Principal component analysis (PCA) was employed for data dimensionality reduction and visualization. A back propagation artificial neural network (BP-ANN) classification model was then constructed to predict storage time. To enhance model transparency, the Shapley additive explanations (SHAP) framework was introduced to quantify the contribution of each input feature to the model’s predictive outcomes. Results During storage, the moisture content of chicken patties decreased significantly (P < 0.05), while water activity showed an upward trend. The lightness value (L*) increased significantly, and the redness value (a*) decreased from –0.09 to –0.25. Texture properties exhibited an overall significant decline: hardness differed significantly among samples with different storage durations (P < 0.05), and chewiness decreased by 61.7%. Sensory scores generally decreased with extended storage but showed limited discriminative power for samples from adjacent storage days. The BP-ANN model achieved an overall prediction accuracy of 0.985, a recall of 0.975, and a mean absolute error (MAE) of 0.246 days. SHAP analysis revealed that the redness value (a*), E-nose sensor signal S3, and adhesiveness were the three core features driving the model’s decisions. Conclusion The integration of the BP-ANN model and SHAP method not only enables accurate discrimination of chicken patties across different storage times but also elucidates the underlying decision-making logic. This approach provides a reliable and interpretable technical tool for shelf-life prediction and quality analysis of meat products. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
