利用高光谱成像技术预测宠物食品的化学成分

摘要：在宠物食品行业中，快速分析以控制宠物食品的营养质量具有重要意义。本研究旨在测试近红外高光谱成像（（HSI） 在确定猫粮和狗粮化学成分含量方面的有效性，并评估混合模型和特征波长对定量模型性能的影响。研究使用参考方法 对 70 份猫粮和 36 份狗粮样本的水分、粗蛋白（CP）、粗脂肪（CF）、粗纤维（CFe）、粗灰分（CA）、钙（Ca）和总磷 （（TP）含量进行了表征。通过偏最小二乘回归（PLSR）建立了涉及猫粮模型和混合模型的定量模型，并使用竞争性自适 应重加权采样（（ CARS）算法选择特征波长。结果显示，除了粗纤维（（ CFe）外，猫粮模型的性能与混合模型相似。在使 用特征波段时，混合模型对粗蛋白（CP）、粗脂肪（CF）、水分和粗纤维（CFe）的预测结果得到优化，其预测决定系数 （（R²p）在0.73到0.96之间，残差预测偏差（RPD）在2.22到5.20之间。然而，总磷（TP）、粗灰分（CA）和钙（Ca） 的预测结果未能满足实际需求。通过优化后的定量模型，能够可视化样本的化学成分分布。本研究结果为宠物食品的快 速在线质量控制提供了理论和技术支持。

关键词：猫粮；狗粮；高光谱成像；偏最小二乘回归

期刊：Microchemical Journal

DOI：https://doi.org/10.1016/j.microc.2024.110903

设备：GaiaField

Predicting the chemical composition of pet food with hyperspectral imaging

Abstract：Performing fast analyses to control the nutritional quality of pet food products is of great interest in the pet food industry. The purpose of this study was to test the efffcacy of near-infrared hyperspectral imaging (HSI) in determining chemical composition content of cat and dog food, and evaluate the inffuence of mixed model and characteristic wavelength on the performance of the quantitative model. Seventy cat food and thirty-six dog food samples were characterised using reference methods for moisture, crude protein (CP), crude fat (CF), crude ffbre (CFe), crude ash (CA), calcium (Ca) and total phosphorus (TP) content. The Partial Least Squares regression (PLSR) was used to establish the quantitative models that involved the cat food model and mixed model. The characteristic wavelength was selected using a competitive adaptive reweighted-sampling (CARS) algorithm. Results showed that, except for CFe, the performance of the cat food model was similar to the mixed model. When using feature band, the prediction results of the mixed models for CP, CF, moisture, and CFe were optimized with R2 p between 0.73 and 0.96 and RPD in the range 2.22–5.20, but the prediction results of TP, CA, and Ca did not meet the actual demand. Optimal quantitative models were used to visualise the sample chemical composition distributions. The results of this study provide theoretical and technical support for the rapid online quality control of pet food.

Keywords：Cat food; Dog food; Hyperspectral imaging; Partial least squares regression