Abstract

Fish fraud is a problematic issue for the industry. For it to be properly addressed will require the use of accurate, rapid, and cost-effective tools. In this work, near infrared reflectance spectroscopy (NIRS) was used to predict nutritional values (protein, lipids, and moisture) as well as to discriminate between sources (farmed vs. wild fish) and conditions (fresh or defrosted fish). Samples of five whitefish species—Alaskan pollock (Gadus chalcogrammu), Atlantic cod (G. morhua), European plaice (Pleuronectes platessa), common sole (Solea solea), and turbot (Psetta maxima)—including farmed, wild, fresh, and frozen ones, were scanned by a low-cost handheld near infrared reflectance spectrometer with a spectral range between 900 and 1700 nm. Several machine learning algorithms were explored for both regression and classification tasks, achieving precisions and coefficients of determination higher than 88% and 0.78, respectively. Principal component analysis (PCA) was used to cluster samples according to classes where good linear discriminations were denoted. Loadings from PCA revealed bands at 1150, 1200, and 1400 nm as the most discriminative spectral regions regarding classification of both source and condition, suggesting the absorbance of OH, CH, CH₂, and CH₃ groups as the most important ones. This study shows the use of NIRS and both linear and non-linear learners as a suitable strategy to address fish fraud and fish QC.

中文翻译：

---

近红外反射光谱法与化学计量学相结合，是一种经济有效，快速且无损的鱼类欺诈控制工具：监测五种常见白鲑物种的来源，状况和营养价值

抽象的

鱼类欺诈对整个行业来说是个问题。为了正确解决此问题，将需要使用准确，快速且具有成本效益的工具。在这项工作中，使用近红外反射光谱法（NIRS）预测营养价值（蛋白质，脂质和水分），并区分来源（养殖鱼类与野生鱼类）和状况（新鲜或解冻的鱼类）。五种白鱼的样本-阿拉斯加狭鳕（Gadus chalcogrammu），大西洋鳕（G. morhua），欧洲（Pleuronectes platesa），普通（Solea solea）和大比目鱼（Psetta maxima））-包括耕种的，野生的，新鲜的和冷冻的，用低成本的手持式近红外反射光谱仪进行扫描，光谱范围在900到1700 nm之间。探索了几种用于回归和分类任务的机器学习算法，其精度和确定系数分别高于88％和0.78。主成分分析（PCA）用于根据表示良好线性判别的类别对样本进行聚类。PCA的负载显示出在1150、1200和1400 nm处的条带是关于源和条件分类的最有区别的光谱区域，表明OH，CH，CH ₂和CH ₃的吸光度组是最重要的组。这项研究表明使用NIRS以及线性和非线性学习器作为解决鱼类欺诈和鱼类质量控制的合适策略。