河蟹的新鲜度是大多数消费者在购买时所考虑的最重要的因素, 挥发性盐基氮（TVB-N）是当前国际通用的评价肉类新鲜度的指标, 但其检测工序繁琐、 耗费时间长, 无法满足当前市场对河蟹新鲜度评价的迫切需求。 因此, 建立一种快速检测河蟹新鲜度的方法是当前急需解决的一大难题。 将购于水产市场的河蟹, 采用聚乙烯充氧袋快速运至实验室, 样本数共126只。 在洁净的工作台上处理后, 将螃蟹分为42个实验样品, 每个样品3只鲜活螃蟹； 42个实验样品放在低温4 ℃的恒温生化培养箱中贮藏, 每天从培养箱中按时取出6个螃蟹样品进行光谱数据采集及新鲜度指标TVB-N的测定, 历时7 d。 采用近红外光谱（NIRS）对贮藏在不同时间下的河蟹新鲜度进行评价, 使用挥发性盐基氮（TVB-N）作为评价河蟹新鲜度的指标, 首先通过比较经五折交叉验证（5-fold CrossValidation）算法、 kennard-stone（KS）算法、 光谱-理化值共生距离（SPXY）算法三种样本划分方法处理后所建模型的预测效果确定最优样本划分方法, 最终采用五折交叉验证（5-fold CrossValidation）算法对样本进行划分。 其中的32个样品被划分为训练集进行模型构建, 其余的10个样品被划分为测试集用于模型检验。 然后在经过五折交叉验证法对样本进行划分的基础上, 分别采用小波变换（WT）、 Savitzky-Golay平滑、 一阶导数法（Db1）、 二阶导数法（Db2）这4种单一算法以及小波变换（WT）与Savitzky-Golay平滑相结合的算法进行预处理, 通过比较预处理后所建模型的预测效果, 确定了小波变换（WT）预处理为最优光谱预处理方法, 从而消除了光谱中的无用信息并提高了信噪比。 再次, 在WT预处理的基础上, 分别采用主成分分析（PCA）法和连续投影 (SPA) 算法提取光谱特征波段, 通过建模比较确定主成分分析（PCA）法为最优波长选择方法, 以所选的16个特征波长作为模型的输入, 不仅提高了模型的运行速度还可以提高模型的稳定性。 最后, 在经过PCA特征提取后, 分别采用偏最小二乘回归（PLSR）算法和多元线性回归（MLR）算法构建TVB-N定量预测模型, 通过比较两种模型的预测效果, 确定了偏最小二乘回归（PLSR）模型为最优建模方法, 最终确定的最优模型为基于WT-PCA-PLSR建立的模型, 模型预测决定系数R2为0.89, 预测均方根误差RMSEP为3.00。 综上所述, 所建立的预测模型具有较高的精度, 可以实现对河蟹新鲜度的快速检测, 具有较好的市场应用前景。

The freshness of river crab is the most important factor that most consumers consider when buying. Total volatile base nitrogen (TVB-N) is a commonly used international index for evaluating meat freshness, However, its detection process is cumbersome and time-consuming, which can not meet the urgent needs of the current market for rapid and objective evaluation of river crab freshness. Therefore, it is an urgent problem to establish a rapid method for detecting freshness of river crabs. A total of 126 crabs purchased from aquatic market were rapidly transported to the laboratory by polyethylene oxygenation bag. After treatment on a clean bench, the crabs were divided into 42 experimental samples, with 3 fresh crabs in each sample; 42 experimental samples were stored in a constant temperature biochemical incubator at low temperature of 4 ℃. 6 crab samples were taken from the incubator on time every day for spectral data collection and freshness index determination for 7 days. In this paper, near infrared spectroscopy (NIRS) was used to evaluate the freshness of river crabs stored at different time, and total volatile base nitrogen (TVB-N) was used as an index to evaluate the freshness of crabs. Firstly, by comparing influence on the model prediction effect of 5-fold Cross Validation, Kennard-stone algorithm and sample set partitioning based on joint X-Y distance algorithm, finally, the 5-fold CrossValidation algorithm was used to divide the samples. 32 samples were used as training sets for model building, and the remaining 10 samples were used as test sets for model testing. Then, on the basis of dividing the samples by five fold cross validation algorithm, wavelet transform (WT), Savitzky-Golay smoothing, first derivative (Db1), second derivative (Db2) and wavelet transform (WT) combined with Savitzky-Golay smoothing were used to pretreat. Wave transform (WT) pretreatment was the best spectral pretreatment method, which eliminated the useless information in the spectrum and improved the signal-to-noise ratio. Once more, on the basis of the WT pretreatment, principal component analysis (PCA) and successive projection algorithm (SPA) were used to extract spectral feature bands, and the principal component analysis (PCA) was used as the optimal wavelength selection method by comparing the model prediction effect. With the selected 16 feature bands as the input of the model, which not only improved the running speed of the model, but also improve the stability of the model. Finally, after PCA feature extraction, by using partial least squares regression (PLSR) and multiple linear regression (MLR) built the TVB-N quantitative prediction model, by comparing the two kinds of model prediction effect to determined the partial least squares regression (PLSR) model for the optimal modeling method, this paper finally determine the optimal model based on WT-PCA-PLSR model, model prediction determination coefficient R2 was 0.89, and the root mean square error of prediction RMSEP was 3.00. In conclusion, the prediction model established in this study has a high accuracy, and this method can realize the rapid detection of the freshness of river crabs, and has a good market application prospect.