中文摘要：

采用近红外（NIR）光谱技术和最小二乘支持向量机（LSSVM）参数优化方法,建立定标预测模型测定鱼粉灰分的含量,采用去趋势校正和标准正交校正（DC-SNV）相结合的方式进行光谱预处理,基于网格搜索法建立LSSVM的参数优化模型,提高NIR光谱定量分析的预测精度。结果表明,LSSVM参数网格搜索方法能够有效地应用于鱼粉NIR光谱模型优化,获得的鱼粉灰分的光谱预测值与化学测定值能较准确的匹配,有利于NIR光谱技术快速检测在养殖饲料产品中的应用。

英文摘要：

The quality of fishmeal becomes the key element research to provide a reasonable balance of nutrition ratio for animal growth.Ash is one of the important indicators for the evaluation of fishmeal quality.Based on near-infrared（NIR）spectroscopic quantitative analytical technology,we utilized least squares support vector machine（LSSVM）algorithm,with parameter optimization,to establish calibration models for the determination of ash contents in fishmeal samples.We proposed the combination method of detrend correction and standard normal variate correction（DC-SNV）to preprocess the spectra and reduce the noises.Then we established optimal LSSVM models in the way of parameter gird searching,to find the optimal mode for the enhancement of NIR predictive accuracy.Results showed that LSSVM parameter grid searching algorithm had the applicability in the NIR quantitative determination for fishmeal quality by obtaining an accurate match on the NIR predictive values and the chemical measured values for fishmeal ash.LSSVM parameter optimization method has the potential in rapid quality detection for the aquaculture feed industry.