摘要:
为探讨近红外高光谱成像技术对鸡蛋种类判别的可行性,采用近红外高光谱(900~1700 nm)成像技术,以223个鸡蛋样本为研究对象,其中富硒鸡蛋74枚、无公害鸡蛋72枚、普通鸡蛋77枚,富硒鸡蛋和无公害鸡蛋为海兰褐鸡蛋,普通鸡蛋为洋鸡蛋。对比S-G卷积平滑、基线校准(Baseline)、标准正态变量变换(SNV)、标准化(Normalize),优选出S-G卷积平滑光谱预处理方法;连续投影算法(SPA)、无信息变量消除法(UVE)、后向间隔偏最小二乘波段选择法(BiPLS)算法提取的特征波长数分别为8、107和155,分别建立全光谱、SPA、UVE、BiPLS的PLS-DA判别模型,结果显示在4种模型中BiPLS-PLS-DA的识别性能要优于FS-PLS-DA、UVE-PLS-DA和SPA-PLS-DA,其校正集正确识别率为95.24%,预测集识别率为78.18%。近红外高光谱成像技术作为一种快速、高效的种类判别技术对鸡蛋种类的判别具有可行性。
Abstract:
In order to explore the feasibility of the near-infrared hyperspectral imaging technique of egg types,the samples were identified by near-infrared hyperspectral(900~1700 nm),among 223 egg samples,74 were selenium-enriched eggs,72 eggs were pollution-free eggs,77 of ordinary eggs for the study object. Selenium-enriched eggs and pollution-free eggs were the Hy-line brown eggs,and ordinary eggs were the organic eggs. Contrast Savitzky-Golay(SG)smoothing,Baseline,Standard Normal Variable(SNV),Normalize,preferred SG convolution smooth spectrum preprocessing method. The number of characteristic wavelengths extracted by Successive Projection Algorithm(SPA),Uninformative Variable Eliminatio(UVE),Backward interval PLS(BiPLS)was 8,107 and 155,respectively. PLS-DA discriminant models of full spectrum,SPA,UVE and BiPLS were established respectively,the results showed that the recognition performance of BiPLS-PLS-DA was better than the whole FS-PLS-DA,UVE-PLS-DA and SPA-PLS-DA in the four models,and the correct recognition rate was 95.24%,the prediction rate was 78.18%. Near-infrared hyperspectral imaging technology as a fast and efficient type of discrimination technology to determine the type of eggs is feasible.
下载:
