Detection method of mutton adulteration based on Si-VISSA characteristic wavelength selection
张雨晴 1王克俭 1司永胜 1淑英2
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 河北农业大学信息科学与技术学院,河北保定 071001
2. 河北农业大学食品科技学院,河北保定 071001
折叠
摘要
为提高羊肉掺假的检测效率、获得更好的预测精度,对区间变量迭代空间收缩法(Interval variable iterative space shrinkage approach,IVISSA)进行改进,以获得更优的建模性能.在IVISSA算法中引入了联合区间偏最小二乘(Synergy interval partial least squares,Si-PLS)的思想,设计了联合区间变量迭代空间收缩法(Synergy interval variable iterative space shrinkage approach,Si-VISSA),首先将全光谱波长划分为若干区间并组合成不同的联合区间,分别建立偏最小二乘模型,通过比较建模结果选出最优区间组合.该算法还改变了局部搜索条件,最优区间组合确定后再进行区间宽度的优化,通过增减区间相邻的波长并判别增减后的建模结果,最终确定区间边界.这些优化有效降低了算法复杂性,提高了运算精度.将IVISSA和Si-VISSA选取的波长分别建立偏最小二乘模型,结果显示,IVISSA波长选择预测集的决定系数和均方根误差分别为0.943和5.918,Si-VISSA波长选择预测集的决定系数和均方根误差分别为0.980和4.996.对比运行时长发现Si-VISSA所用时间仅为原算法的20％.以上表明Si-VISSA有更好的预测性能以及更高的运行效率.
Abstract
In order to improve the detection efficiency of mutton adulteration and obtain better prediction accuracy,the interval variable iterative space shrinkage approach(IVISSA)was improved to obtain better modeling performance.The idea of synergy interval partial least squares(Si-PLS)was introduced into the IVISSA algorithm to obtained the synergy interval variable iterative space shrinkage approach(Si-VISSA).In the initialization part of the algorithm,the full spectrum wavelength was divided into several intervals,that were intervals combined into different joint intervals,and partial least squares models are established respectively,then the optimal interval combination is selected by comparing the modeling results.The algorithm also changed the local search conditions.After the optimal interval combination was determined,the interval width was optimized.By increasing or decreasing the adjacent wavelengths in the interval and judging the modeling results after the increase or decrease,the interval boundary was finally determined.These optimizations effectively reduced the complexity of the algorithm and improved the accuracy of the operation.The partial least squares models were established for the wavelengths selected by IVISSA and Si-VISSA,respectively.The results showed that the coefficient of determination and root mean square error of the IVISSA wavelength selection prediction set were 0.943 and 5.918,respectively,and the determination coefficient of the Si-VISSA wavelength selection prediction set and the root mean square errors are 0.980 and 4.996,respectively.Comparing the running time,it was found that the time used by Si-VISSA was only 20％ of the original algorithm.The above results suggested that Si-VISSA has better prediction performance and higher operating efficiency.
维普
万方数据