电子鼻和随机森林算法快速鉴别野生与 养殖日本真鲈

Rapid identification of wild and farmed Lateolabrax japonicus by electronic-nose technology and random forest algorithm

中文关键词:  电子鼻  随机森林  鉴别  日本真鲈  特征筛选

英文关键词:electronic nose  random forest  identification  Lateolabrax japonicus  feature screening

中文摘要:

      目的 建立电子鼻和随机森林算法快速鉴别野生与养殖日本真鲈的分析方法。方法 采用来源确定且规格不同的日本真鲈, 利用电子鼻中14个金属氧化物传感器获取53份日本真鲈样本(养殖样本25份, 野生样本28份)的特征信号, 构建得到行×列为53×15(含标签列, 野生为1, 养殖为-1)的初始特征矩阵。构建随机森林(randomforest, RF)模型, 并依据袋外错误率(out-of-bag error rate, OOB)对随机森林模型的估计器(决策树)数量和单一决策树最大特征的2个参数进行优化。结果 模型最优估计器数为50, 最大特征数为14, 模型的鉴别准确率达到98.2%。通过该模型, 以贡献率为指标, 对电子鼻传感器进行了特征筛选和排序, 其中S14和S4传感器的贡献率分别为42.9%和36.0%。结论 该技术可以快速鉴别野生和养殖日本真鲈。

英文摘要:

      Objective To establish a rapid identification method for wild and farmed Lateolabrax japonicus by electronic-nose technology and random forest algorithm. Methods Using Lateolabrax japonicus of different sizes with confirmed original materials, feature signals of 53 seabass samples (25 farmed samples, 28 wild samples) were obtained by 14 metal-oxides semiconductor sensors of the electronic-nose. An initial feature matrix formed with row×column as 53×15(labels column included, 1 for wild, -1 for farmed). A random forest model (RF) was constructed, and 2 parameters (estimator number of the RF model and max features of individual decision tree) were optimized according to out-of-bag error rate (OOB). Results The best estimator number was 50, the max feature was 14, and the identification accuracy of the model was 98.2%. According to the model, taking contribution rates as index, the electronic nose senor was selected and ranked, the contribution of S14 and S4 for the identification was 42.9% and 36.0%, respectively. Conclusion This method can rapidly identify the wild and farmed Lateolabrax japonicus.