基于近红外光谱技术鉴别福寿螺卵死活的研究

Identification of the viability status of Pomacea canaliculata eggs based on near⁃infrared spectroscopy technology

目的利用近红外光谱技术,收集存活或死亡福寿螺卵的近红外光谱数据,构建可预测福寿螺卵死活状态的机器学习模型,为准确、快速鉴别福寿螺卵死活和福寿螺防制药物的研发提供新的技术方法。方法采用近红外光谱技术收集福寿螺活卵和死卵原始光谱数据。通过去除异常值和去趋势化分析,对原始数据进行有效清洗。处理后的数据集分别应用支持向量机(SVM)、随机森林(RF)和偏最小二乘判别分析(PLS-DA)3种不同的算法来训练机器学习模型。同时,评估比较3种模型的准确率、精确率、灵敏度和特异度。结果针对当前样本,建立了基于SVM、RF和PLS-DA 3种算法的可预测福寿螺卵死活的机器学习模型,预测准确率分别为90.10%、85.63%和82.41%。其中,采用SVM算法建立的预测模型的准确率最高,且其精确度(94.42%)、灵敏度(89.07%)、特异度(89.52%)均优于RF和PLS-DA方法构建的模型,差异均有统计学意义(F=5.159、2.336、3.071,P均<0.05)。结论近红外光谱技术在检测福寿螺卵的死活状态方面显示出良好的应用潜力,对于福寿螺防治效果评价的应用前景广阔。

Objective To establish a new technical method for accurate and rapid identification of viable and non⁃viable eggs of Pomacea canaliculata based on near⁃infrared spectroscopy technology.Methods Near⁃infrared spectroscopy technology was used to collect raw spectral data of viable and non⁃viable eggs of Pomacea canaliculate.The original data was cleaned by removing outliers and detrending analysis.Three different algorithms,support vector machine(SVM),random forest(RF),and partial least squares discriminant analysis(PLS⁃DA),were applied to the processed data sets to train the machine learning model.At the same time,the accuracy,precision,sensitivity and specificity of the three models were evaluated and compared.Results Based on the current sample,a machine learning model based on SVM,RF and PLS⁃DA was established to predict the death and survival of the eggs,and the prediction accuracy was 90.10%,85.63%and 82.41%,respectively.Among them,the prediction model established by support vector machine algorithm had the highest accuracy,and its accuracy(94.42%),sensitivity(89.07%)and specificity(89.52%)were superior to the model constructed by random forest and partial least squares discriminant analysis method.Conclusion The NIR technique shows a good potential in detecting the dead and alive state of the P.canaliculata eggs,and could be used to evaluate the control effect of the P.canaliculata eggs.