基于高光谱成像技术烟叶田间高效氯氟氰菊酯用量的快速检测技术研究

Rapid Detection of Lambda-cyhalothrin Dosage in Tobacco Field Based on Hyperspectral Imaging Technology

为了能够有效预防烟叶田间用药过量所产生的农药残留危害,加强烟叶生产过程中的农药用量管控,研究了不同用量高效氯氟氰菊酯处理下的田间鲜烟叶高光谱曲线特征并构建了预测模型。应用可见—近红外高光谱成像系统(400~900 nm)采集不同浓度(CK、1∶2000,1∶1000,1∶350)的高效氯氟氰菊酯处理24 h后的鲜烟叶样本的高光谱图像。使用4种常用的光谱预处理方法进行数据预处理,分别采用连续投影法(SPA)、竞争性自适应重加权采样法(CARS)进行数据降维,最后分别构建了最小二乘支持向量机(LSSVM)和随机森林(RF)模型预测模型。结果表明:(1)光谱数据经二阶导数(D2)预处理后,LSSVM和RF模型的准确率均达到100%,经一阶导数(D1)预处理后,RF测试集准确率也达到100%;(2)SPA特征选择数量少于CARS,其中D1预处理后经SPA特征选择的波长数量最少;(3)综合考虑模型复杂度和稳定性,最终选择D1-SPA-RF为最优预测模型,选择11个特征波长,测试集准确率为97.70%。基于高光谱图像的D1-SPA-RF模型预测性能优异,可用于烟叶田间高效氯氟氰菊酯用量的快速检测。

In order to effectively prevent pesticide residue hazards caused by excessive use of pesticides in tobacco field and strengthen pesticide dosage control in tobacco production,the hyperspectral curve characteristics and discrimination model of fresh tobacco leaves treated with different lambda-cyhalothrin doses were studied.The visible near-infrared hyperspectral imaging system(400~900 nm)was used to collect hyperspectral images of fresh tobacco leaf samples treated with different concentrations of lambda-cyhalothrin(CK,1∶2000,1∶1000 and 1∶350)for 24 hours.Four commonly used spectral preprocessing methods were used for preprocessing,followed by continuous projection(SPA)and competitive adaptive reweighted sampling(CARS)to reduce the dimensionality of the data.Then,the least squares support vector machine(LSSVM)and random forest(RF)were used to establish discrimination models.The results show that:(1)After preprocessing with the second derivative(D2),the accuracy of both LSSVM and RF test sets reaches 100%.After preprocessing with the first derivative(D1),the accuracy of the RF test set also reaches 100%.(2)The number of SPA feature selections is less than that of CARS,and the number of wavelengths selected by SPA feature selection after D1 preprocessing is the least.(3)Taking into account the complexity and stability of the model,D1-SPA-RF was ultimately chosen as the optimal discrimination model,with 11 feature wavelengths selected and a test set accuracy of 97.70%.The D1-SPA-RF based on hyperspectral images has superior discriminative performance,and hyperspectral technology can be used for rapid detection of lambda-cyhalothrin doses in tobacco field.