本文已被：浏览 109次 下载 45次
投稿时间：2025-01-09
投稿时间：2025-01-09
中文摘要: 该文选用宁夏枸杞作为试验材料，对枸杞样品经特定浓度的氰戊菊酯、氯氰菊酯、氯氟氰菊酯处理后，进行图像采集与光谱数据分析，并通过多种算法进行样本集划分、光谱预处理、特征波长提取及判别模型构建，最终建立针对菊酯类农药残留的定量检测模型。结果表明，氯氟氰菊酯通过肯纳德-斯通算法（Kennard-Stone，KS）-基线校准（Baseline）-遗传偏最小二乘算法（genetic algorithm and partial least squares，GAPLS）-卷积神经网络（convolutional neural network，CNN）建立的定量预测模型性能最优，校正集和预测集的相关系数分别为0.677、0.571，校正集和预测集的均方根误差分别为0.058、0.065；氰戊菊酯通过随机取样（random sampling，RS）-原始光谱-GAPLS-CNN建立的定量预测模型效果最佳，校正集和预测集的相关系数分别为0.983、0.981，校正集和预测集的均方根误差分别为0.070、0.078；氯氰菊酯通过联合X-Y距离（sample set partitioning based on joint X-Y distances，SPXY）-标准正态变量变换（standard normal variate，SNV）-GAPLS-CNN建立的定量预测模型效果最佳，校正集和预测集的相关系数分别为0.952、0.937，校正集和预测集的均方根误差分别为0.089、0.107。
Abstract:Ningxia wolfberries were selected as the experimental materials. After the wolfberry samples were treated with specific concentrations of fenvalerate，cypermethrin，and lambda-cyhalothrin，image acquisition and spectral data analysis were carried out. Multiple algorithms were used for sample set division，spectrum pretreatment，extraction of characteristic wavelengths，and construction of discriminant models. Finally，a quantitative detection model for pyrethroid pesticide residues was established. The results showed that the quantitative model established by KS-Baseline-GAPLS-CNN had the best prediction performance for lambdacyhalothrin，with the correlation coefficients of 0.677 and 0.571 and the root mean square errors of 0.058 and 0.065 on the calibration set and the prediction set，respectively. The quantitative model established by RS-Raw-GAPLS-CNN had the best prediction performance for fenvalerate，with the correlation coefficients of 0.983 and 0.981 and the root mean square errors of 0.070 and 0.078 on the calibration set and the prediction set，respectively. The quantitative model established by SPXY-SNV-GAPLS-CNN had the best prediction performance for cypermethrin，with the correlation coefficients of 0.952 and 0.937 and the root mean square errors of 0.089 and 0.107 on the calibration set and the prediction set，respectively.
keywords: Ningxia wolfberry fluorescence hyperspectral imaging pyrethroid pesticide residues convolutional neural network quantitative detection
文章编号：202516024 中图分类号： 文献标志码：
基金项目:宁夏市场监督管理厅科技计划项目（2023SJKY0003）；国家市场监管总局科技计划项目（2023MK125）
| 作者 | 单位 |
| 张瑶1，田雨2，吕毅1，锁占杰2，吴龙国2* | 1.宁夏回族自治区食品检测研究院 国家市场监管重点实验室（枸杞及葡萄酒质量安全），宁夏 银川 750021；2.宁夏大学葡萄酒与园艺学院，宁夏 银川 750021 |
引用文本：
津ICP备20000603号-1