当前位置：首页>太原>太原理工大学邸会霞&李晓春‖【JAFC】AI+纳米银膜SERS平台:给农药检测装上“火眼金睛”

太原理工大学邸会霞&李晓春‖【JAFC】AI+纳米银膜SERS平台:给农药检测装上“火眼金睛”

2026-03-19 22:17:55

点击蓝字

关注我们

介绍 /Introduction

准确检测有害农药残留对于保障公众健康至关重要。表面增强拉曼散射（SERS）技术具有潜力，但也存在一些实际限制，比如谱线重叠和基质干扰。为解决这些问题，太原理工大学邸会霞&李晓春等人开发了一种由银纳米星（AgNS）和疏水银纳米岛薄膜（AgNIF）组成的混合基底辅助的 SERS 平台，并结合了局部表面等离子体共振和局部浓度效应，以实现高灵敏度。该平台展示了宽广的线性范围和低检测限，对九种农药的检测结果分别为：单种农药的分类准确率为 99.44%，二元混合物为 98.47%，三元混合物为 98.09%，在掺入样本的番茄样本中为 94.60%。再加上最佳的数据预处理方案，我们的 CNN 模型实现了卓越的分类准确率：单种农药为 99.44%，二元混合物为 98.47%，三元混合物为 98.09%，掺入样本的番茄样本为 94.60%。因此，这项工作展示了一种无需标记、灵敏且准确的农药检测和识别工具，对于指导农药使用和确保食品安全具有巨大潜力。

研究图文/Introduction

Scheme 1. Schematic of the CNN-Assisted SERS Platform for Sensitive and Accurate Detection of Multiple Pesticides

Scheme 2. CNN Architecture for Pesticide Classification, Comprising Three Convolutional, Three Max-Pooling, and Four Fully Connected Layers

Figure 1. Characterization and performance of the AgNS/AgNIF SERS platform. (A) TEM and (B) SEM images (inset: size distribution) of AgNS. (C) UV-vis spectrum of AgNS. (D) Background signal of KI-treated AgNIF. (E) SEM images of AgNIF before and after KI treatment. (F) SERS intensity of 4-MBN at 1578 cm-1 and water contact angle (inset) on KI-treated AgNIF. (G) FDTD simulation of the hybrid platform. (H) SERS spectra and (I) peak intensity comparison of 4-MBN on different substrates. (J, K) RSD values from (J) 40 random points and (K) fivebatches of the hybrid platform.

Figure 2. Representative SERS spectra and characteristic peak assignments of (A) TMTD, (B) PYM, (C) AC, (D) TBZ, (E) PQ, (F) CBZ, (G) ET, (H) FEN, and (I) CTL.

Figure 3. Standard curves for the nine pesticides were established by plotting the intensity of a characteristic SERS peak against concentration. Error bars indicate the standard deviation from three replicate measurements.

Figure 4. Pesticide identification using classification algorithms with different data preprocessing protocols. (A) Representative SERS spectra under different data processing patterns. The identification results of (B) the CNN model and (C) the TSNE cluster analysis. Data preprocessing protocols: (A1-C1) truncation, (A2-C2) smoothing, (A3-C3) baseline correction, (A4-C4) normalization, and (A5-C5) baseline correction + normalization.

Figure 5. CNN-based classification of single pesticides and their mixtures. (A) Evolution of test loss and accuracy across training epochs. (B) Distribution of loss and accuracy from 20 random training-test splits. (C) Comparison of key classification metrics (accuracy, precision, recall, F1 score) among different classification algorithms. (D) Confusion matrix for identifying each pesticide type across 2580 SERS spectra. (E, F) Confusion matrices for (E) single and (F) binary mixture identification. (G) Assessment of CNN performance in predicting single pesticides, binary mixtures and ternary mixtures.

Figure 6. Pesticide detection in tomato samples using the CNN-assisted SERS platform. (A) Concentration-dependent response of TMTD in tomato juice, quantified at 1385 cm-1. The error bars represent the standard deviation of three replicate measurements. (B) Identification performance metrics for each pesticide in tomatoes. (C) Confusion matrix for identifying each pesticide type in tomato juice across 1000 spectra. (D) Confusion matrix for the identification of five single pesticides and their binary/ternary mixtures in tomatoes.

结论 /Conclusion

总之，本研究提出了一种简单、无需标记、灵敏且多功能的基于卷积神经网络（CNN）的表面增强拉曼散射（SERS）平台，用于准确检测多种农药及其混合物。这些农药样品的 SERS 光谱是在具有双重信号放大功能的混合 AgNS/AgNIF 表面增强拉曼光谱仪上通过便携式拉曼光谱仪获取的。我们实施了一种专门的预处理方案，包括数据截断、基线校正和归一化，以有效减少背景噪声、基线漂移和 SERS 强度变化。基于处理后的数据训练的 CNN 模型表现出卓越的识别性能，具有出色的泛化能力和鲁棒性，标准样品的准确率超过 98%，真实番茄样品的准确率为 94.60%。尽管本研究在农药的范围限制、固定的混合比例以及缺乏定量预测方面存在局限性，但这些考虑因素清楚地为我们未来的研究指明了方向。因此，该平台仍具有巨大的潜力，可作为一款强大且可靠的工具，用于监测各种环境和食品污染物，包括抗生素、芳香类污染物以及微塑料等。

文献信息 /Literature information

A Highly Sensitive Silver Nanostars/Silver Nanoisland Films Hybrid SERS Platform Assisted by a Convolutional Neural Network for Accurate Pesticide DetectionHuixia Di,*,† Zhouhao Lei,† Jianing Li, and Xiaochun Li*原文链接：https://doi.org/10.1021/acs.jafc.5c11974