**1. Introduction**

Polycyclic aromatic hydrocarbons (PAHs) are persistent organic pollutants that are nondegradable, highly toxic, mutagenic, and carcinogenic [1,2]. Human exposure to PAHs can occur via the inhalation of polluted air, food intake, and skin contact, of which food intake accounts for more than 90% of cases [3–6]. In particular, fruit and vegetable surfaces tend to attract large deposits of PAHs owing to their long-term exposure to the atmosphere [7,8]. Therefore, it is of great scientific and practical significance to detect PAH residues on fruit and vegetable surfaces because of their strong carcinogenicity and teratogenicity.

In recent years, spectroscopic methods—such as colorimetry, fluorescence spectroscopy, near-infrared spectroscopy, and surface-enhanced Raman spectroscopy (SERS)—have been widely used in PAH analysis because of their efficiency, sensitivity, and automation [9–12]. Although PAHs exhibit macromolecular fluorescence, conventional fluorescence spectra are easily limited by the broadening of emission bands and it can be difficult to distinguish

**Citation:** Qiu, M.; Tang, L.; Wang, J.; Xu, Q.; Zheng, S.; Weng, S. SERS with Flexible β-CD@AuNP/PTFE Substrates for In Situ Detection and Identification of PAH Residues on Fruit and Vegetable Surfaces Combined with Lightweight Network. *Foods* **2023**, *12*, 3096. https://doi.org/10.3390/ foods12163096

Academic Editor: Dapeng Peng

Received: 31 July 2023 Revised: 12 August 2023 Accepted: 15 August 2023 Published: 17 August 2023

**Copyright:** © 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

similar molecules because of their low specificity. SERS is a vibrational spectroscopy technique that provides information on the structural characteristics of molecules, enhances Raman scattering, and has been widely used in fast-trace analysis [13–15]. The key to SERS applications is the fabrication of nanostructures with local surface plasmon resonance as SERS-active substrates, and the interaction between the substrates and targets [16]. However, the adsorption of most PAH molecules onto the surface of metal nanoparticles (NPs) is low and resonance Raman scattering does not occur, hindering the effective SERS detection of PAHs. To solve this problem, many researchers have attempted to prepare functionalised plasma nanostructures by modifying the surfactants, antigens, antibodies, and supramolecules on the NP surfaces to promote target binding to SERS substrates [17,18]. However, these strategies are subject to interference from functionalised molecules during SERS detection.

Our previous studies [19] showed that β-cyclodextrins (β-CDs) modified on the surface of gold NPs (AuNPs) can effectively trap PAHs to form host–guest compounds because of their hydrophobic inner cavities (which exhibit a cyclooligosaccharide structure); β-CDs also exhibit weak Raman scattering properties that can reduce interference. Moreover, the surfaces of most fruit and vegetables are irregular and uneven. To improve the application of SERS technology for the detection of irregular sample surfaces, many researchers have attempted to construct flexible SERS-active substrates by assembling metal NPs on flexible materials—such as polymethylmethacrylate (PMMA), tape, and poly(ethylene terephthalate) (PET), which can be easily wrapped or formed to collect analytes from irregular sample surfaces [20–22]. Although these methods could achieve in situ detection, their sensitivity and stability required further optimisation because of the viscosity of some of the flexible materials that destroyed nanoarray structures during the stripping process. The superhydrophobic film could effectively narrow the gap between NPs under the action of hydrophobicity to generate a large number of hot spots, which could enhance the SERS signal, and was an effective method for introducing molecules into hot spot regions [23,24]. However, most reported superhydrophobic SERS substrates required various nanofabrication techniques—such as electron beam lithography, optical lithography, and reactive ion etching—thus increasing the cost of actual SERS applications [25,26]. Inexpensive polytetrafluoroethylene (PTFE) films, with the advantage of having a low surface tension, could be combined with lubricants to prepare flexible and hydrophobic platforms that contributed to the generation of hot spots and eliminated the effects of coffee rings and viscosity [27]. Consequently, PTFE films exhibited strong practical application potential for the in situ, sensitive, and stable SERS detection of PAH residues on the irregular surfaces of fruits and vegetables.

To achieve rapid, intelligent, and automated analysis, SERS spectra can be combined with deep learning (DL) methods to build a determination model [28,29]. In particular, lightweight DL networks—such as SqueezeNet, Xception, MobileNet, and ShuffleNet developed based on the representative convolutional neural network (CNN)—have been widely used because of small parameters, low computational overhead, and high precision, showing a higher specificity and sensitivity compared to typical chemometric analysis [30–33]. For example, Weng et al. [34] used SqueezeNet to develop regression models for the analysis of chlormequat chloride and acephate; excellent performance was obtained with coefficients of determination (*R*2) of 0.9836 and 0.9826 and root-mean-square errors (RMSEs) of 0.49 and 4.08, respectively. Wang et al. [35] proposed a novel regression model, a lightweight one-dimensional CNN, for predicting the nicotine content in tobacco leaves with *R*<sup>2</sup> and RMSE values of 0.95 and 0.14, respectively. These results demonstrated that lightweight networks were suitable for the rapid, accurate analysis of SERS spectra. Consequently, SqueezeNet, MobileNet, and ShuffleNet were used to build classification models for the analysis of various PAH residues on fruit and vegetable surfaces.

In summary, this study aims to develop a method for the in situ detection and identification of various PAH residues on fruit and vegetable surfaces using flexible β-CD@AuNP/PTFE substrates and lightweight DL networks (Figure 1). The β-CD@AuNP/PTFE was prepared

by assembling β-CD@AuNPs on a flexible PTFE film coated with perfluorinated liquid and the PAHs were detected based on the flexible substrate. SqueezeNet, MobileNet, and ShuffleNet were used to construct an intelligent analysis model combined with SERS spectra to classify various PAH residues on the fruit and vegetable surfaces.

**Figure 1.** Schematic diagram of the flexible β-CD@AuNP/PTFE combined with lightweight networks to detect PAH residues on fruit and vegetable surfaces.

#### **2. Materials and Methods**
