Lincomycin rapid detection method based on surface enhanced Raman spectroscopy

Abstract

The invention discloses a rapid lincomycin detection method based on surface enhanced Raman spectroscopy. The rapid lincomycin detection method comprises the following steps: preparing silver sol containing coarse silver nanoparticles with the average particle size of about 50nm; putting the silver sol containing the rough silver nanoparticles into a centrifugal machine to be treated, forming supernate containing silver nanoparticle scraps and reaction impurities on the upper layer, removing the supernate according to the proportion that the supernate accounts for 80-95% of the total volume of the centrifuged liquid, and then putting the residual sedimentation part into an ultrasonic cleaning machine to be subjected to ultrasonic dispersion, the silver nanoparticles with the uniform particle size are obtained and serve as an SERS enhanced substrate; mixing the SERS enhanced substrate with a lincomycin sample according to a volume ratio of 1: 1-3: 1 to obtain a detection mixed solution; performing SERS (Surface Enhanced Raman Scattering) detection on the detection mixed solution under set detection conditions to obtain a mixed solution SERS spectrum; extracting a characteristic Raman peak from the SERS spectrum of the mixed solution; and based on the extracted lincomycin characteristic Raman peak, carrying out modeling analysis of a regression algorithm so as to carry out quantitative detection of lincomycin.

Description

Technical Field

[0001] This application relates to the field of antibiotic detection technology, and more specifically, to a rapid detection method for lincomycin based on surface-enhanced Raman spectroscopy. Background Technology

[0002] Lincomycin, a commonly used lincosamide antibiotic, is widely used in the prevention and treatment of livestock and poultry diseases, clinical treatment, and microbial fermentation production. Excessive use or improper discharge can lead to residues in environmental water bodies, fermentation broths, and other substrates, threatening not only ecological safety but also potentially causing public health risks such as bacterial resistance. Therefore, rapid and accurate detection of lincomycin is of significant practical importance.

[0003] In the development of lincomycin detection technology, several detection schemes have emerged. The first is high-performance liquid chromatography (HPLC), which separates lincomycin from matrix impurities using a chromatographic column and performs quantitative analysis using a UV or fluorescence detector. This requires complex pretreatment such as sample extraction and purification, and the detection cycle typically takes 1-2 hours. The second is liquid chromatography-mass spectrometry (LC-MS), which combines the advantages of chromatographic separation and qualitative / quantitative mass spectrometry. It offers high detection sensitivity, but the equipment is expensive, the operation is cumbersome, and it requires professional technicians for maintenance. The third is enzyme-linked immunosorbent assay (ELISA), which detects lincomycin based on the specific binding reaction between antigen and antibody. It is relatively simple to operate, but there is a risk of cross-reaction, the results are easily affected by matrix interference, and the kit has a short shelf life.

[0004] Existing detection methods generally suffer from technical limitations: HPLC and LC-MS methods have long detection cycles and complex pretreatment processes, making them difficult to meet the needs of rapid on-site detection; while ELISA methods are simple to operate, their specificity and stability are insufficient, making it impossible to accurately quantify low concentrations of lincomycin residues. These technical shortcomings limit the applicability of existing methods in scenarios such as real-time monitoring of environmental water bodies and online monitoring of fermentation processes, necessitating a technical solution that combines rapid detection with high sensitivity and specificity. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a rapid detection method for lincomycin based on surface-enhanced Raman spectroscopy, which at least alleviates the aforementioned technical problems.

[0006] The technical solutions provided in this application are as follows: A rapid detection method for lincomycin based on surface-enhanced Raman spectroscopy includes the following steps: Step (1) Preparation of crude silver nanoparticle silver sol: Silver nitrate was reduced with neutral hydroxylamine to prepare silver sol containing crude silver nanoparticles with an average particle size of about 50 nm. Step (2), SERS-enhanced substrate preparation: Silver sol containing coarse silver nanoparticles is placed in a centrifuge for treatment. After centrifugation, the silver nanoparticles settle to the bottom, and a supernatant containing silver nanoparticle debris and reaction impurities is formed on the upper layer. The supernatant is removed at a ratio of 80% to 95% of the total volume of the liquid after centrifugation. The remaining sediment is placed in an ultrasonic cleaner for ultrasonic dispersion to obtain silver nanoparticles with uniform particle size, which serve as the SERS-enhanced substrate. Step (3), SERS detection of the mixture: The SERS-enhanced substrate and the lincomycin sample are mixed at a volume ratio of 1:1 to 3:1 to obtain the detection mixture; the detection mixture is subjected to SERS detection under the set detection conditions to obtain the SERS spectrum of the mixture; Step (4), Characteristic Raman Peak Extraction: Extract lincomycin from the SERS spectrum of the mixture. The characteristic Raman peaks of C–N stretching vibrations at the location and The C–H bending vibration characteristic Raman peak at the location; Step (5), Spectral modeling and quantification: Based on the extracted characteristic Raman peaks of lincomycin, the Raman signals related to lincomycin in the SERS spectrum of the mixture are modeled and analyzed by regression algorithm in order to perform quantitative detection of lincomycin.

[0007] The rapid detection method for lincomycin based on surface-enhanced Raman spectroscopy provided in this application has the following technical advantages: In terms of detection efficiency, this method eliminates the need for complex sample pretreatment procedures. Detection can be performed simply by directly mixing the lincomycin sample with a silver nanoparticle SERS-enhanced substrate. Combined with a 500ms scan time, the entire detection cycle is shortened by more than 90% compared to HPLC and more than 95% compared to LC-MS, fully meeting the needs for rapid on-site detection and online monitoring of the fermentation process. Its core logic lies in the signal amplification characteristics of SERS technology. Through the localized surface plasmon resonance effect of silver nanoparticles, the Raman signal of lincomycin is directly amplified, eliminating the need for time-consuming separation and purification steps.

[0008] Regarding specificity and stability, this method involves extracting lincomycin in... (C–N stretching vibration) and The characteristic Raman peak of (C–H bending vibration) effectively avoids interference from other impurities in the matrix, significantly improving specificity compared to the ELISA method. Simultaneously, the preparation of a uniformly sized SERS-enhanced substrate through centrifugation and ultrasonic dispersion ensures that the relative standard deviation of the characteristic peak intensity is less than 10%, resulting in significantly improved detection stability compared to traditional Raman detection methods. This design specifically addresses the issues of insufficient specificity and poor stability in existing methods, ensuring the reliability of low-concentration lincomycin detection results.

[0009] In terms of applicable scope, this method supports the detection of various sample types, including lincomycin aqueous solution, environmental water mixture, and fermentation broth supernatant, and its quantitative range covers a wide range. This method can meet the monitoring needs of high-concentration fermentation broth and also achieve the detection of low-concentration environmental water residues. Compared with the strict limitations of HPLC and LC-MS methods on sample matrices, this method enhances its adaptability to complex matrices and broadens the scope of detection scenarios through pH optimization (5.22) and spectral data preprocessing (baseline removal, smoothing, and normalization).

[0010] Compared to traditional detection methods, which often rely on "separation-detection" or "specific binding-detection" approaches, this method, based on the SERS enhancement mechanism, employs a "signal amplification-feature extraction-modeling quantification" approach, simultaneously balancing detection efficiency, specificity, and applicability. The preparation and optimization of the silver nanoparticle substrate ensures the stability of signal amplification, the precise extraction of characteristic Raman peaks guarantees detection specificity, and regression algorithm modeling achieves quantitative accuracy, providing a superior technical option for lincomycin detection. Attached Figure Description

[0011] Figure 1 : A schematic flowchart of the rapid detection method for lincomycin based on surface-enhanced Raman spectroscopy of this invention; Figure 2 Characterization images of silver nanoparticles; A is the SEM image of silver nanoparticles, and B is the UV spectrum of silver nanoparticles. Figure 3 Raman spectra of silver nanoparticles and lincomycin added sequentially; Figure 4 SERS spectra of lincomycin and silver sol solutions at different volume ratios; Figure 5 SERS spectra of lincomycin under different pH conditions; Figure 6 SERS substrate homogeneity verification diagram; A is the SERS spectrum of ten parallel measurements of lincomycin, B is... A bar chart showing the intensity of the SERS peak; Figure 7 SERS spectra of lincomycin standard solutions at different concentrations; A is the SERS spectrum, and B is the standard curve. Figure 8 SERS spectrum of lincomycin in environmental lake water matrix; A is SERS spectrum, B is standard curve; Figure 9 SERS spectra of lincomycin supernatant at different fermentation times; Figure 10Scatter plot of prediction results from PLSR and CNN algorithms. Detailed Implementation

[0012] A rapid detection method for lincomycin based on surface-enhanced Raman spectroscopy includes the following steps: Step (1) Preparation of crude silver nanoparticle silver sol: Silver nitrate was reduced with neutral hydroxylamine to prepare silver sol containing crude silver nanoparticles with an average particle size of about 50 nm. Step (2), SERS-enhanced substrate preparation: Silver sol containing coarse silver nanoparticles is placed in a centrifuge for treatment. After centrifugation, the silver nanoparticles settle to the bottom, and a supernatant containing silver nanoparticle debris and reaction impurities is formed on the upper layer. The supernatant is removed at a ratio of 80% to 95% of the total volume of the liquid after centrifugation. The remaining sediment is placed in an ultrasonic cleaner for ultrasonic dispersion to obtain silver nanoparticles with uniform particle size, which serve as the SERS-enhanced substrate. Step (3), SERS detection of the mixture: The SERS-enhanced substrate and the lincomycin sample are mixed at a volume ratio of 1:1 to 3:1 to obtain the detection mixture; the detection mixture is subjected to SERS detection under the set detection conditions to obtain the SERS spectrum of the mixture; Step (4), Characteristic Raman Peak Extraction: Extract lincomycin from the SERS spectrum of the mixture. The characteristic Raman peaks of C–N stretching vibrations at the location and The C–H bending vibration characteristic Raman peak at the location; Step (5), Spectral modeling and quantification: Based on the extracted characteristic Raman peaks of lincomycin, the Raman signals related to lincomycin in the SERS spectrum of the mixture are modeled and analyzed by regression algorithm in order to perform quantitative detection of lincomycin.

[0013] Optionally, the specific operation for preparing the crude silver nanoparticle silver sol in step (1) includes: mixing 4 mL of 6 mmol / L hydroxylamine hydrochloride aqueous solution with 4 mL of 5 mmol / L sodium hydroxide aqueous solution to obtain a neutral hydroxylamine solution; adding the neutral hydroxylamine solution dropwise to 72 mL of 1 mmol / L silver nitrate aqueous solution and reacting at room temperature for 5 min; then adding 800 μL of 1% wt trisodium citrate solution and continuing to stir the reaction for 4 h to obtain a yellow-green silver sol containing crude silver nanoparticles with an average particle size of about 50 nm.

[0014] Optionally, the stirring speed for continuing the reaction for 4 hours in step (1) is 200~500 r / min.

[0015] In step (1), the preparation of crude silver nanoparticles and the control of stirring speed are achieved by precisely controlling the composition ratio, feeding sequence and stirring parameters of the reaction system. This enables the preparation of crude silver nanoparticles with an average particle size of about 50 nm and uniform dispersion, laying the core material basis for the high activity and stability of the subsequent SERS-enhanced substrate.

[0016] The determination of the ratio of 4 mL 6 mmol / L hydroxylamine hydrochloride to 4 mL 5 mmol / L sodium hydroxide, the dropwise addition of neutral hydroxylamine solution, the sequential addition of 800 μL 1% wt trisodium citrate, and the stirring speed of 200~500 r / min is based on a comprehensive consideration of the nucleation and growth kinetics of silver nanoparticles and the interfacial stability mechanism. From the perspective of the reduction system, the neutral hydroxylamine system (pH=7.0±0.2) formed by mixing hydroxylamine hydrochloride and sodium hydroxide in a 1:1 volume ratio can precisely control the reduction activity of hydroxylamine, avoiding the problems of rapid silver ion deposition in a strongly alkaline environment or insufficient reduction efficiency in an acidic environment. The concentration ratio of 6 mmol / L hydroxylamine hydrochloride to 1 mmol / L silver nitrate makes the molar ratio of reducing agent to silver ions reach 1:3, which ensures complete reduction of silver ions and avoids the occupation of active sites on the particle surface due to excessive reducing agent residue. From a timing control perspective, the process of reducing nucleation before adding the stabilizer differs from the traditional extensive "simultaneous addition" method. This avoids interference from trisodium citrate on silver crystal nucleus formation, ensuring the integrity of crystal nucleus growth. The dosage of 800 μL 1%wt trisodium citrate in a molar ratio of 0.52:1 to silver ions forms a monomolecular adsorption layer on the particle surface, preventing aggregation through electrostatic repulsion without covering active sites. Regarding the stirring rate, the range of 200–500 r / min is an optimization based on the mass transfer efficiency of the reaction system and the stability of crystal nucleus growth. This avoids localized concentration unevenness caused by excessively low rates while preventing crystal nucleus breakage caused by excessively high rates.

[0017] The neutral hydroxylamine system is the core component for achieving gentle and uniform reduction of silver ions. 1. Mechanism of component concentration and ratio matching: A neutral hydroxylamine reduction system was constructed by mixing 6 mmol / L hydroxylamine hydrochloride aqueous solution and 5 mmol / L sodium hydroxide aqueous solution in a 1:1 volume ratio. This concentration ratio was determined through extensive experimental screening. When the concentration of hydroxylamine hydrochloride is below 4 mmol / L, the reduction rate is too slow and the reduction of silver ions is incomplete. When it is above 8 mmol / L, excess hydroxylamine is prone to decomposition to produce nitrogen gas, which causes the system to foam and destroys the crystal nucleus growth environment. The concentration of 5 mmol / L sodium hydroxide forms a precise neutralization with hydroxylamine hydrochloride, so that the pH of the mixed system is stable at 7.0±0.2. At this time, hydroxylamine exists in a neutral molecule form with moderate reducing activity, which can realize the gradual reduction of silver ions and avoid rapid nucleation and aggregation under strong reducing environment. The mixing process should be carried out at room temperature (25±2℃) with magnetic stirring at 300r / min for 10min to ensure that the two solutions are fully homogenized and form a stable neutral hydroxylamine reduction environment. These mixing parameters are based on specific optimizations of acid-base neutralization reaction kinetics. Insufficient stirring or temperature fluctuations will lead to local pH deviations and cause uneven silver ion reduction rates.

[0018] 2. Mass transfer control of dropwise addition: 72 mL of 1 mmol / L silver nitrate aqueous solution was added dropwise to the neutral hydroxylamine solution at a rate of 1-2 drops / second. The drop rate was precisely controlled using a constant-pressure dropping funnel, with each drop having a volume of 0.05 mL, ensuring a constant molar amount of silver ions entering the reaction system per unit time. (mol / s). Dropwise addition avoids sudden spikes in silver ion concentration in localized areas, allowing silver ions to diffuse evenly in the reduction system and fully contact hydroxylamine molecules for simultaneous nucleation, maintaining the particle size distribution index (PDI) below 0.25. A single-stage addition would result in excessively high local silver ion concentrations, leading to a massive, instantaneous nucleation and the formation of broadly distributed particles with diameters of 20–100 nm, which would not meet the requirements of the subsequent SERS substrate. During the dropwise addition, the reaction system gradually changes color from colorless and transparent to light yellow, indicating the beginning of silver crystal nucleation; this color change is a direct indicator of the initiation of nucleation.

[0019] The sequential addition of trisodium citrate is key to controlling the morphology and dispersibility of silver nanoparticles: 1. Kinetic adaptation of stabilizer addition timing: After the silver nitrate was added, the reaction continued for 5 minutes at room temperature, and then 800 μL of 1% wt trisodium citrate aqueous solution was added. This timing design is based on the growth law of silver crystal nuclei - the first 5 minutes are the stage of rapid reduction of silver ions and formation of crystal nuclei. At this time, the crystal nuclei are about 5~10 nm in diameter and have not yet formed a stable morphology. If trisodium citrate is added in advance, it will be adsorbed on the surface of the crystal nuclei, inhibiting the growth of specific crystal faces and causing morphological distortion. After 5 minutes of reaction, the crystal nuclei have been initially stabilized. At this time, the stabilizer can guide the crystal nuclei to grow into spherical shapes through the specific binding of carboxyl groups to the surface of silver crystal nuclei, and prevent the crystal nuclei from agglomerating through electrostatic repulsion, so that the particles grow uniformly to about 50 nm. The dosage of 800 μL 1%wt trisodium citrate was precisely calculated, and its molar ratio with silver ions in the system was 0.52:1. This ratio can form a dense monomolecular adsorption layer on the particle surface, with an absolute zeta potential of over 35 mV, ensuring dispersion stability. If the dosage is insufficient (molar ratio <0.3:1), the stabilization effect will be poor, and the particles will easily agglomerate; if the dosage is excessive (molar ratio >0.8:1), it will form a multilayer coating, blocking the SERS active sites on the particle surface.

[0020] 2. Enhanced interfacial interaction during the growth stage: After adding trisodium citrate, the system color gradually changed from light yellow to milky white, eventually forming a yellowish-green opaque sol. This color change corresponds to the growth and stabilization process of silver nanoparticles—the milky white stage represents rapid nucleus growth (particle size 10~30nm), and the yellowish-green stage represents saturation of stabilizer adsorption on the particle surface (particle size stabilized at around 50nm). To enhance the binding between the stabilizer and the particle surface, a specific stirring rate needs to be started immediately after addition to ensure uniform diffusion of trisodium citrate throughout the system, avoiding uneven dispersion caused by insufficient local stabilizer concentration. The stirring rate during this stage is consistent with the rate during the subsequent 4-hour growth process, creating a stable growth environment.

[0021] Precise control of stirring speed from 200 to 500 rpm, and dynamic adaptation of stirring speed, ensure uniform growth and dispersion of silver nanoparticles. 1. Mechanism and Range Determination of Stirring Rate: The stirring rate range of 200~500 r / min is determined based on the mass transfer requirements and crystal nucleus growth characteristics of the reaction system. 200 r / min is the minimum effective rate, ensuring uniform mixing of the reducing agent and silver ions in the system, avoiding local concentration gradients, and enabling synchronous crystal nucleus growth. 500 r / min is the maximum safe rate; exceeding this rate will generate excessive shear force, causing the formed silver crystal nuclei to break, forming small-diameter fragments, and destroying particle size uniformity. Different stirring rates are adapted to different growth stages: the initial 2 hours are the rapid crystal nucleus growth stage, using a rate of 300~400 r / min, which provides sufficient mass transfer efficiency while avoiding crystal nucleus damage caused by excessively high rates. The following 2 hours are the particle stabilization stage, using a rate of 200~300 r / min to reduce the probability of collisions between particles, consolidate the stabilizer adsorption layer, and improve dispersion stability.

[0022] 2. Process Control During Stirring: A magnetic stirrer with a 3cm stir bar was used to ensure uniform stirring by matching the reaction vessel (250mL beaker). During stirring, the reaction system temperature was maintained at a stable 25±2℃ to avoid fluctuations in stirring efficiency. Simultaneously, the distance between the stir bar and the bottom of the vessel was controlled at 0.5cm to prevent particle deposition. High-speed camera monitoring showed that within the range of 200~500r / min, the system flow field was uniform, the silver crystal nucleus growth rate was consistent, and the particle size variation coefficient was ≤15%, significantly better than the results of constant 100r / min (variance coefficient ≥30%) or 600r / min (variance coefficient ≥25%).

[0023] By employing a neutral hydroxylamine system for directional reduction, time-sequential stabilizer regulation, and synergistic fractional stirring rate, the average particle size of the prepared crude silver nanoparticles was precisely controlled to 50±5 nm, with a PDI ≤ 0.25, a sphericity ≥ 90%, and excellent dispersibility, providing high-quality raw materials for the subsequent preparation of SERS-enhanced substrates. Without this control method, the silver nanoparticles would have a wide particle size distribution (30~80 nm) and irregular morphology, leading to increased difficulty in subsequent centrifugation and fractionation, and a 60% decrease in the success rate of preparing substrates with uniform particle size. Furthermore, the silver nanoparticles prepared using this method have abundant surface active sites and exhibit stable binding with trisodium citrate.

[0024] Uniform coarse silver nanoparticles, after centrifugation (5500 rpm for 10 min), efficiently remove debris and impurities. Following ultrasonic dispersion, they form a uniformly sized SERS-enhanced substrate with consistent interparticle spacing and sufficient "hot spot" density. If the particle size is uneven, stratification will occur after centrifugation, making it impossible to obtain a uniform substrate. Silver nanoparticles around 50 nm exhibit high matching degree with the localized surface plasmon resonance effect of 785 nm excitation light, significantly enhancing the Raman signal of lincomycin. and Characteristic peak intensity increased by more than 3 times, detection limit as low as At the same time, the high dispersion of particles avoids the uneven distribution of "hot spots" caused by agglomeration.

[0025] Optionally, the centrifuge processing parameters in step (2) are: rotation speed 5500 rpm, centrifugation time 10 min; ultrasonic dispersion time 5 min.

[0026] Optionally, the detection conditions set in step (3) are: laser wavelength 785nm, scanning time 500ms, and laser power 200mW.

[0027] Optionally, the lincomycin sample in step (3) includes an aqueous solution of lincomycin, a mixture of environmental water containing lincomycin, or a supernatant of fermentation broth containing lincomycin.

[0028] Optionally, the pH value of the mixture is detected to be 5.22 in step (3).

[0029] Optionally, the characteristic Raman peaks of lincomycin extracted in step (4) are specifically as follows: The characteristic Raman peaks of C–N stretching vibrations at the location and The Raman peaks characteristic of C–H bending vibration at the location.

[0030] Optionally, in the spectral modeling and quantification process of step (5), the quantitative detection range of lincomycin is set to... Furthermore, the relative standard deviation of the SERS intensity of the characteristic Raman peak of lincomycin extracted based on the SERS spectrum of the mixed solution during this process is less than 10%.

[0031] In step (2), centrifugation and ultrasonic dispersion are carried out. Through precise design of centrifugation and ultrasonic parameters, the impurities and unqualified particle sizes in the crude silver nanoparticles are efficiently removed, and a uniformly dispersed SERS-enhanced substrate with uniform particle size (50±5nm) is obtained, which provides structural support for the subsequent high-sensitivity detection of lincomycin.

[0032] The determination of the centrifugal speed of 5500 rpm, the centrifugation time of 10 min, and the ultrasonic dispersion time of 5 min is based on a comprehensive consideration of the physical properties of silver nanoparticles and the SERS activity requirements: From the perspective of centrifugal separation mechanism, a speed of 5500 rpm can generate a centrifugal force of about 5000 g, which can completely settle silver nanoparticles of about 50 nm, while keeping debris and reaction impurities with a particle size of <20 nm (such as unreduced silver ions and excess trisodium citrate) in the supernatant, thus achieving precise classification; if the speed is lower than 5000 rpm, the centrifugal force is insufficient, the 50 nm particles do not settle completely, and the separation efficiency decreases by 40%; if it is higher than 6000 rpm, the centrifugal force is too large, which will cause particle agglomeration and destroy the particle size uniformity. From the perspective of ultrasonic dispersion mechanism, an ultrasonic duration of 5 minutes can effectively break up the weak aggregates on the surface of particles after centrifugation without damaging the crystal structure and surface active sites of silver nanoparticles. If the ultrasonic duration is less than 3 minutes, the aggregates cannot be completely dispersed and the "hot spots" are unevenly distributed. If it exceeds 8 minutes, the ultrasonic cavitation effect will cause oxidation of the particle surface and SERS activity will decrease by more than 30%. This parameter combination is the optimal solution for balancing sorting efficiency, dispersion stability and activity retention rate.

[0033] The core of centrifuge processing lies in achieving efficient separation of silver nanoparticles from impurities: Mechanism adaptation and equipment requirements for centrifugation parameters: A high-speed refrigerated centrifuge (temperature control accuracy ±1℃) was selected, with a rotation speed of 5500 rpm and a centrifugation time of 10 min, and the centrifugation temperature controlled at 4℃. The 4℃ low-temperature environment can reduce the Brownian motion intensity of silver nanoparticles, reduce the probability of particle collision and aggregation, and at the same time avoid the desorption of trisodium citrate stabilizer caused by temperature rise; the synergy of the 5500 rpm rotation speed and 10 min time ensures that the centrifugal force and centrifugation time are optimally matched, ensuring that the sedimentation efficiency of 50nm silver nanoparticles is ≥95%, and the impurity removal rate in the supernatant is ≥90%. During centrifugation, the centrifuge tubes need to be placed symmetrically, with a distance deviation from the centrifuge shaft of ≤2mm, to avoid particle sedimentation deviation caused by uneven centrifugal force. This equipment adaptation and parameter setting is based on fluid dynamics optimization. If a regular centrifuge or no low-temperature control is used, it will result in a wide particle size distribution after sorting (PDI>0.3).

[0034] Precise control of supernatant removal after centrifugation: After centrifugation, the supernatant volume should be removed at a ratio of 80% to 95% of the total liquid volume after centrifugation, with 90% being the preferred ratio. The supernatant should be precisely aspirated using a pipette, retaining the bottom 10% sediment and a small amount of supernatant. This ratio design avoids secondary particle agglomeration caused by excessive supernatant removal while ensuring thorough removal of impurities. During operation, the centrifuge tube should be tilted at 30°, and the pipette aspiration rate should be 0.5 mL / min to avoid shearing forces generated during aspiration damaging the settled particles. Randomly discarding the supernatant can result in a particle loss rate exceeding 20%.

[0035] The core of ultrasonic dispersion lies in breaking down agglomerates while preserving particle activity: An ultrasonic cleaner with a power of 100W and a frequency of 40kHz was selected, and the ultrasonic duration was precisely controlled at 5 minutes. The 100W power generates a moderate cavitation effect, breaking up weak aggregates with a particle size ≤200nm without damaging the crystal structure of the silver nanoparticles. The 40kHz frequency avoids localized temperature increases caused by concentrated ultrasonic energy (the system temperature rise during ultrasonication is ≤3℃), ensuring that the trisodium citrate stabilizer does not desorb. During ultrasonication, the container containing the particle suspension is placed in the center of the ultrasonic tank, with the liquid level matching the tank level, ensuring uniform energy transfer. Excessive ultrasonic power (>150W) or excessively low frequency (<30kHz) will lead to increased surface defects on the particles and decreased SERS activity.

[0036] After sonication, the silver nanoparticle suspension was allowed to stand at room temperature for 10 minutes to allow the surface charge of the particles to redistribute and form a stable electric double layer structure, with the absolute value of the zeta potential reaching above 35 mV. This ensured that the dispersion did not exhibit significant aggregation within 7 days of storage at 4°C. Vibration should be avoided during the standing process, otherwise it would disrupt the electric double layer balance and cause the particles to re-aggregate.

[0037] The detection conditions in step (3) are optimized by precisely designing the laser parameters, sample type and pH value to achieve efficient excitation and stable acquisition of lincomycin Raman signal, providing reliable data support for subsequent quantitative analysis. This is achieved through the coupling optimization of photophysics and molecular interaction theory.

[0038] A 785nm near-infrared laser was selected as the excitation source. This wavelength achieves optimal energy matching with the localized surface plasmon resonance (LSPR) peak (approximately 430nm) of silver nanoparticles, maximizing the excitation of the local electric field in the substrate's "hot spot" region. This results in a Raman signal enhancement efficiency for lincomycin that is more than three times higher than that of 532nm excitation light. Simultaneously, the 785nm wavelength effectively suppresses fluorescence interference between lincomycin and the water matrix, increasing the signal-to-noise ratio (SNR) from 20 at 532nm to over 80. Using a 633nm wavelength increases fluorescence interference and reduces the SNR by 50%; using a 1064nm wavelength results in insufficient excitation energy and a 60% signal intensity attenuation.

[0039] Setting the scan time to 500ms and the laser power to 200mW balances signal strength and detection efficiency. The 200mW power provides sufficient excitation energy for lincomycin... and The characteristic peak intensity reaches the detection threshold; if the power is below 150mW, the signal strength is insufficient, indicating a low concentration. Lincomycin cannot be detected; if the laser power is higher than 250mW, the laser thermal effect will cause lincomycin molecules to degrade, resulting in distortion of the characteristic peak shape. A scan time of 500ms can effectively accumulate Raman signals, reduce random noise, and will not cause sample dehydration due to excessive scan time, thus improving detection efficiency by 1 time compared to a 1s scan time.

[0040] To address interference from humic acid and suspended particulate matter in environmental water bodies, the suspended particulate matter needs to be removed by filtration through a 0.45μm filter membrane before detection. Subsequently, the pH=5.22 adjustment condition is used. This pH value can inhibit the adsorption of humic acid and silver nanoparticles, reducing the intensity of interference signals by more than 70%. No complex extraction process is required, which solves the technical pain point of large matrix interference in environmental water bodies in traditional detection.

[0041] Compatibility optimization of fermentation broth supernatant: The fermentation broth supernatant contains culture medium components (such as glucose and amino acids), which do not exhibit significant Raman signals under 785nm excitation and will not interfere with the identification of lincomycin characteristic peaks; simultaneously, the high viscosity of the fermentation broth was adapted by detecting it after a 10-fold dilution, and the mixing ratio of the diluted sample with the SERS substrate remained unchanged. This ensures that the detection sensitivity is not affected.

[0042] pH=5.22 is the range near the isoelectric point of lincomycin molecules. At this pH, the piperazine ring in the lincomycin molecule is in a weakly protonated state, forming electrostatic attraction and hydrogen bonding with the carboxyl group of trisodium citrate on the surface of silver nanoparticles. The adsorption efficiency is 65% higher than that at neutral pH (7.0). At the same time, this pH value can maintain the stability of the zeta potential on the surface of silver nanoparticles, avoid particle aggregation, and increase the density of "hot spots" by 50%.

[0043] A pH meter with an accuracy of ±0.01 was used, and 0.1 mol / L hydrochloric acid solution or sodium hydroxide solution was selected as the adjusting reagent. During the adjustment process, the reagent was added dropwise at a rate of 0.05 mL / min, while magnetic stirring was performed at a rate of 200 rpm to ensure uniform pH change of the mixture and avoid local pH abrupt changes that could alter the molecular structure of lincomycin. After adjusting to pH=5.22, the system was allowed to stand for 10 min to allow the pH to stabilize (fluctuation ≤±0.02). If the adjustment rate is too fast or the stirring is insufficient, pH drift will occur, and the adsorption efficiency will decrease by 30%.

[0044] In step (4), the directional extraction of the characteristic Raman peak of lincomycin is achieved by clearly defining the mechanism adaptation and identification criteria of the characteristic peak of lincomycin, so as to realize the accurate extraction of lincomycin signal in complex matrix and provide specific guarantee for quantitative analysis.

[0045] (C–N stretching vibration) and Selection of characteristic peaks (C–H bending vibration): From the perspective of molecular vibrational mechanisms, The peak corresponds to the C–N stretching vibration of the piperazine ring in lincomycin. The peaks correspond to the C–H bending vibration of the benzene ring. Both peaks are strong characteristic peaks with peak intensities greater than 1000 a.u. and full width at half maximum (FWHM). And in There are no other impurity peaks interfering within the sampling range; if selected The nearby weak peaks are easily interfered with by humic acid in the environmental water or proteins in the fermentation broth, resulting in a quantitative deviation of more than 25%. From the perspective of SERS enhancement adaptation, the vibration modes of these two peaks have the highest coupling efficiency with the local electric field of the "hot spot" region of silver nanoparticles, and the signal enhancement factor is more than 2 times higher than that of other peaks.

[0046] Before acquiring the SERS spectrum of the mixture, a background scan was performed on a blank SERS substrate (not mixed with the lincomycin sample). The scanning parameters were consistent with those of the sample detection (785nm laser, 500ms scan time, 200mW power). Then, the background signal was subtracted by the spectral analysis software to eliminate the weak Raman interference of the substrate itself. The criterion for identifying characteristic peaks is: peak position deviation. The peaks are symmetrical and have no obvious shoulders. The criterion for identifying characteristic peaks is: peak position deviation. Peak intensity is The peak value is 0.8 to 1.2 times that of the target peak. This standard can effectively distinguish lincomycin from matrix interference signals, with a specific recognition rate of over 98%.

[0047] In step (5), the quantitative detection range and signal stability control of lincomycin are achieved through precise design of the quantitative range and multiple controls of signal stability. arrive Precise quantification.

[0048] The determination of the quantitative range and the relative standard deviation (RSD) of SERS intensity <10% indicates that, from the perspective of the concentration response mechanism, this range covers the residual concentrations in environmental water bodies. ) and fermentation broth intermediates ( The common concentration range of lincomycin was observed, and within this range, the adsorption amount of lincomycin on the SERS substrate showed a good linear relationship with the concentration. If the upper limit of the range is lower than This cannot cover the high-concentration detection requirements of fermentation broth; if the lower limit is higher than... This method cannot meet the requirements for detecting trace residues in environmental water bodies. In terms of signal stability, the requirement of RSD < 10% can ensure the consistency of test results across different batches. This requirement is achieved through methods such as substrate homogeneity control and standardization of test conditions, which is a significant improvement over traditional SERS detection (RSD ≥ 15%).

[0049] Design and validation of concentration gradients: Lincomycin standard concentration solutions were set up with a 10-fold gradient. Three parallel samples were set up for each concentration. The standard solution was prepared with ultrapure water. The environmental water sample was mixed with the SERS substrate at a ratio of 1:1. The supernatant of the fermentation broth was diluted 10 times before mixing to ensure that the sample concentration fell within the quantitative range. This gradient design can effectively cover the linear response range and avoid linear deviation caused by excessive concentration intervals.

[0050] Verification method for linear relationship: Plot the logarithm of the standard solution concentration on the x-axis, and... or A standard curve was established with the characteristic peak area as the ordinate, and linear regression analysis was performed using the least squares method, requiring a linear correlation coefficient R² ≥ 0.91. This verification method can accurately reflect the correlation between concentration and signal intensity, providing a reliable model for quantitative calculation.

[0051] Nine detection points were randomly selected from the SERS substrates prepared in the same batch, to... For lincomycin samples, the characteristic peak intensity must be measured, and the RSD must be <8%. When testing samples of the same concentration using substrates prepared in different batches, the RSD must be <10%. This control ensures the consistency of substrate performance. The ambient temperature is maintained during the testing process. To avoid signal drift caused by temperature fluctuations; the laser focal point diameter is controlled at 200μm, and sapphire is used before each detection. Characteristic peaks are calibrated to ensure the stability of laser intensity and focus position. These standardized operations can reduce signal fluctuations by 40% and ensure RSD < 10%.

[0052] In step (3), the synergy between the 785nm laser and pH=5.22 increases the intensity of the lincomycin characteristic peak by 65%, achieving a signal-to-noise ratio (SNR) ≥80, which is significantly optimized compared to traditional detection conditions (SNR ≤30). In step (4), the specific selection of the characteristic peak avoids matrix interference, enabling the lincomycin identification accuracy in complex samples to reach over 98%, thus solving the technical pain point of insufficient specificity in complex matrices in traditional SERS detection. The dispersion stability in step (2) and the signal stability in step (3) result in a higher linear correlation coefficient for quantitative detection in step (5). Detection limit as low as The spiked recovery rate remained at 90%-110%, meeting the accuracy requirements for environmental monitoring and fermentation process monitoring; The quantitative range covers common concentration ranges in practical applications, broadening the applicable scenarios of the detection method.

[0053] Optionally, the regression algorithm modeling analysis in step (5) includes spectral data preprocessing and data modeling. The preprocessing optimization and noise reduction are used to purify the mixed liquid SERS spectrum so that the lincomycin-related Raman signal in the mixed liquid SERS spectrum can achieve a state of reduced background interference and improved signal stability. The correlation between the characteristic Raman peak of lincomycin and the concentration of lincomycin is established through the data modeling.

[0054] Optionally, the spectral data preprocessing method includes spectral band truncation, least squares smoothing for baseline removal, Savitzky-Golay filter smoothing, and normalization at predetermined bands. Spectral band truncation is used to extract the band containing the characteristic peaks of the mixed-liquid SERS spectrum for preliminary dimensionality reduction and to focus the lincomycin-related Raman signal. Least squares smoothing for baseline removal is used to perform baseline correction on the truncation-based mixed-liquid SERS spectrum to eliminate background interference and amplify the characteristic Raman peaks of lincomycin. Savitzky-Golay filter smoothing is used to smooth the baseline-removed mixed-liquid SERS spectrum to reduce signal noise and improve the stability of the lincomycin-related Raman signal. Normalization at predetermined bands is used to unify the signal scale of the smoothed mixed-liquid SERS spectrum to ensure consistency of the lincomycin-related Raman signal modeling data.

[0055] Optionally, the spectral band interception interval is This range covers the band containing the characteristic Raman peak of lincomycin, and is used to focus the lincomycin-related Raman signal in the SERS spectrum of the mixture.

[0056] Optionally, the objective function for baseline removal using the least squares method is:

[0057] in: It is the i-th data point in the SERS spectrum of the mixture, used to reflect the original superposition information of the lincomycin-related Raman signal and the background in this spectrum; It is the i-th data point of the estimated baseline of the mixed liquid SERS spectrum, used to characterize the background interference signal in the spectrum; It is a weighting factor used to adjust the processing weight of the deviation between the characteristic Raman peak region of lincomycin and the baseline region in the SERS spectrum of the mixed solution, so as to highlight the lincomycin-related signal; It is a smoothing parameter used to control the smoothness of the SERS spectral baseline of the mixture; Weighting factors The calculation method is as follows:

[0058] Where p is an asymmetric parameter used to optimize the lincomycin characteristic peak in the SERS spectrum of the mixture. Priority for handling deviations in the region; The baseline removal parameters are set as follows: smoothing parameter 100, asymmetry parameter 0.01, and number of iterations 10. The baseline removal operation is used to eliminate background interference in the SERS spectrum of the mixture and highlight the characteristic Raman peaks of lincomycin to adapt to subsequent modeling.

[0059] This application's least-squares smoothing baseline removal method constructs a computational logic with dual-objective constraints. Combining the distribution differences between the characteristic signal and background interference in the lincomycin SERS spectrum, it designs an asymmetric weight adjustment mechanism and a dynamic baseline smoothing control strategy to achieve precise removal of background interference and complete preservation of the lincomycin characteristic signal. This technique differs from traditional least-squares methods that only focus on the overall smoothness of the baseline. By strengthening the differentiated processing between the lincomycin characteristic peak region and the baseline region, it solves the baseline correction distortion problem caused by weak characteristic signals and strong background interference in the SERS spectrum of low-concentration lincomycin, providing a high-purity signal foundation for subsequent lincomycin characteristic Raman peak extraction and concentration modeling.

[0060] Preferably, the specific implementation process of the least squares smoothing baseline removal in step (5) of spectral data preprocessing is as follows: First, the SERS spectrum of the mixed liquid after band interception is labeled with data points, and each data point is identified as yi, where i represents the sequence number of the data point in the SERS spectrum of the mixed liquid, corresponding to The physical meaning of yi, which is the continuous wavenumber position within the intercepted interval, is the superposition value of the Raman signals related to the C–N stretching vibration and C–H bending vibration of the lincomycin molecule at that wavenumber position, and the background of the detection environment and the interference signals of the substrate itself. Its value comes from the raw light intensity data collected by the SERS detection equipment, and the value range is related to the sensitivity of the detection equipment (e.g., between 100 and 10000 light intensity units), directly reflecting the superposition intensity of the lincomycin signal and the interference signal.

[0061] Preferably, in a scenario, in the specific technical implementation of the least squares smoothing baseline removal in step (5) of spectral data preprocessing, an initial estimated baseline bi sequence is constructed based on the labeled yi sequence, where bi corresponds one-to-one with yi. The initial value of each bi is obtained by averaging the yi sequence using a sliding window. The length of the sliding window is set to a common odd number (e.g., 15 data points). The wavenumber range, which is related to the half-width at half maximum (FWHM) of the characteristic peak of lincomycin. The initial bi sequence is adapted to ensure that each bi accurately reflects the average level of the background interference signal around the corresponding yi. The initial bi sequence constitutes an initial estimated baseline signal set specifically adapted for lincomycin detection, avoiding misjudgment of the characteristic peak region by the baseline due to improper window size.

[0062] Preferably, in the specific implementation of the least squares smoothing baseline removal technique in step (5) of spectral data preprocessing, a dynamic calculation rule for the weighting factor ωi is designed. The role of ωi is to adjust the processing weight of the deviation between each yi and the corresponding bi. Its value is determined by the numerical relationship between yi and bi. When yi is greater than bi, it indicates that lincomycin may be present at that wavenumber position. C–N stretching vibration or The characteristic signal of C–H bending vibration is obtained, where ωi takes the value of the asymmetric parameter p. When yi is less than or equal to bi, it indicates that the wavenumber position is mainly used to detect environmental background and base interference signals. In this case, ωi takes the value of 1 / p, where the asymmetric parameter p is a value greater than 0 and less than 1 (e.g., 0.01). Through this asymmetric design, the region that may contain lincomycin characteristic signals is given higher weight in the calculation, thereby highlighting the influence of lincomycin-related Raman signals and avoiding the characteristic signals being overfitted by the baseline.

[0063] Preferably, in the specific implementation of the least squares smoothing method for baseline removal in step (5) spectral data preprocessing, a smoothing parameter λ is introduced to control the smoothness of the estimated baseline bi. The smoothing parameter λ is a positive number (e.g., 100). Its physical meaning is to constrain the variation amplitude between adjacent data points of the baseline, specifically adapting to the background fluctuation characteristics of the lincomycin SERS spectrum. The background interference of lincomycin detection mainly comes from substrate scattering and ambient light, and the fluctuation is relatively gentle. The value setting of λ needs to balance the smoothing effect and feature retention. The larger the value of λ, the higher the smoothness of the baseline, which can effectively suppress the random fluctuation of ambient light. The smaller the value of λ, the closer the baseline is to the fluctuation trend of the original signal, which can avoid the weak feature signal of lincomycin being misjudged as background, and ultimately avoid the loss of lincomycin feature signal due to excessive smoothing or the retention of too much background noise due to insufficient smoothing.

[0064] Preferably, in the specific technical implementation of the least squares smoothing baseline removal in step (5) of spectral data preprocessing, the calculation logic of the objective function L is constructed. The objective function L consists of two parts, the first part being... This is the sum of the squared deviations of the superimposed signal yi from the estimated baseline bi at each wavenumber position, multiplied by the corresponding weighting factor ωi. The purpose of this part is to minimize the deviation between the lincomycin-related Raman signal and the baseline, thereby enhancing... , Signal weights in regions with characteristic peaks; the second part is That is, three adjacent estimated baseline data points The sum of the squared second-order difference multiplied by the smoothing parameter λ serves to constrain the curvature variation of the estimated baseline, preventing sharp fluctuations in the baseline near the characteristic peak of lincomycin and preventing the baseline from wrapping or cutting off the characteristic peak. The two parts are added together to form the complete objective function L. Its core purpose is to highlight the lincomycin-related signal while ensuring the smoothness of the baseline and adapting to the signal distribution characteristics of the lincomycin SERS spectrum.

[0065] Preferably, the specific implementation process of the least squares smoothing baseline removal in step (5) of spectral data preprocessing is as follows: An iterative optimization method is used to solve for the bi sequence that minimizes the objective function L. The initial input for the iteration is the initial estimated baseline bi sequence adapted for lincomycin detection, the set weight factor ωi calculation rules, the smoothing parameter λ, and the asymmetric parameter p. In each iteration, ωi at each wavenumber position is first calculated based on the current bi sequence, with a focus on strengthening... , The weights of the surrounding areas are assigned, and then the total value is calculated by substituting them into the objective function L. The value of bi is then adjusted by gradient descent to reduce the value of the objective function L. The number of iterations is set to a fixed integer (e.g., 10 times). The bi sequence after each iteration is used as the input for the next iteration to ensure that the estimated baseline bi sequence gradually approaches the real background interference signal in the lincomycin detection scenario, avoiding baseline deviation caused by insufficient iteration or feature signal distortion caused by excessive iteration.

[0066] Preferably, in the specific implementation of the least squares smoothing baseline removal technique in step (5) of spectral data preprocessing, after the iteration, the final optimized estimated baseline bi sequence is obtained. This bi sequence specifically corresponds to the background interference signal for lincomycin detection. The original superimposed signal yi sequence is subtracted from the final bi sequence to obtain the lincomycin pure Raman signal sequence after removing the background interference. In this signal sequence, lincomycin is detected in... The characteristic Raman peaks of C–N stretching vibrations at the location and The ratio of the signal intensity of the Raman peak of the C–H bending vibration characteristic to the background interference is significantly improved, and the peak shape profile is completely preserved, providing a high-quality lincomycin-specific signal input for subsequent Savitzky-Golay filter smoothing processing.

[0067] Optionally, the Savitzky-Golay filter smoothing parameters are set to: window length 15, polynomial order 2, and the smoothing operation is used to reduce the noise of the lincomycin-related Raman signal after baseline removal in the mixed SERS spectrum.

[0068] The Savitzky-Golay filter smoothing technique described in this application essentially achieves selective signal noise suppression by designing a sliding window and polynomial fitting rules adapted to the SERS spectral characteristics of lincomycin, while preserving the complete contour of the characteristic Raman peaks of lincomycin. This technique differs from traditional Savitzky-Golay filters that use a fixed window and a universal polynomial order; it combines the width of the characteristic peaks of lincomycin (… and Half peak width By studying the characteristics of noise distribution, this study solved the problem of difficulty in identifying characteristic peaks caused by the superposition of noise and weak signals in the SERS spectrum of low-concentration lincomycin, providing a highly stable signal basis for subsequent normalization processing and modeling analysis.

[0069] Preferably, the specific implementation process of Savitzky-Golay filter smoothing in step (5) spectral data preprocessing is as follows: First, the lincomycin pure Raman signal sequence after baseline removal smoothed by least squares method is normalized. Each data point in this signal sequence corresponds to the Raman signal intensity at a specific wavenumber position, and the wavenumber interval is consistent with the acquisition accuracy of the SERS detection device (e.g., This forms a continuous one-dimensional signal vector, with its elements arranged in ascending order of wavenumber, covering... The cutoff interval.

[0070] Preferably, in a scenario, in the specific technical implementation of Savitzky-Golay filter smoothing in step (5) of spectral data preprocessing, the length parameter of the sliding window is designed. The window length adopts an odd number of data points, and the selection is based on the wavenumber span corresponding to the half-width of the characteristic Raman peak of lincomycin. This ensures that the window can cover enough signal points to achieve effective fitting, but will not cause distortion of the characteristic peak profile due to an excessively large window. For example, a window length of 15 data points is selected, corresponding to... The wavenumber range is matched with the half-width at half-maximum of the characteristic peak of lincomycin to avoid insufficient noise suppression due to an excessively small window length (e.g., 7 data points) or excessively large window length (e.g., 25 data points) leading to attenuation of the characteristic peak value.

[0071] Preferably, in the specific technical implementation of Savitzky-Golay filter smoothing in step (5) spectral data preprocessing, the order of the polynomial fitting is determined. The polynomial order is selected based on the changing trend of the lincomycin SERS signal. A second-order polynomial is used for fitting. This order can accurately capture the parabolic contour of the lincomycin characteristic peak, while avoiding the overfitting problem caused by higher-order polynomials (such as third-order and above), preventing false signal fluctuations in the noise region, and also distinguishing it from the limitation of first-order polynomials (linear fitting) that cannot accurately restore the curved contour of the characteristic peak. For example, the second-order polynomial can completely preserve the curve of the characteristic peak. The peak exhibits a symmetrical rising and falling trend, and the similarity between the fitted peak shape and the original signal peak shape is higher than 95%.

[0072] Preferably, in the specific implementation of Savitzky-Golay filter smoothing in step (5) spectral data preprocessing, the sliding window movement rules are constructed. The designed sliding window starts from the starting data point of the normalized one-dimensional signal vector and moves sequentially along the wavenumber increasing direction, moving one data point at a time to form a continuous overlapping window. Each window contains N consecutive data points (N is the window length, such as 15). During the window movement, it is ensured that no signal point at any wavenumber position is missed, so that the signal in the entire intercept interval can be smoothed.

[0073] Preferably, in the specific technical implementation of Savitzky-Golay filter smoothing in step (5) spectral data preprocessing, a second-order polynomial is fitted to N data points within each sliding window. During the fitting process, the wavenumber position of each data point within the window is taken as the independent variable, and the corresponding Raman signal intensity is taken as the dependent variable. The coefficients of the second-order polynomial (including the coefficients of the quadratic term, the coefficients of the first term, and the constant term) are calculated using the least squares method to ensure that the fitted curve can fit the true signal trend within the window to the maximum extent, while suppressing signal fluctuations caused by random noise. For example, for data containing... The window at the peak of the characteristic peak allows the fitted curve to accurately pass through the peak point while maintaining the left-right symmetry of the peak.

[0074] Preferably, the specific implementation process of Savitzky-Golay filter smoothing in step (5) of spectral data preprocessing is as follows: extract the intensity value of the data point corresponding to the center position of the window in the fitting curve of each sliding window, and use it as the smoothed signal intensity. The center position of the window is the (N+1) / 2th data point in the window (N is an odd number). For example, the center of the window with 15 data points is the 8th data point. This operation makes each original signal point replaced by the intensity value optimized by the fitting curve of the corresponding window, forming a smoothed lincomycin Raman signal sequence. and The intensity fluctuation of the characteristic peaks is reduced by more than 40% compared to the original baseline-removed signal, and the peak position deviation is ≤ ±1. The noise signal intensity attenuation is ≥60%, and the half-width variation of the characteristic peak is ≤5%.

[0075] Preferably, in the specific technical implementation of Savitzky-Golay filter smoothing in step (5) spectral data preprocessing, the smoothed lincomycin Raman signal sequence is subjected to integrity verification to ensure that the wavenumber range and number of data points of the signal sequence are consistent with the original baseline-removed signal, and that there are no missing or misaligned data points. At the same time, it is verified that the key parameters of the characteristic Raman peak (peak value, half-peak width, peak shape symmetry) have not undergone significant distortion. The smoothed signal sequence after verification is used as the input for subsequent normalization processing at special bands, providing a stable characteristic basis for signal scale unification.

[0076] Optionally, normalization at specific wavelengths is applied to the sapphire Raman probe in... Using the characteristic peaks as a reference, the normalized data range is controlled within... The normalization is used to unify the scale of lincomycin-related Raman signals in the SERS spectra of different mixtures.

[0077] The special band normalization technique in this application essentially involves selecting a stable reference characteristic peak in the lincomycin detection scenario, constructing a calculation rule with a unified signal scale, and combining it with the signal distribution range of the lincomycin SERS spectrum to standardize the Raman signals of different batches and matrix samples. This solves the problem of poor signal comparability caused by improper selection of reference peaks or unreasonable scale range in traditional normalization methods. It provides a highly consistent data foundation for subsequent partial least squares regression models or convolutional neural network modeling, ensuring that the model has a high degree of consistency in predicting the concentration of samples from different sources.

[0078] Preferably, the specific implementation process of normalization at special bands in step (5) of spectral data preprocessing is as follows: first, the reference peak of the lincomycin Raman signal sequence smoothed by the Savitzky-Golay filter is located, and the Raman probe sapphire is used to locate the reference peak. The characteristic peak at a certain point is used as a reference. This characteristic peak originates from the inherent vibration mode of sapphire material and is unaffected by lincomycin signals or the detection environment. It exhibits stable peak intensity, symmetrical peak shape, and a half-peak width. The reason for choosing this peak as a reference is that its wavenumber position is located at... Within the cutoff interval, and with lincomycin The characteristic peaks have no overlapping interference and can independently reflect the stability of the detection system.

[0079] Preferably, in a scenario, in the specific technical implementation of normalization at special bands in step (5) spectral data preprocessing, the SERS spectrum of each mixture is extracted. The intensity value of the reference peak, denoted as I_ref, is either the peak height or the peak area. Peak area is preferred to reduce errors caused by slight shifts in peak shape, for example, through integration. to The signal area within the interval is used as I_ref to ensure the stability of the reference intensity and avoid random errors caused by taking only the peak height.

[0080] Preferably, in the specific technical implementation of normalization at special bands in step (5) spectral data preprocessing, the normalization coefficient k of each mixture SERS spectrum is calculated. The normalization coefficient k is the ratio of the reference peak standard intensity value to the current spectral reference peak intensity value I_ref. The reference peak standard intensity value is obtained by repeatedly measuring the SERS spectrum of a blank sapphire probe. The average peak area (e.g., 10,000 light intensity units). This standard value serves as a unified reference benchmark, aligning the intensity scales of different spectra to this benchmark. For example, when the I_ref of a sample spectrum is 8000 light intensity units... At that time, the normalization coefficient k = 10000 / 8000 = 1.25.

[0081] Preferably, in the specific implementation of normalization at special bands in step (5) of spectral data preprocessing, the intensity value I_i of each data point in the smoothed lincomycin Raman signal sequence is multiplied by the normalization coefficient k to obtain the pre-normalized signal sequence. This operation unifies the reference peak intensity of different sample spectra to a standard intensity value, eliminating the overall signal strength differences caused by factors such as fluctuations in laser power of the detection equipment and slight probe position offsets. For example, in a sample containing lincomycin... The original intensity of the peak was 5000 light intensity units. After multiplying by the normalization factor of 1.25, the normalized intensity was 6250 light intensity units.

[0082] Preferably, in the specific technical implementation of normalization at special bands in step (5) of spectral data preprocessing, the signal sequence I'_i after preliminary normalization is scaled, and the intensity values ​​of all data points are mapped to a fixed range of [-1,1]. The mapping method is linear transformation, and the maximum value of the signal sequence after preliminary normalization is calculated. and minimum value Through formula To achieve scale conversion, for example, the signal intensity range after initial normalization is 1000 to 9000 light intensity units. If the intensity of a certain data point is 5000 light intensity units, then the mapped value is 2×(5000-1000) / (9000-1000)-1=0. The reason for choosing this range is that the intensity fluctuation of the lincomycin SERS signal after preprocessing can be completely covered, and it meets the requirements of the subsequent modeling algorithm for the scale of the input data, avoiding signal saturation due to the range being too narrow or noise amplification due to the range being too wide.

[0083] Preferably, the specific implementation process of normalization at special bands in step (5) of spectral data preprocessing is as follows: for the scale-compressed signal sequence Validity verification is performed, and abnormal data points exceeding the range of [-1,1] are removed. If the proportion of abnormal data points exceeds 1% of the total data points, the normalization coefficient k and scale mapping parameters are recalculated, and the accuracy of reference peak extraction is checked to ensure that the normalized data sequence has no obvious abnormalities. For example, strong interference signal points caused by sample contamination during the detection process are effectively removed, ensuring the integrity and rationality of the normalized lincomycin Raman signal sequence.

[0084] Preferably, in the specific technical implementation of normalization at special bands in step (5) spectral data preprocessing, the normalized lincomycin Raman signal sequence As the final preprocessing result, this sequence The characteristic Raman peaks of C–N stretching vibrations at the location and The intensity ratio of the Raman peaks of the C–H bending vibration characteristics remained unchanged, and the intensity fluctuation of the same concentration signal from different batches of samples was reduced by more than 60% compared with that before normalization. This provided input data with uniform scale and strong comparability for subsequent data modeling, which increased the determination coefficient (R²) of the partial least squares regression model to above 0.91, and kept the root mean square error (RMSE) of the convolutional neural network at a low level.

[0085] Optionally, the data modeling uses a partial least squares regression model or a convolutional neural network, and the data modeling is used to calculate the lincomycin concentration based on the characteristic Raman peak signal of lincomycin in the SERS spectrum of the mixed solution.

[0086] Optionally, the calculation process of the partial least squares regression model specifically includes: Select an initial weight vector (u), which is a weight vector used to initially correlate the SERS spectral signal of the mixture with the concentration of lincomycin; Calculate vectors , where (X) is the lincomycin-related Raman signal matrix after preprocessing of the SERS spectrum of the mixture, and (t) is the projection of the signal matrix in the (u) direction, which is used to extract key signal features that can be associated with lincomycin concentration; Calculate the load vector , where (Y) is the actual concentration matrix of lincomycin in the mixture, (p) is used to reflect the contribution of each Raman signal in the signal matrix (X) to the key feature (t), and (q) is used to reflect the correlation between the concentration matrix (Y) and the key feature (t); Update the initial weight vector This is used to optimize the correlation between the weight vector (u) and the concentration matrix (Y); By extracting multiple components through multiple iterations, a linear correlation model between the lincomycin-related Raman signal and the lincomycin concentration in the SERS spectrum of the mixed solution was established, thereby enabling the quantitative detection of lincomycin.

[0087] The partial least squares regression model designed in this application essentially constructs a feature extraction and correlation modeling logic adapted to the SERS spectral characteristics of lincomycin. Combining the high-dimensional coupling relationship between spectral signals and concentration data, it designs a hierarchical iterative component extraction strategy to solve the multicollinearity problem that easily occurs in traditional linear regression models with high-dimensional spectral data. This achieves a precise linear correlation between lincomycin-related Raman signals and concentration, providing stable and reliable model support for the quantitative detection of lincomycin in different matrices and concentration ranges. This model differs from the generalized design of traditional partial least squares regression models by strengthening the weight allocation and iterative optimization of lincomycin characteristic peak signals, making the model more precise in its application to... The characteristic signal has a higher response sensitivity and improves the consistency of concentration prediction by more than 30% compared with the general model.

[0088] Preferably, the specific implementation process of the partial least squares regression model in step (5) data modeling is as follows: First, construct the lincomycin-related Raman signal matrix X after preprocessing the SERS spectrum of the mixed solution and the actual concentration matrix Y of lincomycin in the mixed solution. The number of rows in the signal matrix X corresponds to the number of samples (e.g., 8 batches of samples with different concentrations), and the number of columns corresponds to the number of wavenumber points in the preprocessed spectrum (covering...). The range (e.g., 1379 wavenumber points) is defined as follows: In matrix X, the element in the m-th row and n-th column represents the normalized Raman signal intensity of the m-th lincomycin sample at the n-th wavenumber point. This intensity includes signal information from the C–N stretching vibration, C–H bending vibration, and other weak vibrations of lincomycin. The concentration matrix Y is a single-column matrix with the same number of rows as the signal matrix X. The element in the m-th row of matrix Y represents the actual concentration value of the m-th lincomycin sample, covering a concentration range of... This ensures that the model training data can cover the concentration range of the actual detection scenario.

[0089] Preferably, in a scenario, in the specific technical implementation of the partial least squares regression model in step (5) data modeling, an initial weight vector u is selected. The weight vector u is a column vector with a length consistent with the number of columns in the signal matrix X (i.e., the number of wavenumber points). Its physical meaning is to assign initial weights to the lincomycin-related Raman signal at each wavenumber point, for the initial association between the signal matrix X and the concentration matrix Y. The value of the initial weight vector u is not randomly set, but is specifically assigned based on the wavenumber position of the lincomycin characteristic peak. and The wavenumber points where the characteristic peak is located are assigned a higher initial weight (e.g., 0.8~1.0), while other wavenumber points are assigned a lower initial weight (e.g., 0.1~0.3). The reason for this design is to highlight the dominant role of the characteristic peak signal in the concentration correlation, avoid the interference of noise signals from irrelevant wavenumber points in the modeling, and differ from the method of equally assigning initial weights in traditional models, so that the model can quickly focus on the key signal.

[0090] Preferably, in the specific technical implementation of the partial least squares regression model in step (5) data modeling, the vector is calculated. , where Xᵀ is the transpose of the signal matrix X, and the length of vector t is the same as the number of rows in the signal matrix X (i.e., the number of samples). Its physical meaning is the projection of the signal matrix X onto the direction of the initial weight vector u, used to extract key signal features that can correlate with lincomycin concentration. This projection process compresses the high-dimensional spectral signal into a low-dimensional feature vector t, where each element represents the comprehensive feature intensity of the corresponding sample. This intensity integrates effective information from the lincomycin characteristic peak signal and other relevant signals, such as in a sample... When the intensity of the feature peak is high, the corresponding t vector element values ​​also increase, achieving effective aggregation of key signal features.

[0091] Preferably, in the specific implementation of the partial least squares regression model in step (5) data modeling, the loading vectors p and q are calculated. The length of the loading vector p is consistent with the number of columns in the signal matrix X. Its physical meaning is to reflect the contribution of the Raman signal at each wavenumber point in the signal matrix X to the key feature vector t. The wavenumber point corresponding to the element with a higher value in the p vector indicates that the signal at that wavenumber point has a stronger correlation with the concentration of lincomycin. For example, The corresponding p-vector elements are typically in the top 10%; the loading vector q is a single-element vector, its physical meaning reflecting the correlation between the concentration matrix Y and the key feature vector t. The closer the absolute value of q is to 1, the stronger the linear correlation between the extracted key feature t and the concentration. During the calculation process, The influence of dimensions is eliminated by autocorrelation normalization of the t vector. This ensures that the range of values ​​for the load vector is consistent, which facilitates the convergence of subsequent iterative optimizations.

[0092] Preferably, in the specific technical implementation of the partial least squares regression model in step (5) data modeling, the initial weight vector u=Yq is updated, and the product of the concentration matrix Y and the loading vector q is used as the new weight vector u, thereby optimizing the weight vector's correlation with concentration. The updated weight vector u will further strengthen the wavenumber point signal weights that are highly correlated with lincomycin concentration, for example... The weights of wavenumber points around the feature peaks are further increased due to the optimization of the q value, while the weights of irrelevant wavenumber points are continuously reduced, making the features extracted in subsequent iterations more focused on the specific signal of lincomycin. Compared with the fixed weight strategy of traditional models, the convergence speed of the model is improved by 40%.

[0093] Preferably, the specific implementation process of the partial least squares regression model in step (5) data modeling is as follows: multiple components are extracted through multiple iterations. The iteration termination condition is set as the number of extracted components reaching a preset threshold (e.g., 8 components) or the change in the load vector q between two adjacent iterations being less than a set threshold (e.g., 0.001). Each iteration repeats the process of "calculating the projection vector t - calculating the load vectors p and q - updating the weight vector u". Each component t extracted in each iteration corresponds to a new set of key features. Different components focus on the correlation between the signal and concentration of different vibration modes of lincomycin. For example, the first two components mainly focus on Strong characteristic peak signals are observed, and subsequent components focus on supplementary correlations with other weak vibrational signals. All extracted components constitute a component matrix T, where the number of rows in matrix T is the same as the number of samples, and the number of columns is the number of extracted components. This matrix integrates the core concentration-related features of the lincomycin spectral signal.

[0094] Preferably, in the specific technical implementation of the partial least squares regression model in step (5) data modeling, a linear correlation model is constructed based on the extracted component matrix T and concentration matrix Y, and the model coefficient β is calculated by the least squares method to establish the lincomycin concentration prediction formula: ,in The model is used to predict the concentration matrix, with ε representing the error term. After model construction, the model is optimized and validated using a training set, validation set, and test set with a 7:2:1 ratio. The training set is used to determine the model coefficients β, the validation set is used to adjust the number of components extracted in the iterations to avoid overfitting, and the test set is used to evaluate the model's predictive performance and ensure the model's coefficient of determination. The root mean square error (RMSE) is at a low level (e.g., below 900). The finally constructed linear correlation model can extract the corresponding components by projection calculation based on the signal matrix X after preprocessing the SERS spectrum of the new lincomycin sample, and substitute them into the prediction formula to obtain the predicted value of lincomycin concentration, realizing the direct conversion from spectral signal to concentration.

[0095] Optionally, the convolutional neural network specifically includes: Basic feature extraction layer: used to extract basic Raman features related to lincomycin from the SERS spectrum of the mixture, and output the basic Raman feature map of lincomycin; Feature detail extraction layer: used to further extract lincomycin-related detail Raman features from the basic Raman feature map of lincomycin, and output the lincomycin detail Raman feature map; Detail feature normalization layer: Used to perform batch normalization processing on the lincomycin detail Raman feature map, stabilize the feature signal intensity, and output normalized lincomycin detail Raman feature map; Normalized Feature Flattening Layer: Used to convert the normalized lincomycin detail Raman feature map into a one-dimensional lincomycin feature vector; First-mapped fully connected layer: used to map the features of the one-dimensional lincomycin feature vector, enhance the non-linear expression of the features through the ReLU activation function, and output the lincomycin feature vector after the first mapping; The second-mapping fully connected layer is used to further map the lincomycin feature vector after the first mapping. The ReLU activation function is used to optimize the correlation between features and concentration, and the output is the lincomycin feature vector after the second mapping. Lincomycin concentration output layer: This layer processes the lincomycin feature vector after the second mapping and outputs the predicted lincomycin concentration. L1 regularization is used to adjust the stability of the predicted concentration.

[0096] Optionally, the parameters of the convolutional neural network are set as follows: optimizer is adam, learning rate is 0.1, epochs is 200, and batch size is 32.

[0097] Optionally, the core metric for model evaluation is the coefficient of determination (R²). 2 The indicators are the root mean square error (RMSE) and the root mean square error (RMSE), which are used to verify the accuracy of the prediction; the formula for calculating RMSE is:

[0098] The calculation formula is: n is the number of samples. This represents the actual concentration of lincomycin in the i-th mixed sample. Let be the predicted concentration of lincomycin in the i-th mixed sample. The average concentration of lincomycin in all mixed sample samples.

[0099] Optionally, the dataset for modeling and analysis in step (5) is divided into a training set, a validation set, and a test set, with a ratio of 7:2:1.

[0100] The convolutional neural network designed in this application essentially constructs a hierarchical extraction and mapping architecture adapted to the SERS spectral features of lincomycin. Combining the low-dimensional features and high-dimensional correlation characteristics of lincomycin Raman signals, it designs targeted feature extraction kernels and regularization strategies to solve the problems of poor feature extraction generalization and easy overfitting of concentration prediction in traditional convolutional neural networks in spectral data processing. It achieves accurate nonlinear mapping from lincomycin-related Raman signals to concentration, reducing the root mean square error of concentration prediction by more than 30% compared with general convolutional neural networks, and providing a highly stable model support for the quantitative detection of lincomycin in complex matrices.

[0101] Preferably, the specific implementation process of the basic feature extraction layer of the convolutional neural network in step (5) data modeling is as follows: the normalized lincomycin Raman signal sequence after passing through a special band is used as input. This signal sequence is a one-dimensional vector with a length corresponding to The number of wavenumber points in the interval (e.g., 1379), where each element in the vector represents the normalized Raman signal intensity of the corresponding wavenumber point, including lincomycin. C–N stretching vibration, The C–H bending vibration and other related vibration signals are analyzed. The basic feature extraction layer uses a general-range number of convolutional kernels (e.g., 16), each with a short-scale size (e.g., 6×1), designed to accommodate the wavenumber span (half-peak width) of the lincomycin characteristic peak. This method can accurately capture local intensity changes of feature peaks, avoiding feature blurring caused by long-scale convolution kernels. The stride of the convolution kernel is set to 1. Pointwise convolution operations are performed on the input one-dimensional lincomycin Raman signal sequence. Each convolution kernel outputs a feature map. The outputs of all convolution kernels are combined to form the lincomycin basic Raman feature map. The number of rows in this feature map is the same as the length of the input signal sequence, and the number of columns is the same as the number of convolution kernels. Each element represents the basic feature intensity extracted by the convolution kernel at the corresponding position. For example, for... The convolution kernel of the feature peak will output a higher feature intensity value in the corresponding wavenumber region.

[0102] Preferably, in the specific technical implementation of the feature detail extraction layer of the convolutional neural network in step (5) data modeling, the basic Raman feature map of lincomycin is used as input. This layer has fewer convolutional kernels than the basic feature extraction layer (e.g., 8), and each convolutional kernel has a shorter scale (e.g., 3×1). This design is used to capture the detailed changes of the lincomycin feature peaks in the basic feature map, such as peak shape symmetry and peak shoulder intensity. These detailed information are potentially related to the concentration of lincomycin and are information that the basic feature extraction layer has not fully explored. The convolution operation adopts a padding strategy to keep the feature map size consistent with the input and avoid the loss of edge details. After convolution, the feature map is nonlinearly transformed by the ReLU activation function, and the elements with feature values ​​less than 0 are set to 0. While retaining effective features, the nonlinear expression ability of the model is enhanced, and the lincomycin detailed Raman feature map is output. The dimension of this feature map is consistent with the output of the basic feature extraction layer, but the feature information is more focused on the specific details of lincomycin.

[0103] Preferably, in the specific implementation of the detail feature normalization layer of the convolutional neural network in step (5) data modeling, the lincomycin detail Raman feature map is received as input. This layer standardizes the value of each feature channel through batch normalization, calculates the mean and variance of each feature channel in the batch of samples (for example, 32 samples as a batch), and converts the feature values ​​into a distribution with a mean of 0 and a variance of 1. This operation can stabilize the feature signal intensity, avoid model training oscillations caused by signal scale differences between different batches of samples, and accelerate the network convergence speed. After batch normalization, the expressive power of the feature is restored through linear transformation, and the normalized lincomycin detail Raman feature map is output. The value distribution of each channel feature in this feature map is more concentrated, effectively suppressing the gradient vanishing problem and providing stable input for the feature mapping of the subsequent fully connected layer.

[0104] Preferably, in step (5) of data modeling, when implementing the normalized feature flattening layer of the convolutional neural network, the two-dimensional normalized lincomycin detail Raman feature map is converted into a one-dimensional lincomycin feature vector. The conversion method is to arrange all elements of the feature map in row-major order, and the vector length is the product of the number of rows and columns of the feature map (e.g., 1379×8=11032). This one-dimensional vector integrates the basic and detail features of lincomycin, and each element represents an extracted and normalized feature value, realizing the dimensionality reduction and integration of high-dimensional features, adapting to the input requirements of the fully connected layer, such as dispersing features in different channels. Feature information is integrated into the same vector.

[0105] Preferably, in the specific technical implementation of the first-mapping fully connected layer of the convolutional neural network in step (5) data modeling, a one-dimensional lincomycin feature vector is used as input. This layer has a general number of neurons (e.g., 64). Each neuron is connected to all elements of the input vector. The input value of each neuron is calculated through linear transformation. The dimension of the weight matrix of the linear transformation is the length of the input vector × the number of neurons, and the dimension of the bias vector is consistent with the number of neurons. The ReLU activation function is applied to the input value of each neuron, and the part of the input value less than 0 is set to 0. While retaining the effective feature mapping information, the nonlinear expression ability of the model is enhanced, so that the model can capture the complex nonlinear relationship between lincomycin features and concentration. The output is the lincomycin feature vector after the first mapping. The length of the vector is consistent with the number of neurons in this layer (e.g., 64), and each element represents the feature value after the first nonlinear mapping.

[0106] Preferably, the specific implementation process of the second-mapping fully connected layer of the convolutional neural network in step (5) data modeling is as follows: The lincomycin feature vector after the first mapping is received as input. This layer also has the same number of neurons as the first-mapping fully connected layer (e.g., 64). A linear transformation is performed using another set of weight matrices and bias vectors to further optimize the correlation between the features and the lincomycin concentration, filtering out feature information that contributes more to concentration prediction. After the linear transformation, the ReLU activation function is applied for non-linear processing, outputting the lincomycin feature vector after the second mapping. The vector length is still 64. This vector focuses more on core features directly related to concentration than the feature vector after the first mapping, such as strengthening… The correlation between characteristic peak intensity and concentration is analyzed to suppress interference from irrelevant noise features.

[0107] Preferably, in the specific technical implementation of the lincomycin concentration output layer of the convolutional neural network in step (5) data modeling, the lincomycin feature vector after the second mapping is used as input. This layer has one neuron, which is connected to all elements of the input vector. Through linear transformation, the high-dimensional feature vector is mapped to a single output value, which initially corresponds to the predicted value of lincomycin concentration. To avoid model overfitting, L1 regularization is introduced in this layer to apply a penalty term to the weight parameters of the linear transformation. The coefficient of the penalty term is set to a small positive number (e.g., 0.001), so that the weight parameters tend to be sparsely distributed, reducing the influence of redundant features on the prediction result. Finally, the predicted value of lincomycin concentration adjusted by L1 regularization is output. The unit of this predicted value is consistent with the actual concentration of lincomycin (M), covering... The quantitative detection range.

[0108] Preferably, in the specific technical implementation of the overall training and optimization of the convolutional neural network in step (5) data modeling, a training set, validation set, and test set with a ratio of 7:2:1 are used for model training and evaluation. The training set is used to update the network weight parameters, the validation set is used to monitor the model overfitting and adjust the number of training iterations, and the test set is used to evaluate the final prediction performance. The model selects the adaptive momentum estimation (Adam) optimizer, the learning rate is set to a value within a general range (e.g., 0.1), the batch size is set to a general number of samples (e.g., 32), and the number of training iterations is set to a value within a general range (e.g., 200 times). During training, the root mean square error (RMSE) is used as the loss function, and the weight matrix and bias vector of each layer are updated layer by layer through the backpropagation algorithm. Training is stopped when the loss function value of the validation set no longer decreases after multiple iterations (e.g., 10 times). Finally, the trained convolutional neural network is obtained. The network has a determination coefficient (R²) ≥ 0.98 for the test set, and the root mean square error (RMSE) is at a low level (e.g., below 560), which can accurately output the concentration prediction values ​​of lincomycin samples with different concentrations.

[0109] The following specific examples are given. Example 1

[0110] 1. Preparation of silver nanoparticles Synthesizing silver nanoparticles with a particle size of approximately 50±10 nm: Prepare a 1% (w / w) trisodium citrate aqueous solution, a 6 mmol / L hydroxylamine hydrochloride aqueous solution, and a 5 mmol / L sodium hydroxide aqueous solution beforehand. Prepare a 1 mmol / L silver nitrate aqueous solution. Take 4 mL of 6 mmol / L hydroxylamine hydrochloride solution and 4 mL of 5 mmol / L sodium hydroxide solution and mix them thoroughly. Then add 72 mL of 1 mmol / L silver nitrate aqueous solution and stir the reaction at room temperature for 5 min (stirring speed of 200 r / min). The solution changes from colorless and transparent to light yellow. Add 800 μL of 1% trisodium citrate solution and stir the reaction for 20 min. The solution changes from light yellow to milky white, and then to a yellow-green opaque liquid.

[0111] The silver nanoparticles prepared by chemical reduction had a particle size of approximately 50 ± 10 nm. The particle size and shape were characterized by scanning electron microscopy (SEM) (see [link to SEM]). Figure 2 ).like Figure 2 As shown, A is the SEM characterization image of silver nanoparticles, with the horizontal axis representing the scale bar (100 nm) and the vertical axis representing the scale bar (100 nm). The image clearly shows that the silver nanoparticles are regular spherical with a particle size concentrated at 50 ± 10 nm and no obvious agglomeration. B is the ultraviolet spectrum of silver nanoparticles, with the horizontal axis representing the wavelength (nm) and the vertical axis representing the absorbance (au). The ultraviolet absorption peak is located at 430 nm, indicating that the silver nanoparticles have a uniform particle size, which meets the requirements for the preparation of SERS-enhanced substrates.

[0112] like Figure 3 The image shows the Raman spectra of silver nanoparticles and lincomycin added sequentially, with the horizontal axis representing wavenumber ( ). The vertical axis represents Raman intensity (au). The figure contains three core curves: one for the pure silver nanoparticle system, one for the pure lincomycin system, and one for the lincomycin + silver nanoparticle mixture. The curve for the pure silver nanoparticle system (bottom) shows no obvious characteristic peak, indicating no Raman signal interference from the substrate; the curve for the pure lincomycin solution (middle) shows a weak characteristic peak; and the curve for the mixture (top) shows... The significantly enhanced intensity of the characteristic peaks proves that the measured peaks are the Raman characteristic peaks of lincomycin, indicating that the SERS enhancement effect of silver nanoparticles is significant.

[0113] 2. Optimal ratio of lincomycin to silver nanoparticles By fixing the concentrations of silver sol and lincomycin solution, adjusting the volume ratio of silver sol to lincomycin solution, and scanning the surface-enhanced Raman spectrum of lincomycin, the optimal ratio between silver sol and lincomycin solution was determined.

[0114] The concentration of the silver sol was fixed at the stock solution concentration, and the concentration of the lincomycin solution was [missing value]. The volume ratio of lincomycin to silver sol was adjusted to 5:1, 4:1, 2:1, 1:1, 1:2, and 1:3, and SERS spectra were collected. (See attached image.) Figure 4 .like Figure 4 The image shows SERS spectra of lincomycin and silver sol solutions at different volume ratios. The horizontal axis represents the wavenumber. The vertical axis represents Raman intensity (au). The six curves in different colors are labeled with volume ratios of 5:1, 4:1, 2:1, 1:1, 1:2, and 1:3, respectively. Each curve is focused. Characteristic peak regions. It can be seen that the curve with a volume ratio of 1:2 has the highest intensity at the two characteristic peaks, and the intensity of the curves with other ratios decreases sequentially, further verifying that the optimal volume ratio of silver sol to lincomycin solution is 1:2.

[0115] 3. Optimal pH of silver sol and lincomycin solution The amount of silver nanoparticles added was fixed at 200 μL. The pH of the mixture of silver sol and lincomycin solution was adjusted to 1.8, 2.1, 2.52, 3.02, 4.35, 5.22, 5.88, 6.46, 7.19, 8.53, and 9.16, respectively. The surface-enhanced Raman spectrum of lincomycin was scanned to determine the optimal pH.

[0116] like Figure 5 The image shows the SERS spectra of lincomycin measured under different pH conditions. The horizontal axis represents the wavenumber. The vertical axis represents Raman intensity (au). The 11 different colored curves in the figure correspond to detection systems with pH values ​​of 1.8, 2.1, 2.52, 3.02, 4.35, 5.22, 5.88, 6.46, 7.19, 8.53, and 9.16, respectively. All curves are expressed in terms of... The two characteristic peaks are the core feature peaks. When pH=5.22, the intensities of both characteristic peaks are the highest among all curves, proving that this pH value is the optimal condition for lincomycin detection. This result is consistent with the adaptation design of the protonation state of lincomycin molecules and the surface charge of silver nanoparticles.

[0117] 4. Homogeneity of lincomycin and SERS substrate Take silver sol and concentration of The amount of silver nanoparticle sol added to the lincomycin solution was fixed at 200 μL, the pH was adjusted to 5.22, and SERS spectra were randomly collected ten times on the substrate with an excitation wavelength of 785 nm. The uniformity of the SERS substrate was determined by comparing the intensity of the lincomycin SERS peaks collected in ten parallel samples.

[0118] like Figure 6 As shown, A represents ten parallel measurements of lincomycin (…). The SERS spectrum of ) with the horizontal axis representing the wavenumber ( The vertical axis represents Raman intensity (au), and the 10 curves show high overlap with no significant shift; B represents ten parallel measurements of lincomycin at... The bar chart shows the peak intensity of SERS samples, with the horizontal axis representing the number of measurements (1-10) and the vertical axis representing the peak intensity (au). The RSD is marked as 0.306% above the bars. It can be seen that the characteristic peak intensity fluctuations of the 10 parallel samples are extremely small, with a relative standard deviation of less than 10%, indicating good homogeneity of the SERS substrate, suitable for quantitative analysis. This is consistent with the effect of uniform particle size control techniques using centrifugation and ultrasonic dispersion.

[0119] 5. SERS detection of lincomycin Take silver sol and concentration of The amount of silver nanoparticle sol added to the lincomycin solution was fixed at 200 μL, the pH was adjusted to 5.22, and the SERS spectrum was collected under excitation wavelength of 785 nm. The SERS spectra of different concentrations of lincomycin were measured.

[0120] like Figure 7 As shown, A represents the SERS spectra of different concentrations of lincomycin (excitation at 785 nm), with the horizontal axis representing the wavenumber (…). The vertical axis represents Raman intensity (au), and the five curves from top to bottom correspond to concentrations. The higher the concentration, the stronger the characteristic peak intensity; B represents lincomycin in... The graph shows the relationship between SERS intensity and the negative logarithm of concentration at a given location, with the horizontal axis representing... (Lincomycin concentration, mol / L), with the ordinate representing peak intensity (au), and the linear fitting curve equation in the figure is: This indicates that in Within the specified range, the characteristic peak intensity exhibits a good linear relationship with the negative logarithm of the concentration, meeting the linearity requirements for quantitative analysis. Example 2

[0121] 6. Lincomycin detection in environmental water samples Silver sol was mixed with lincomycin solutions of different concentrations. The amount of silver nanoparticle sol added was fixed at 200 μL, and the pH was adjusted to 5.22. SERS spectra of lincomycin at different concentrations were collected under excitation wavelength of 785 nm.

[0122] like Figure 8 As shown, A represents the SERS spectra (785 nm excitation) of lincomycin at different concentrations prepared using environmental lake water. The horizontal axis represents the wavenumber. The vertical axis represents Raman intensity (au), and the five curves from top to bottom correspond to concentrations. The lake water matrix did not obscure the characteristic peaks; B is its... The graph shows the relationship between SERS intensity and the negative logarithm of concentration at a given location, with the horizontal axis representing... (Lincomycin concentration, mol / L), with the ordinate representing peak intensity (au), and the linear fitting curve equation is: This demonstrates that the method still has good quantitative effects in complex water environments.

[0123] This invention involves mixing silver nanoparticles with lincomycin to detect the Raman intensity of lincomycin. Due to the plasmon effect on the surface of the silver sol, the Raman characteristic peak of lincomycin is enhanced, enabling quantitative analysis of lincomycin. Under optimized conditions of mixing ratio, dosage, and pH, the linear detection range of lincomycin is [insert range here]. The relative standard deviation of SERS intensity is less than 10%. Silver sol as a SERS substrate has the advantages of low cost, controllability, and high detection efficiency. Moreover, the method is simple to operate, accurate and sensitive, does not require complicated operating techniques, and can meet the requirements of large-scale and rapid analysis and detection.

[0124] Example 3 7. Detection of Lincomycin Fermentation Supernatant Silver sol was mixed with lincomycin fermentation supernatant of different fermentation times. The amount of silver nanoparticle sol added was fixed at 200 μL. SERS spectra of lincomycin fermentation supernatant of different concentrations were collected under excitation wavelength of 1064 nm.

[0125] like Figure 9 The image shows the SERS spectra (1064 nm excitation) of lincomycin supernatant with different fermentation times. The horizontal axis represents the Raman shift. The vertical axis represents intensity (au), and the 10 curves from top to bottom correspond to fermentation times of 27h, 51h, 75h, 99h, 123h, 148h, 172h, 196h, 220h, and 244h, respectively. (See figure.) (C–N stretching vibration) and The intensity of the characteristic peak (C–H bending vibration) first increases and then stabilizes with fermentation time, which is consistent with the accumulation pattern of lincomycin synthesis. Meanwhile, interference peaks such as amide II band and phosphate group (PO) do not affect the identification of characteristic peaks, proving that this method can be used for online monitoring of the fermentation process.

[0126] Example 4 8. Lincomycin Data Modeling and Analysis Eight batches of lincomycin solutions with different concentrations were prepared for modeling and predictive analysis. The dataset was divided into training, validation, and test sets in a ratio of 7:2:1. In this experiment, the first, second, third, fourth, and sixth batches of data were used as the training set, the fifth and eighth batches as the validation set, and the seventh batch as the test set.

[0127] First, all datasets are preprocessed, and the spectral bands are truncated to the specified range. The baseline removal algorithm was configured with a smoothing parameter of 100, an asymmetry parameter of 0.01, and 10 iterations to bring the spectral baseline close to the connecting lines at the bottom of each peak, thereby amplifying the characteristic peaks. The baseline-removed data was then smoothed using a Savitzky-Golay filter with a window length of 15 and a polynomial order of 2. Each data point's neighborhood was fitted with 15 consecutive data points to create a second-order polynomial for smoothing. The smoothed data was then processed using... Normalization is performed at the point to control the data range between [-1, 1].

[0128] Then, Partial Least Squares (PLSR) and Convolutional Neural Network (CNN) models were built for result prediction. The CNN model consisted of a first layer composed of 16 6×6 convolutional kernels; a second layer composed of 8 3×3 convolutional kernels to extract detailed features; a third layer a batch normalization layer to accelerate training and improve model stability; a fourth layer a flattening layer to flatten the multi-dimensional feature map into a one-dimensional vector; the fourth and fifth layers were fully connected layers with 64 neurons each, using ReLU as the activation function; and the sixth layer was the output layer, consisting of one neuron, with L1=0.001 regularization. The Adam optimizer was selected, with a learning rate of 0.1, 200 epochs, and a batch size of 32.

[0129] like Figure 10 The image shows a scatter plot of the original standard values ​​and the prediction results from the PLSR and CNN algorithms. The horizontal axis represents the actual concentration of lincomycin (label), and the vertical axis represents the predicted concentration (predict). The diagonal line in the plot is the ideal prediction line for y=x. Blue dots represent the prediction results from the PLSR algorithm, and red dots represent the prediction results from the CNN algorithm. The coefficient of determination for the PLSR algorithm is also shown. The root mean square error (RMSE) was 894.2399; the CNN algorithm... Both algorithms predict points that are close to the ideal line, but the CNN algorithm has better prediction accuracy.

[0130] To further highlight the technical advantages of the detection method of this application, three comparative examples are set up, using existing conventional detection techniques and variations of the method of this application, respectively, and compared with the core performance indicators of Examples 1-4 of this application, as follows: I. Proportional Design Comparative Example 1: Traditional High Performance Liquid Chromatography (HPLC) Detection Method Detection principle: A C18 column (4.6 mm × 250 mm, 5 μm) was used with methanol-0.05 mol / L potassium dihydrogen phosphate solution (volume ratio 40:60, pH=4.5) as the mobile phase, the flow rate was 1.0 mL / min, the column temperature was 30 ℃, and the UV detection wavelength was 210 nm.

[0131] Sample preparation: Take 1 mL each of lincomycin standard solution, environmental lake water sample, and fermentation supernatant, add 3 mL of methanol for vortex extraction for 10 min, centrifuge at 8000 r / min for 10 min, and then filter the supernatant through a 0.45 μm organic phase filter membrane before testing.

[0132] Detection process: Sequentially inject a series of standard curve solutions ( The sample solution and peak areas were recorded, a standard curve was established, and the sample concentration was calculated.

[0133] Comparative Example 2: Unoptimized SERS substrate detection method Detection principle: Silver nanoparticles were prepared by direct reduction of silver nitrate with sodium citrate (without centrifugation, fractionation and ultrasonic dispersion), and were directly used as the SERS enhancement substrate. The remaining detection conditions were the same as those in Example 1 of this application (laser wavelength 785nm, scanning time 500ms, mixing ratio 1:2, pH=5.22).

[0134] Substrate preparation: Boil 10 mL of 1 mmol / L silver nitrate aqueous solution, quickly add 1 mL of 1% wt trisodium citrate solution, continue boiling for 30 min, cool and use directly as substrate, with a particle size distribution of 30~100 nm (PDI=0.42).

[0135] The testing procedure is the same as in Examples 1-3 of this application, testing the standard solution, environmental water sample, and fermentation supernatant respectively.

[0136] Comparative Example 3: Single PLSR Modeling and Detection Method (No Spectral Preprocessing) Detection principle: The SERS substrate and detection conditions prepared in Example 1 of this application are used, but the spectral data are directly extracted without baseline removal, smoothing, and normalization preprocessing. Characteristic peak intensities were modeled and quantified using the PLSR model.

[0137] Data processing: From the raw SERS spectrum... The peak intensity at a given point is directly used as the input variable, and the lincomycin concentration is used as the output variable. The training set, validation set, and test set are divided into a 7:2:1 ratio to establish the PLSR model.

[0138] The testing procedure was the same as in Example 4 of this application. Eight batches of lincomycin solutions with different concentrations were tested to evaluate the predictive performance of the model.

[0139] Comparison table of detection indicators

[0140] III. Comparative Explanation 1. Comparison with Comparative Example 1 (HPLC method) Detection efficiency: The total time for a single sample (pretreatment + detection) of the method in this application is only 1.5~11.5 min, which is more than 90% shorter than that of the HPLC method (75~100 min). It does not require complicated pretreatment steps such as extraction and filtration, which solves the pain points of long detection cycle and cumbersome operation of the HPLC method. It is more suitable for rapid on-site detection and online monitoring of fermentation process.

[0141] Sensitivity and accuracy: The detection limit of the method in this application ( Although slightly higher than HPLC method ( ), but it already meets the requirements for residual environmental water bodies (usually) ) and fermentation broth monitoring ( It meets the actual needs of HPLC method, and the spiked recovery rate and detection deviation are close to those of HPLC method, and the accuracy meets the quantitative requirements.

[0142] Applicability: The method of this application does not require expensive chromatographic instruments and professional operators, and the equipment is more portable. It can directly detect complex matrix samples such as fermentation supernatant, while the HPLC method requires additional dilution for high-viscosity fermentation broth, and the operation is more complicated.

[0143] 2. Comparison with Comparative Example 2 (Unoptimized SERS substrate) Substrate uniformity and stability: In this application, the silver nanoparticles were made uniform in size (50±5nm, PDI=0.25) and the RSD was only 0.306%~1.25% by centrifugation (5500 rpm / 10 min) and ultrasonic dispersion (5 min). In contrast, the substrate of Comparative Example 2 had a wide particle size distribution (30~100nm), resulting in large fluctuations in the SERS signal. This makes it impossible to achieve stable quantification of low-concentration samples.

[0144] Detection range and sensitivity: The lower limit of linear detection of the method in this application is extended to... Comparative Example 2 ( The improvement was an order of magnitude, and the recovery rate of spiked environmental water samples increased by more than 10%. This is attributed to the more uniform "hot spot" density of the optimized substrate, resulting in more stable Raman signal enhancement efficiency (enhancement factor). ).

[0145] Adaptability to complex matrices: Comparative Example 2 showed a deviation of 18.5% in the detection of fermentation supernatant due to uneven substrate particle size. However, the method of this application controlled the deviation within 8.2% through substrate optimization and targeted extraction of characteristic peaks, making it more suitable for complex matrix samples.

[0146] 3. Comparison with Comparative Example 3 (without spectral preprocessing) Modeling accuracy: This application effectively eliminates background interference and signal noise through a three-level preprocessing process of "baseline removal-smoothing-normalization". The R² of the CNN model reaches 0.9875 and the RMSE is as low as 557.13. In contrast, Comparative Example 3, due to the lack of processing of spectral data, has a larger modeling error (RMSE=1538.79) caused by baseline drift and noise in the original signal, and the R² is only 0.8864.

[0147] Characteristic peak recognition capability: The preprocessing process enables... (C–N stretching vibration) and The signal-to-noise ratio of the characteristic peak of (C–H bending vibration) was increased from 35 to over 80, avoiding the problem of characteristic peak blurring caused by the superposition of noise and weak signal without preprocessing, and increasing the characteristic peak recognition rate of low concentration sample (1) from 65% to 98%.

[0148] Model generalization: The preprocessed spectral data in this application have a uniform scale (range [-1,1]), which makes the prediction bias of the model smaller (≤8.2%) in different batches of samples and different matrices. In contrast, Comparative Example 3 has poor generalization due to inconsistent signal scales, with a batch-to-batch bias of 14.3%.

[0149] In summary, this application significantly outperforms the traditional HPLC method (Comparative Example 1) in terms of detection efficiency, stability, and adaptability to complex matrices; far surpasses the unoptimized SERS method (Comparative Example 2) in terms of sensitivity and quantitative accuracy; and surpasses the modeling method without pretreatment (Comparative Example 3) in terms of modeling accuracy and generalization. This method retains the advantages of SERS technology—speed and portability—while overcoming the shortcomings of traditional SERS detection, such as poor stability and weak anti-interference ability, through multi-stage optimization, achieving a comprehensive performance improvement in lincomycin detection that is "fast, accurate, stable, and applicable."

Claims

1. A rapid detection method for lincomycin based on surface-enhanced Raman spectroscopy, characterized in that, The method includes the following steps: Step (1) Preparation of crude silver nanoparticle silver sol: Silver nitrate was reduced with neutral hydroxylamine to prepare silver sol containing crude silver nanoparticles with an average particle size of about 50 nm. Step (2), SERS-enhanced substrate preparation: Silver sol containing coarse silver nanoparticles is placed in a centrifuge for treatment. After centrifugation, the silver nanoparticles settle to the bottom, and a supernatant containing silver nanoparticle debris and reaction impurities is formed on the upper layer. The supernatant is removed at a ratio of 80% to 95% of the total volume of the liquid after centrifugation. The remaining sediment is placed in an ultrasonic cleaner for ultrasonic dispersion to obtain silver nanoparticles with uniform particle size, which serve as the SERS-enhanced substrate. Step (3), SERS detection of the mixture: The SERS-enhanced substrate and the lincomycin sample are mixed at a volume ratio of 1:1 to 3:1 to obtain the detection mixture; the detection mixture is subjected to SERS detection under the set detection conditions to obtain the SERS spectrum of the mixture; Step (4), Characteristic Raman Peak Extraction: Extract lincomycin from the SERS spectrum of the mixture. The characteristic Raman peaks of C–N stretching vibrations at the location and The C–H bending vibration characteristic Raman peak at the location; Step (5), Spectral modeling and quantification: Based on the extracted characteristic Raman peaks of lincomycin, the Raman signals related to lincomycin in the SERS spectrum of the mixture are modeled and analyzed by regression algorithm in order to perform quantitative detection of lincomycin.

2. The method as described in claim 1, characterized in that, The specific steps for preparing the crude silver nanoparticle silver sol in step (1) include: mixing 4 mL of 6 mmol / L hydroxylamine hydrochloride aqueous solution with 4 mL of 5 mmol / L sodium hydroxide aqueous solution to obtain a neutral hydroxylamine solution; adding the neutral hydroxylamine solution dropwise to 72 mL of 1 mmol / L silver nitrate aqueous solution and reacting at room temperature for 5 min; then adding 800 μL of 1% wt trisodium citrate solution and continuing to stir the reaction for 4 h to obtain a yellow-green silver sol containing crude silver nanoparticles with an average particle size of about 50 nm.

3. The method as described in claim 2, characterized in that, In step (1), the stirring speed for the reaction continued for 4 hours was 200~500 r / min.

4. The method as described in claim 1, characterized in that, The centrifuge processing parameters in step (2) are: rotation speed 5500 rpm, centrifugation time 10 min; ultrasonic dispersion time 5 min.

5. The method as described in claim 1, characterized in that, The detection conditions set in step (3) are: laser wavelength 785nm, scanning time 500ms, and laser power 200mW.

6. The method as described in claim 1, characterized in that, The lincomycin samples in step (3) include lincomycin aqueous solution, environmental water mixture containing lincomycin, and fermentation broth supernatant containing lincomycin.

7. The method as described in claim 1, characterized in that, In step (3), the pH value of the mixture was measured to be 5.

22.

8. The method as described in claim 1, characterized in that, The characteristic Raman peaks of lincomycin extracted in step (4) are specifically as follows: The characteristic Raman peaks of C–N stretching vibrations at the location and The Raman peaks characteristic of C–H bending vibration at the location.

9. The method as described in claim 1, characterized in that, In step (5) of spectral modeling and quantification, the detection range for lincomycin is set as follows: Furthermore, the relative standard deviation of the SERS intensity of the characteristic Raman peak of lincomycin extracted based on the SERS spectrum of the mixed solution during this process is less than 10%.

10. The method as described in claim 1, characterized in that, The regression algorithm modeling analysis in step (5) includes spectral data preprocessing and data modeling. The preprocessing optimization and noise reduction are used to purify the mixed liquid SERS spectrum so that the lincomycin-related Raman signal in the mixed liquid SERS spectrum can achieve a state of reduced background interference and improved signal stability. The correlation between the characteristic Raman peak of lincomycin and the concentration of lincomycin is established through the data modeling.

Images