Quantum dot fluorescence sensing method for food ingredient detection
Patent Information
- Application Number
- CN202611123151.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-28
- Publication Date
- 2026-10-13
AI Technical Summary
[0005]本发明提供了一种用于食品成分检测的量子点荧光传感方法，旨在解决单一荧光传感在检测结构相似成分时特异性不足的技术问题
[0040]1、通过构建多通道量子点荧光传感阵列，同时捕获荧光强度、寿命与峰位三类正交响应信号，提升了对结构相似食品成分的感知维度；
Smart Images
Figure CN122901145A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of analytical detection technology, specifically relating to a quantum dot fluorescence sensing method for detecting food components. Background Technology
[0002] With the increasing demands for food safety supervision, rapid and accurate detection technologies for food components have become crucial for safeguarding public health and ensuring industry compliance. Fluorescence sensing technology, due to its high sensitivity, rapid response, and ease of operation, is widely used in the detection of harmful substances, additives, and nutrients in food. Quantum dots, as a novel type of fluorescent nanoprobe, improve the signal quality and detection limit of fluorescence sensing due to their tunable emission wavelength, broad excitation and narrow emission spectrum, and excellent photostability. However, when faced with structurally similar homologues or isomers, detection strategies relying solely on quantum dot fluorescence intensity or emission peak position are insufficient to effectively distinguish target analytes, leading to insufficient specificity and severe cross-reactivity, thus limiting their practical application in complex food matrices.
[0003] Sensing strategies based on multimodal signal fusion offer a new approach to improving detection specificity. Surface-enhanced Raman scattering (SERS) technology, through the local electromagnetic field enhancement effect, can acquire molecular vibrational fingerprint information and has extremely high resolution for minute differences in chemical structure. If quantum dot fluorescence sensing can be synergistically integrated with SERS technology to simultaneously capture the electronic transition characteristics (fluorescence) and vibrational energy level characteristics (Raman) of target molecules, it is expected to construct a more comprehensive molecular recognition spectrum from different physical dimensions. Existing dual-modal sensing systems mostly employ step-by-step measurement or independent signal processing methods, failing to achieve simultaneous spatiotemporal acquisition of fluorescence and Raman signals and deep semantic fusion, resulting in insufficient exploitation of information redundancy or complementarity.
[0004] Traditional classification methods struggle to effectively integrate heterogeneous high-dimensional features such as fluorescence intensity, fluorescence lifetime, Raman shift, and peak intensity ratio. They also lack adaptive decision-making mechanisms to address the complexity of food components and matrix interference, resulting in weak model generalization and high misclassification rates. When faced with real-world samples exhibiting similar compositions, large concentration fluctuations, or the presence of coexisting interfering substances, existing single-modal or simple fusion strategies fail to consistently output high-confidence classification results, hindering the development of high-throughput, high-precision on-site food component detection technologies. Therefore, a novel quantum dot fluorescence sensing method integrating dual-modal simultaneous sensing and intelligent multi-source feature fusion is urgently needed to overcome the technical challenge of specifically identifying structurally similar components. Summary of the Invention
[0005] This invention provides a quantum dot fluorescence sensing method for food component detection, aiming to solve the technical problem of insufficient specificity of single fluorescence sensing when detecting structurally similar components. This method constructs a multi-channel quantum dot fluorescence sensing array, combines a multi-source feature extraction and fusion mechanism, and introduces a decision tree classification model based on information gain ratio optimization to achieve high-precision and high-specificity identification and quantitative analysis of structurally highly similar components in food.
[0006] This invention provides a quantum dot fluorescence sensing method for detecting food components, comprising:
[0007] A fluorescence sensing array composed of quantum dots with at least three different surface modifications was constructed, each quantum dot exhibiting differentiated response characteristics to the target food component;
[0008] The food sample solution to be tested is brought into contact with each quantum dot in the fluorescence sensing array to obtain the fluorescence intensity response signal, fluorescence lifetime decay curve and fluorescence peak shift data of each quantum dot under the irradiation of the excitation light source.
[0009] The fluorescence intensity response signal, fluorescence lifetime decay curve, and fluorescence peak shift data are subjected to baseline correction, noise filtering, and normalization to form a standardized multidimensional response feature vector.
[0010] Principal component analysis was used to reduce the dimensionality of the multidimensional response feature vector, and the first three principal components were extracted as the core discriminant features.
[0011] The core discriminative features are input into a pre-trained decision tree classification model. The decision tree classification model uses information gain rate as the node splitting criterion, constructs a multi-layer classification path through recursive splitting, and outputs the category determination result and concentration estimate of the component to be tested.
[0012] Preferably, a fluorescence sensing array composed of quantum dots with at least three different surface modifications is constructed, including:
[0013] One or more combinations of cadmium sulfide quantum dots, cadmium selenide quantum dots, and cadmium telluride quantum dots are selected.
[0014] The surface ligands of quantum dots were modified with thioglycolic acid, cysteine and glutathione, respectively, to regulate the affinity and selectivity of quantum dots with different food components;
[0015] The quantum dots are fixed on a porous silica microsphere carrier, and a stable load is achieved through the dual effects of physical adsorption and coordination bonding.
[0016] Preferably, the food sample solution to be tested is brought into contact with each quantum dot in the fluorescence sensing array to obtain the fluorescence intensity response signal, fluorescence lifetime decay curve, and fluorescence peak shift data of each quantum dot under excitation light source irradiation, including:
[0017] A wavelength-tunable ultraviolet-visible continuous light source is used as the excitation source;
[0018] Fluorescence intensity response signals were acquired using a time-resolved fluorescence detection mode.
[0019] Fluorescence lifetime decay curves were obtained by time-correlated single-photon counting, and fast decay components and slow decay components and their corresponding amplitudes were obtained by fitting with a double exponential decay model.
[0020] Fluorescence peak shift data were determined by Gaussian fitting, and the shift relative to the blank control group was calculated.
[0021] Preferably, the fluorescence intensity response signal, fluorescence lifetime decay curve, and fluorescence peak shift data are subjected to baseline correction, noise filtering, and normalization processing to form a standardized multidimensional response feature vector, including:
[0022] The fluorescence intensity response signal, after baseline correction, is expressed as: ,in The fluorescence intensity after the sample is added. The fluorescence intensity of the blank solution;
[0023] Fluorescence lifetime weighted average As a lifetime characteristic parameter;
[0024] and These represent the amplitudes of the fast decay component and the slow decay component, respectively. and The corresponding fluorescence lifetime
[0025] Fluorescence peak shift The feature vector is directly incorporated at the nanometer level; for the three quantum dots, a 9-dimensional original feature vector is formed, denoted as and . , , , These are the baseline-corrected fluorescence intensity response values of the three quantum dots. , , These are the weighted average fluorescence lifetimes of the three quantum dots, respectively. , , These represent the fluorescence peak shifts of the three quantum dots.
[0026] Preferably, principal component analysis is used to reduce the dimensionality of the multidimensional response feature vector, and the first three principal components are extracted as core discriminant features, including:
[0027] The original 9-dimensional eigenvectors are centered, and the covariance matrix is constructed and eigenvalues are decomposed.
[0028] Select the top three principal components whose cumulative variance contribution rate is not less than 90%;
[0029] Each sample is mapped to the first three principal components, forming a three-dimensional discriminant space, and the three-dimensional principal component score vector is obtained as the core discriminant feature.
[0030] Preferably, the core discriminative features are input into a pre-trained decision tree classification model. The decision tree classification model uses information gain rate as the node splitting criterion, constructs a multi-layer classification path through recursive splitting, and outputs the category determination result and concentration estimate of the component to be tested, including:
[0031] Collect a standard sample dataset covering the target food ingredients and their structural analogs, with no fewer than 50 samples for each ingredient class;
[0032] For each sample, the aforementioned sensing and feature extraction process is performed to generate the corresponding three-dimensional principal component feature vector;
[0033] Using component categories as labels, a decision tree is constructed using the C4.5 algorithm, with the information gain ratio as the splitting criterion.
[0034] The model performance is evaluated through cross-validation, requiring an accuracy of at least 95% in distinguishing structurally similar components.
[0035] Preferably, the target food ingredient is a sweetener in food additives, specifically including sodium saccharin, acesulfame potassium, aspartame, sucralose, and neotame; the structurally similar ingredients refer to compounds with similar molecular skeletons and small differences in functional groups, with a molar mass difference of less than 50 grams per mole, and a retention time overlap of more than 70% under conventional chromatographic conditions.
[0036] Preferably, when outputting concentration estimates, for components whose categories have been determined, a partial least squares regression model is used to establish a quantitative relationship between the principal component scores of the components and the actual concentrations. The determination coefficient of the regression model is not less than 0.98, and the prediction relative error is controlled within 5%.
[0037] Preferably, the method further includes a system calibration step: before each test, the quantum dot sensing array is tested for response stability using a standard mixed solution. If the fluorescence intensity variation coefficient of any quantum dot is greater than 5%, the sensor regeneration procedure is triggered to restore its initial response performance through ultrasonic cleaning and ligand reconstruction.
[0038] Preferably, the entire detection process is completed within the microfluidic chip, which integrates a sample inlet, a mixing reaction chamber, a sensing detection area, and a waste liquid outlet. The detection area is equipped with an optical window, and the fluid drive uses a constant pressure pump.
[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0040] 1. By constructing a multi-channel quantum dot fluorescence sensing array, three types of orthogonal response signals—fluorescence intensity, lifetime, and peak position—we can simultaneously capture these signals, thereby enhancing our perception of structurally similar food components.
[0041] 2. Principal component analysis is used to compress and denoise high-dimensional response features, retaining the most discriminative information; a decision tree algorithm based on information gain rate is introduced to avoid the misjudgment problem of traditional thresholding or simple pattern recognition in complex component systems, and to distinguish between highly similar sweeteners such as sodium saccharin and acesulfame potassium.
[0042] 3. Combining partial least squares regression model, qualitative identification and quantitative analysis are completed simultaneously, with a detection limit as low as 10 nanograms per milliliter and a linear range covering 0.1 micrograms per milliliter to 100 micrograms per milliliter;
[0043] 4. The entire method requires no complex pretreatment, has a short detection cycle, and is suitable for rapid on-site screening and online monitoring of food safety, overcoming the technical problems of poor specificity and susceptibility to interference of single fluorescence sensing. Attached Figure Description
[0044] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;
[0045] Figure 2 This is a schematic diagram of the core principle framework of decision tree classification based on multi-source feature fusion and information gain ratio optimization in this invention;
[0046] Figure 3 This is a flowchart illustrating the logical flow of the multi-channel quantum dot fluorescence sensing array response signal acquisition and multi-dimensional feature construction in this invention.
[0047] Figure 4 This is a flowchart illustrating the logical flow of principal component analysis dimensionality reduction and three-dimensional discriminative feature space mapping in this invention.
[0048] Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow between the microfluidic chip integrated system and the embedded decision tree model deployment in this invention. Detailed Implementation
[0049] refer to Figures 1 to 5This invention provides a quantum dot fluorescence sensing method for food component detection, aiming to solve the technical problem of insufficient specificity of single fluorescence sensing when detecting structurally similar components. The method constructs a multi-channel quantum dot fluorescence sensing array to simultaneously acquire three types of orthogonal response signals: fluorescence intensity, fluorescence lifetime, and fluorescence peak position, forming high-dimensional raw response data. Subsequently, the raw data undergoes standardized preprocessing and principal component analysis for dimensionality reduction, extracting the most discriminative core features. Finally, the dimensionality-reduced features are input into a decision tree classification model using information gain rate as the splitting criterion, achieving accurate qualitative identification of structurally similar food components. A partial least squares regression model is then used to complete quantitative concentration estimation.
[0050] The method strictly follows the following sequence of steps:
[0051] S1, Construct a fluorescence sensing array composed of quantum dots with at least three different surface modifications;
[0052] S2, the food sample solution to be tested is brought into contact with each quantum dot in the fluorescence sensing array to obtain the fluorescence intensity response signal, fluorescence lifetime decay curve and fluorescence peak shift data of each quantum dot under the irradiation of the excitation light source.
[0053] S3, perform baseline correction, noise filtering and normalization on the above three types of raw response data to form a standardized multidimensional response feature vector;
[0054] S4, Principal component analysis is used to reduce the dimensionality of the multidimensional response feature vector, and the first three principal components are extracted as core discriminant features.
[0055] S5. The core discriminative features are input into a pre-trained decision tree classification model. The decision tree classification model uses the information gain rate as the node splitting criterion, constructs a multi-layer classification path through recursive segmentation, and finally outputs the category determination result and concentration estimate of the component to be tested.
[0056] In step S1, a fluorescence sensing array composed of at least three different surface-modified quantum dots is constructed. The quantum dots are selected from one or more combinations of cadmium sulfide quantum dots, cadmium selenide quantum dots, and cadmium telluride quantum dots, with a particle size ranging from 2 nm to 6 nm. The surface ligands of each quantum dot are modified with thioglycolic acid, cysteine, and glutathione, respectively, to regulate their affinity and selectivity with different food components. The surface of the thioglycolic acid-modified quantum dots is rich in carboxyl functional groups, exhibiting a strong electrostatic attraction to amino-containing sweeteners such as aspartame; the cysteine-modified quantum dots contain both amino and carboxyl groups, allowing them to form hydrogen bonds with the sulfonyl groups in sodium saccharin; and the glutathione-modified quantum dots, due to their thiol and carboxyl groups, exhibit specific recognition ability for polyhydroxy structures such as sucralose.
[0057] The differentiated surface chemical environments of the three quantum dots enable them to generate non-redundant fluorescence response modes when facing structurally similar components, thus forming the basis for multi-channel sensing. The quantum dots are immobilized on a porous silica microsphere carrier with a pore size of 10 to 30 nanometers and a specific surface area of not less than 200 square meters per gram. The quantum dots are stably loaded onto the inner wall of the carrier pores through a combination of physical adsorption and coordination bonding, preventing aggregation or detachment during liquid flow or stirring, thus ensuring the long-term stability and repeatability of the sensing signal.
[0058] In step S2, the food sample solution to be tested is brought into contact with each quantum dot in the fluorescence sensing array to obtain the fluorescence intensity response signal, fluorescence lifetime decay curve, and fluorescence peak shift data of each quantum dot under the illumination of the excitation source. The excitation source is a wavelength-tunable ultraviolet-visible continuous light source, with an excitation wavelength covering the range of 300 nm to 450 nm and a step interval of 5 nm. For each quantum dot, its fluorescence emission spectrum at different excitation wavelengths is scanned sequentially, and the wavelength position and intensity value corresponding to the maximum emission intensity are recorded. The fluorescence intensity response signal is acquired using a time-resolved fluorescence detection mode, with the integration time set to 10 ms to 500 ms and the sampling frequency set to 1000 Hz to suppress background scattering interference and improve the signal-to-noise ratio. The fluorescence lifetime decay curve is obtained by time-correlated single-photon counting, with a system time resolution better than 100 picoseconds and a cumulative photon count of not less than 10,000 in a single measurement. The decay curve is fitted using a double exponential decay model, expressed as:
[0059] ;
[0060] For time fluorescence intensity at time, and These represent the amplitudes of the fast decay component and the slow decay component, respectively. and This represents the corresponding fluorescence lifetime. The weighted average fluorescence lifetime. Defined as:
[0061] ;
[0062] The parameters comprehensively reflect the polarity, viscosity, and interaction strength between the quantum dot's microenvironment and the target molecule. Fluorescence peak shift data are determined by Gaussian fitting to determine the peak wavelength, and the shift relative to the blank control group (i.e., the buffer solution without sample) is calculated. The shift direction and amplitude are directly related to the electron donor-acceptor interaction or steric hindrance effect between the target molecule and the quantum dot surface ligands. The entire detection process is completed within a microfluidic chip, which integrates a sample inlet, a mixing reaction chamber, a sensing detection area, and a waste outlet. The reaction chamber volume is 10 μL, ensuring sufficient contact between the sample and the sensing array while minimizing reagent consumption. The detection area features a quartz optical window for vertical excitation light introduction and lateral fluorescence signal acquisition. A constant-pressure pump is used for fluid drive, with the flow rate controlled between 0.5 μL / s and 5 μL / s to ensure consistent reaction time and signal reproducibility.
[0063] In step S3, the three types of raw response data are preprocessed, including baseline correction, noise filtering, and normalization, to form a standardized multidimensional response feature vector. Baseline correction uses a polynomial fitting method to eliminate instrument drift and background fluorescence interference; the fitting order is dynamically adjusted according to the signal smoothness, typically second or third order. Noise filtering uses a wavelet thresholding algorithm, selecting the Daubechies fourth-order wavelet basis function, and a soft thresholding strategy to preserve signal abrupt changes. Normalization uses the response value of the blank control group as a benchmark, and relativizes the three types of response parameters for each quantum dot. Specifically, the fluorescence intensity response is expressed as... , The fluorescence intensity after the sample is added. Fluorescence intensity of blank solution; weighted average of fluorescence lifetime. The fitting results are used directly because they are absolute physical quantities; fluorescence peak shift. The eigenvectors are directly incorporated at the nanometer level. Thus, for the three types of quantum dots, a total of 9 original eigenvectors are formed, denoted as... , , , These are the baseline-corrected fluorescence intensity response values of the three quantum dots. , , These are the weighted average fluorescence lifetimes of the three quantum dots, respectively. , , These represent the fluorescence peak shifts of the three quantum dots.
[0064] In step S4, principal component analysis (PCA) is used to reduce the dimensionality of the original 9-dimensional feature vector, extracting the first three principal components as core discriminant features. PCA obtains a set of orthogonal principal component axes by centering and eigenvalue decomposition of the covariance matrix of the original feature matrix. Let the original feature matrix be... , If the sample size is , then the centered matrix is: , This represents the mean vector for each feature dimension. The covariance matrix is also shown. The nine eigenvalues are arranged in descending order as follows: The corresponding eigenvectors form the projection matrix. The cumulative variance contribution rate of the first three principal components. Defined as:
[0065] ;
[0066] For the first 1 eigenvalue, requiring 2 eigenvalues, ... This ensures that the main part of the original information is preserved. The three-dimensional principal component score vector for each sample is:
[0067] ;
[0068] The scores for the first, second, and third principal components are respectively. This represents the original 9-dimensional feature vector for a single sample. The 3D vector forms the coordinate points in the discriminant space, effectively compressing the data dimensionality while enhancing inter-class separability.
[0069] In step S5, the core discriminative features are input into a pre-trained decision tree classification model. This model uses information gain rate as the node splitting criterion, constructs multi-layer classification paths through recursive splitting, and finally outputs the category determination result and concentration estimate of the component to be tested. The training process of the decision tree classification model includes: collecting a standard sample dataset covering the target food component and its structural analogues. The target component is a sweetener in food additives, specifically including sodium saccharin, acesulfame potassium, aspartame, sucralose, and neotame; structurally similar components refer to compounds with similar molecular skeletons and small differences in functional groups, with a molar mass difference of less than 50 g / mol and a retention time overlap of more than 70% under conventional chromatographic conditions. There are no fewer than 50 samples for each component, covering a concentration range from 0.1 μg / mL to 100 μg / mL. The aforementioned sensing and feature extraction process is performed on each sample to generate a corresponding three-dimensional principal component feature vector. A decision tree is constructed using the C4.5 algorithm with the component category as the label, and the splitting criterion is information gain rate. Defined as:
[0070] ;
[0071] Information gain , The set of all possible values for attribute A A subset of samples where attribute A has a value of v. Let S be the total number of samples in the sample set. for The number of samples, for Shannon entropy; split information The minimum number of leaf node samples is set to 5, and the maximum tree depth is set to 8 to prevent overfitting. After model training, its classification accuracy is evaluated using 10-fold cross-validation, requiring an accuracy of at least 95% in distinguishing structurally similar components. In actual deployment, the decision tree classification model is embedded in an ARM Cortex-A72 architecture processor with a clock speed of 1.8 GHz and four gigabytes of RAM. The model inference time is no more than 200 milliseconds, supporting real-time online detection.
[0072] After outputting the category determination results, the concentration estimation module is activated for the components whose categories have been determined. This concentration estimation module uses a partial least squares regression model to establish a quantitative relationship between the principal component score and the actual concentration of the component. Partial least squares regression constructs a latent variable space by maximizing the covariance between the principal component score and the concentration label, effectively handling the problem of feature collinearity. The coefficient of determination of the regression model...
[0073] The value should be no less than 0.98, and the relative prediction error should be controlled within 5%. The concentration estimate and category determination results are output simultaneously to form a complete qualitative-quantitative detection report.
[0074] The method also includes a system calibration step: before each detection, the response stability of the quantum dot sensing array is verified using a standard mixed solution. The standard mixed solution contains five target sweeteners, each at a concentration of 10 μg / mL. The fluorescence intensity response of each quantum dot to the standard solution is measured, and its coefficient of variation is calculated. If the fluorescence intensity coefficient of variation of any quantum dot is greater than 5%, a sensor regeneration procedure is triggered. The regeneration procedure includes: first, rinsing the sensing area with deionized water for 5 minutes at a flow rate of 3 μL / s; then, injecting a 52% alcohol solution for ultrasonic cleaning for 2 minutes at a power of 40 W; finally, introducing a buffer solution containing the original ligands at a flow rate of 1 μL / s for 10 minutes to reconstruct the ligands on the quantum dot surface and restore its initial response performance. After calibration, the system automatically records the current baseline parameters for normalization processing in subsequent sample detections.
[0075] In summary, this invention captures multimodal fluorescence responses using a multi-channel quantum dot array, compresses feature dimensions using principal component analysis, achieves highly specific classification by combining an information gain rate-optimized decision tree, and completes precise quantification using partial least squares regression. The entire process is integrated within a microfluidic chip, enabling real-time analysis with an embedded processor. The detection cycle is less than 5 minutes, and the detection limit is as low as 10 nanograms per milliliter. It is suitable for rapid on-site screening and online monitoring of food safety, overcoming the problems of poor specificity and susceptibility to interference associated with single fluorescence sensors.
[0076] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A quantum dot fluorescence sensing method for detecting food components, characterized in that, include: A fluorescence sensing array composed of quantum dots with at least three different surface modifications was constructed, each quantum dot exhibiting differentiated response characteristics to the target food component; The food sample solution to be tested is brought into contact with each quantum dot in the fluorescence sensing array to obtain the fluorescence intensity response signal, fluorescence lifetime decay curve and fluorescence peak shift data of each quantum dot under the irradiation of the excitation light source. The fluorescence intensity response signal, fluorescence lifetime decay curve, and fluorescence peak shift data are subjected to baseline correction, noise filtering, and normalization to form a standardized multidimensional response feature vector. Principal component analysis was used to reduce the dimensionality of the multidimensional response feature vector, and the first three principal components were extracted as the core discriminant features. The core discriminative features are input into a pre-trained decision tree classification model. The decision tree classification model uses information gain rate as the node splitting criterion, constructs a multi-layer classification path through recursive splitting, and outputs the category determination result and concentration estimate of the component to be tested.
2. The quantum dot fluorescence sensing method for food component detection according to claim 1, characterized in that, Constructing a fluorescence sensing array composed of quantum dots with at least three different surface modifications, including: One or more combinations of cadmium sulfide quantum dots, cadmium selenide quantum dots, and cadmium telluride quantum dots are selected. The surface ligands of quantum dots were modified with thioglycolic acid, cysteine and glutathione, respectively, to regulate the affinity and selectivity of quantum dots with different food components; The quantum dots are fixed on a porous silica microsphere carrier, and a stable load is achieved through the dual effects of physical adsorption and coordination bonding.
3. The quantum dot fluorescence sensing method for food component detection according to claim 2, characterized in that, The food sample solution to be tested is brought into contact with each quantum dot in the fluorescence sensing array to obtain the fluorescence intensity response signal, fluorescence lifetime decay curve, and fluorescence peak shift data of each quantum dot under the irradiation of the excitation source, including: A wavelength-tunable ultraviolet-visible continuous light source is used as the excitation source; Fluorescence intensity response signals were acquired using a time-resolved fluorescence detection mode. Fluorescence lifetime decay curves were obtained by time-correlated single-photon counting, and fast decay components and slow decay components and their corresponding amplitudes were obtained by fitting with a double exponential decay model. Fluorescence peak shift data were determined by Gaussian fitting, and the shift relative to the blank control group was calculated.
4. The quantum dot fluorescence sensing method for food component detection according to claim 3, characterized in that, The fluorescence intensity response signal, fluorescence lifetime decay curve, and fluorescence peak shift data are subjected to baseline correction, noise filtering, and normalization to form a standardized multidimensional response feature vector, including: The fluorescence intensity response signal, after baseline correction, is expressed as: , The fluorescence intensity after the sample is added. The fluorescence intensity of the blank solution; Fluorescence lifetime weighted average As a lifetime characteristic parameter; and These represent the amplitudes of the fast decay component and the slow decay component, respectively. and The corresponding fluorescence lifetime Fluorescence peak shift The feature vector is directly incorporated at the nanometer level; for the three quantum dots, a 9-dimensional original feature vector is formed, denoted as and . , , , These are the baseline-corrected fluorescence intensity response values of the three quantum dots. , , These are the weighted average fluorescence lifetimes of the three quantum dots, respectively. , , These represent the fluorescence peak shifts of the three quantum dots.
5. The quantum dot fluorescence sensing method for food component detection according to claim 4, characterized in that, Principal component analysis (PCA) is used to reduce the dimensionality of the multidimensional response feature vector, and the first three principal components are extracted as core discriminant features, including: The original 9-dimensional eigenvectors are centered, and the covariance matrix is constructed and eigenvalues are decomposed. Select the top three principal components whose cumulative variance contribution rate is not less than 90%; Each sample is mapped to the first three principal components, forming a three-dimensional discriminant space, and the three-dimensional principal component score vector is obtained as the core discriminant feature.
6. The quantum dot fluorescence sensing method for food component detection according to claim 5, characterized in that, The core discriminative features are input into a pre-trained decision tree classification model. The decision tree classification model uses information gain as the node splitting criterion, constructs multi-layer classification paths through recursive splitting, and outputs the category determination result and concentration estimate of the component to be tested, including: Collect a standard sample dataset covering the target food ingredients and their structural analogs, with no fewer than 50 samples for each ingredient class; For each sample, the aforementioned sensing and feature extraction process is performed to generate the corresponding three-dimensional principal component feature vector; Using component categories as labels, a decision tree is constructed using the C4.5 algorithm, with the information gain ratio as the splitting criterion. The model performance is evaluated through cross-validation, requiring an accuracy of at least 95% in distinguishing structurally similar components.
7. The quantum dot fluorescence sensing method for food component detection according to claim 6, characterized in that, The target food ingredients are sweeteners in food additives, specifically including sodium saccharin, acesulfame potassium, aspartame, sucralose, and neotame; the structurally similar ingredients refer to compounds with similar molecular skeletons, minimal differences in functional groups, a molar mass difference of less than 50 grams per mole, and a retention time overlap of more than 70% under conventional chromatographic conditions.
8. The quantum dot fluorescence sensing method for food component detection according to claim 7, characterized in that, When outputting concentration estimates, for components whose categories have been determined, a partial least squares regression model is used to establish a quantitative relationship between the principal component scores of the components and the actual concentrations. The determination coefficient of the regression model is not less than 0.98, and the prediction relative error is controlled within 5%.
9. The quantum dot fluorescence sensing method for food component detection according to claim 8, characterized in that, The method also includes a system calibration step: before each test, the quantum dot sensing array is checked for response stability using a standard mixed solution. If the fluorescence intensity variation coefficient of any quantum dot is greater than 5%, the sensor regeneration procedure is triggered, and its initial response performance is restored through ultrasonic cleaning and ligand reconstruction.
10. The quantum dot fluorescence sensing method for food component detection according to claim 9, characterized in that, The entire detection process is completed within the microfluidic chip, which integrates a sample inlet, a mixing reaction chamber, a sensing detection area, and a waste liquid outlet. The detection area is equipped with an optical window, and the fluid is driven by a constant pressure pump.