Intelligent wine detection method based on multispectral fusion

Abstract

The invention discloses an intelligent liquor detection method based on multispectral fusion, which comprises multispectral excitation and acquisition, spectral data fusion processing and multi-index parallel analysis, and relates to the technical field of liquor quality detection. According to the intelligent wine detection method based on multispectral fusion, five-dimensional index analysis of the content of the saccharum arundinaceum, the alcoholic strength, the authenticity, the brand and the year can be synchronously completed through a three-stage detection process of ultraviolet excitation, fluorescence capture and near-infrared verification. Detection data is encrypted and transmitted to a transaction platform through a block chain, and a reliable spectrum traceability voucher is formed by using the non-tampering characteristic of the block chain technology, so that the core pain points of pricing basis deficiency, difficulty in authenticity identification, year false bidding and the like in Baijiu transaction are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of liquor quality detection, and specifically to an intelligent liquor detection method based on multi-spectral fusion. Background Art

[0002] In the liquor trading industry, the continued growth of the market and escalating consumer demand are driving higher demands for precision, efficiency, and comprehensiveness in liquor quality testing. Traditional liquor testing technologies have numerous limitations. For example, sensory evaluation relies on manual experience, is highly subjective, and inefficient, making it difficult to rapidly test the large volumes of liquor samples required by liquor trading centers. While chemical analysis offers high accuracy, the testing process is complex and time-consuming, requiring specialized laboratory equipment and technicians, making it difficult to rapidly test on-site. When it comes to authenticating liquor, existing technologies struggle to quickly and accurately identify the endless stream of counterfeit products on the market. The lack of scientific, precise, and rapid methods for authenticating liquor vintages makes it difficult to accurately determine the vintage during sauce-flavored liquor transactions, leading to a lack of reliable basis for pricing and severely impacting market fairness and regulation. When benchmarking quality against Moutai Feitian, there's currently no comprehensive indicator modeling and comparison system. Many liquor companies lack a scientific basis for pricing, leading to market price chaos.

[0003] Existing technologies for alcohol content testing struggle to meet the demands of refined trading, failing to achieve the high-precision detection requirements of 0.1 vol. Furthermore, faced with the vast number and variety of nearly 10,000 liquor samples encountered in liquor trading centers, rapid and accurate brand identification technology remains a major industry gap, significantly limiting the efficiency and accuracy of liquor trading. These challenges are intertwined across every stage of the liquor trading process, including storage, grading, trading, display, blending, and storage inspection, severely hindering the healthy and efficient development of the liquor trading industry. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an intelligent wine detection method based on multi-spectral fusion, which solves the above-mentioned problems.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent wine detection method based on multi-spectral fusion, comprising the following steps: S1. Multispectral excitation and acquisition: The wine samples were excited in sequence using a three-wavelength UV LED light source of 265 nm, 275 nm, and 310 nm in a marquee mode, with an excitation time interval of 0.5 seconds. The fluorescence emission spectrum in the range of 340-1100nm and the near-infrared spectrum in the four bands of 850nm, 900nm, 950nm and 1000nm were collected simultaneously to generate four original spectral curves; The four-band near-infrared detection unit includes four detection bands: 850nm, 900nm, 950nm, and 1000nm.

[0006] S2. Spectral data fusion processing: De-noising (Savitzky-Golay filtering) and normalization processing are performed on the original spectral curve; Extract characteristic wavelengths: To detect the amount of tar, the double peak intensity at 425nm excited by 265nm, the shoulder peak intensity at 485nm excited by 310nm, and the hydroxyl absorption intensity at 950nm need to be extracted; to detect the alcohol content, the third-order harmonic absorption slope of the O-H bond at 900-1000nm needs to be extracted; The feature data is fused using the weighted average method, and the weights are dynamically allocated according to the feature signal-to-noise ratio; S3. Multi-index parallel analysis: Through the three-level process of "ultraviolet excitation-fluorescence capture-near infrared verification", the similarity of Maotai content, alcohol content, authenticity determination, brand logo and year value can be output simultaneously within 30 seconds.

[0007] Preferably, the detection of the amount of tartar content specifically includes: Establish a Moutai benchmark spectral library, including the 265nm / 425nm double peak area ratio, 310nm / 485nm shoulder peak offset, and 950nm absorption peak half width; Calculate similarity: \text{Similarity\%}=α·\frac{\|A_{\text{Sample}}-A_{\text{Benchmark}}\|}{A_{\text{Benchmark}}}+β·\text{SIM}(F_{\text{Sample}},F_{\text{Benchmark}}); Where α=0.4, β=0.6, SIM is the fluorescence spectrum correlation coefficient, and the output is 70%-100% quantitative results.

[0008] Preferably, the alcohol content detection includes: Fit the OH bond third-order harmonic absorption curve in the near-infrared band of 900-1000nm and calculate the integral area S OH ; The background fluorescence intensity of ethanol at 275 nm is 275 Correction interference: \text{Alcohol content}=k_1·S_{OH}+k_2·I_{275}+b; (k1=0.78,k2=-0.15,b=3.2), achieving ±0.1vol% accuracy.

[0009] Preferably, the brand identification includes: The pre-stored brand fingerprint database includes the 265nm / 365nm peak height ratio and 310nm / 415nm peak position difference of Wuliangye, and the 310nm / 450nm peak half width of Luzhou Laojiao; A one-dimensional convolutional neural network (CNN) model was constructed: the input layer received fused spectral data, the convolution kernel size was 5×1, the pooling layer used maximum pooling, and the fully connected layer output the brand classification probability.

[0010] Preferably, the year identification includes: Measure the fluorescence lifetime decay rate τ of humic acid under 275nm excitation (time resolution 0.1ns) and obtain the decay slope dτ / dt; Calculate the red shift of the fluorescence peak of ester excited at 310 nm (accuracy ±0.5 nm); Through regression model: \text{Year}=e^{a·(dτ / dt)+b·Δλ+c} (a=0.32, b=1.75, c=-2.1), interpreting the aging period of 3-30 years, with an error of ±1 year.

[0011] Preferably, the data security processing includes: Hash the spectral feature values ​​to generate a digital fingerprint; Through encrypted transmission on the Hyperledger Fabric blockchain platform, a spectral traceability certificate containing a timestamp and device ID is generated on the transaction chain.

[0012] The present invention also includes an intelligent wine detection device based on multi-spectral fusion, including: Optical modules: UV light source: 265nm, 275nm, 310nm LED array (power 10mW±5%), triggered in a cyclic manner in a marquee mode; Fluorescence acquisition: Hamamatsu H11706-01 photomultiplier tube, detection wavelength 340-1100 nm; Near-infrared unit: InGaAs detector (response range 850-1700nm), bandpass filter limited to four bands of 850 / 900 / 950 / 1000nm; Control Module: Embedded processor (ARMCortex-A72) with built-in multi-spectral fusion algorithm; The communication interface supports 4G / Bluetooth dual-mode transmission and blockchain encryption.

[0013] Preferably, the control logic of the marquee mode is: Timing control: 265nm light → delay 100ms to collect fluorescence → off → 275nm light → delay 100ms to collect → off → 310nm light → delay 100ms to collect; A single cycle takes 300ms and is repeated 10 times to take the average value of the spectrum.

[0014] The present invention provides an intelligent wine detection method based on multispectral fusion. Compared with the existing technology, it has the following advantages: 1. This intelligent liquor detection method based on multi-spectral fusion is fast and efficient: through a unique three-level detection flow and advanced algorithms, it can complete the five-dimensional index analysis of liquor content, alcohol content, authenticity, brand, and year within 30 seconds. Compared with traditional detection methods, it greatly improves the detection efficiency and meets the needs of liquor trading centers for rapid detection of large quantities of liquor samples. 2. This intelligent liquor detection method based on multi-spectral fusion is accurate and reliable: it integrates multi-spectral fusion technology to detect and analyze liquor from multiple dimensions. Combined with targeted algorithms and models, the alcohol content detection accuracy reaches ±0.1vol%. Year identification can accurately analyze aging years of 3-30 years. Maotai content detection can accurately quantify and output similarity percentages. Brand recognition can achieve millisecond-level retrieval of a brand library of 10,000 brands, ensuring the accuracy and reliability of test results and providing a scientific pricing basis for liquor transactions.

[0015] 3. This intelligent liquor detection method based on multi-spectral fusion has comprehensive and diverse functions: it realizes one-stop detection of key indicators of liquor, covering Maotai content, alcohol content, authenticity, brand, year and other aspects, meeting the diverse detection needs in scenarios such as liquor trading, storage, and grading, and providing a comprehensive detection solution for the liquor industry.

[0016] 4. This intelligent alcohol detection method based on multi-spectral fusion is easy to carry and operate: it is designed as a portable device, which is convenient for use in different scenarios. The operation process is simple and does not require professional laboratory equipment and complex operating skills. It lowers the threshold for use and is conducive to its widespread promotion and application in the liquor industry.

[0017] 5. This intelligent liquor detection method based on multi-spectral fusion has secure and reliable data: the detection data is encrypted and transmitted to the trading platform through blockchain, forming an unalterable spectral traceability certificate, ensuring the security and credibility of the data, effectively solving the data trust problem in liquor trading, and promoting the standardized development of the liquor trading market. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the steps of the method of the present invention; Figure 2 Schematic diagram of the spectrum curve of the method of the present invention; Figure 3 This is a schematic diagram of the structural method flow of the present invention; Figure 4 This is a schematic diagram of the CNN model architecture of the present invention. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] See also Figure 1-4 The embodiment of the present invention provides a technical solution: an intelligent wine detection method based on multi-spectral fusion, comprising the following steps: The following steps are involved: S1. Multispectral excitation and acquisition: The wine samples were excited in sequence using a three-wavelength UV LED light source of 265 nm, 275 nm, and 310 nm in a marquee mode, with an excitation time interval of 0.5 seconds. The three-wavelength ultraviolet excitation light sources are 265nm, 275nm, and 310nm. The 265nm ultraviolet light source can specifically excite aromatic amino acids and nucleotides in wine samples, which play an important role in the formation of liquor flavor and quality characterization; the 275nm band targets phenols and polycyclic aromatic hydrocarbons, whose content and types are closely related to the taste and quality of liquor; and the 310nm wavelength effectively captures the fluorescence characteristics of flavonoids and B vitamins, which have a significant impact on the antioxidant properties and other quality characteristics of liquor. The three work together to form a three-dimensional fluorescence excitation-emission matrix, accurately analyzing the composition of trace organic matter in liquor from multiple dimensions and constructing a unique optical fingerprint of liquor.

[0021] The fluorescence emission spectrum in the range of 340-1100nm and the near-infrared spectrum in the four bands of 850nm, 900nm, 950nm and 1000nm were collected simultaneously to generate four original spectral curves; The four-band near-infrared detection unit includes four detection bands: 850nm, 900nm, 950nm, and 1000nm.

[0022] S2. Spectral data fusion processing: De-noising (Savitzky-Golay filtering) and normalization processing are performed on the original spectral curve; Extract characteristic wavelengths: To detect the amount of tar, the double peak intensity at 425nm excited by 265nm, the shoulder peak intensity at 485nm excited by 310nm, and the hydroxyl absorption intensity at 950nm need to be extracted; to detect the alcohol content, the third-order harmonic absorption slope of the O-H bond at 900-1000nm needs to be extracted; The feature data is fused using the weighted average method, and the weights are dynamically allocated according to the feature signal-to-noise ratio; The four LED light sources in these four wavelength bands were sequentially illuminated in a revolving pattern, and the corresponding spectral curves were collected to generate four spectral curves. Data fusion of these four spectral curves is then performed. The specific fusion process is shown in the figure below: First, after the spectral data is input, spectral curves at different wavelengths (such as 265nm, 275nm, 310nm, and a wide-spectrum LED of 340-1100nm) are collected. Data preprocessing, including noise removal and spectral normalization, is then performed to improve data quality. Feature extraction is then performed, selecting relevant wavelengths and calculating eigenvalues. Next, in the data fusion step, a weighted averaging method is used to combine the feature data, which involves determining weights, calculating weights, and normalizing them. Principal component analysis and eigenvector combination can also be used to further optimize the data. Finally, a classification model is constructed and parameter testing is performed to output the final results.

[0023] S3. Multi-index parallel analysis: Through the three-level process of "ultraviolet excitation-fluorescence capture-near infrared verification", the similarity of Maotai content, alcohol content, authenticity determination, brand logo and year value can be output simultaneously within 30 seconds.

[0024] Preferably, the detection of the amount of tartar content specifically includes: Establish a Moutai benchmark spectral library, including the 265nm / 425nm double peak area ratio, 310nm / 485nm shoulder peak offset, and 950nm absorption peak half width; Calculate similarity: \text{Similarity\%}=α·\frac{\|A_{\text{Sample}}-A_{\text{Benchmark}}\|}{A_{\text{Benchmark}}}+β·\text{SIM}(F_{\text{Sample}},F_{\text{Benchmark}}); Where α=0.4, β=0.6, SIM is the fluorescence spectrum correlation coefficient, and the output is 70%-100% quantitative results.

[0025] To determine Moutai content, a multi-spectral fusion similarity algorithm was developed by comparing the characteristic fluorescence peaks of Moutai Feitian under excitation at 265 / 275 / 310nm, such as the doublet at 425nm under 265nm excitation and the shoulder at 485nm under 310nm excitation, as well as the near-infrared hydroxyl absorption pattern. This algorithm compares the spectral data of the wine sample to a pre-established Moutai benchmark spectral database and quantifies the percentage of similarity between the sample and the Moutai benchmark (70%-100%), providing scientific and accurate data support for wine companies to benchmark Moutai prices.

[0026] Preferably, the alcohol content detection includes: Fit the OH bond third-order harmonic absorption curve in the near-infrared band of 900-1000nm and calculate the integral area S OH ; The background fluorescence intensity of ethanol at 275 nm is 275 Correction interference: \text{Alcohol content}=k_1·S_{OH}+k_2·I_{275}+b; (k1=0.78,k2=-0.15,b=3.2), achieving ±0.1vol% accuracy.

[0027] Alcohol content detection utilizes the third-order harmonic frequency characteristics of the OH bond in the near-infrared (850-1000nm) band, combined with ethanol background fluorescence correction technology at 275nm in the ultraviolet region. The absorption characteristics of the OH bond in the near-infrared spectrum have a specific correspondence with alcohol content. By analyzing the spectral data in this band and correcting for the background fluorescence of ethanol at 275nm in the ultraviolet region, interference from other factors is eliminated, achieving an accuracy of ±0.1vol% for alcohol content detection, meeting the high-precision requirements for alcohol content detection in the liquor trade.

[0028] Preferably, the brand identification includes: The pre-stored brand fingerprint database includes the 265nm / 365nm peak height ratio and 310nm / 415nm peak position difference of Wuliangye, and the 310nm / 450nm peak half width of Luzhou Laojiao; A one-dimensional convolutional neural network (CNN) model was constructed: the input layer received fused spectral data, the convolution kernel size was 5×1, the pooling layer used maximum pooling, and the fully connected layer output the brand classification probability.

[0029] The brand recognition system uses a three-wavelength ultraviolet fluorescence fingerprint, such as Wuliangye's characteristic peaks of 365nm / 415nm when excited at 265nm, and Luzhou Laojiao's characteristic peak of 450nm when excited at 310nm, to match the near-infrared absorption spectrum with a convolutional neural network. The system pre-establishes a spectral database containing tens of thousands of brands and uses the powerful pattern recognition capabilities of convolutional neural networks to rapidly analyze and match the input wine sample spectral data, achieving millisecond-level retrieval of wine sample brands, greatly improving the efficiency and accuracy of liquor brand recognition. Preferably, the year identification includes: Measure the fluorescence lifetime decay rate τ of humic acid under 275nm excitation (time resolution 0.1ns) and obtain the decay slope dτ / dt; Calculate the red shift of the fluorescence peak of ester excited at 310 nm (accuracy ±0.5 nm); Through regression model: \text{Year}=e^{a·(dτ / dt)+b·Δλ+c} (a=0.32, b=1.75, c=-2.1), interpreting the aging period of 3-30 years, with an error of ±1 year.

[0030] The vintage identification module utilizes a two-parameter regression model based on the fluorescence lifetime decay rate of humic acids under 275nm excitation (time-resolved detection) and the red shift of ester fluorescence under 310nm excitation. Humic acids and esters undergo specific changes with age. By monitoring these changes in fluorescence characteristics and analyzing them with the two-parameter regression model, it is possible to accurately determine the aging period of 3-30 years, providing a scientific and reliable method for vintage identification.

[0031] Preferably, the data security processing includes: Hash the spectral feature values ​​to generate a digital fingerprint; Through encrypted transmission on the Hyperledger Fabric blockchain platform, a spectral traceability certificate containing a timestamp and device ID is generated on the transaction chain.

[0032] 1. Check the hardware configuration of the device Optical modules: UV excitation unit: Three-wavelength LED array (265nm / 275nm / 310nm, power 10mW±5%, half-peak width <5nm), with marquee timing controlled by an STM32 microcontroller (lighting sequence: 265nm→275nm→310nm, single wavelength excitation time 100ms, interval 50ms); Fluorescence acquisition: Hamamatsu H11706-01 photomultiplier tube (response wavelength 200-900 nm) with OceanOptics USB4000 fiber optic spectrometer (detection range 340-1100 nm, resolution 0.5 nm); Near-infrared unit: InGaAs detector (Thorlabs DET20C2, response 850-2200 nm) with four-channel bandpass filter (center wavelength 850 / 900 / 950 / 1000 nm, bandwidth ±10 nm); Control Module: Main control chip: ARMCortex-A72 processor Temperature control component: The cuvette tank has a built-in Pt100 temperature sensor and semiconductor cooling chip to maintain a constant temperature of 25±0.5℃.

[0033] 2. Testing process steps Step 1: Sample pretreatment Take 2.0 ml of the liquor sample to be tested and inject it into a quartz cuvette (optical path 10 mm); Start the temperature control system and trigger the optical detection after the temperature stabilizes to 25℃.

[0034] Step 2: Multispectral Data Acquisition Stage ① UV fluorescence excitation: t = 0-100ms: Light up the 265nm LED and collect the fluorescence spectrum from 340-650nm (focus on recording the double peak intensities A1 and A2 at 425nm); t=150-250ms: Light up the 275nm LED and collect the 350-600nm spectrum (record the humic acid fluorescence decay curve, time resolution 0.1ns); t=300-400ms: Light up the 310nm LED and collect the spectrum from 400-700nm (capture the 485nm shoulder peak intensity B and the ester fluorescence peak position λ m ); Phase 2 Near-infrared verification: t=450-550ms: sequentially collect the absorption intensities of the four bands C1-C4 at 850nm, 900nm, 950nm, and 1000nm; Cycle mechanism: Repeat the above process 10 times and take the average value of the spectral data.

[0035] Step 3: Spectral data processing 31. Data Preprocessing ① Denoising The original spectral curve was smoothed using a Savitzky-Golay filter with a window width of 11 data points and a polynomial order of 3. This method can effectively eliminate high-frequency noise (such as circuit noise and ambient light interference) while preserving the spectral peak characteristics.

[0036] For the time domain signals of the fluorescence spectrum (e.g., the decay curve of humic acid excited at 275 nm), an exponentially weighted moving average algorithm (EWMA) was additionally applied to smooth the random fluctuations of the time-resolved fluorescence signal.

[0037] ② Normalization Each spectral curve is normalized to its maximum value: the intensity value at each wavelength point is divided by the global maximum value of the curve to scale all data to the [0,1] interval.

[0038] The four-band near-infrared data need to be baseline corrected separately: the background absorption value of each band is deducted based on the intensity of the 850nm band.

[0039] 32. Feature Extraction ① Characteristics related to the amount of tar Extract the double peak intensity (A1, A2) at 425 nm from the fluorescence spectrum excited at 265 nm, and calculate the peak area ratio: R1=A1 / (A1+A2); Locate the 485nm shoulder peak from the 310nm excited spectrum and calculate its half-maximum width (FWHM) and relative intensity (B / Bmao, Bmao is the Moutai reference value); The hydroxyl absorption intensity C3 was read from the 950nm near-infrared band, and its deviation from the reference value was calculated: ΔC3=|C3-C3mao| / C3mao.

[0040] ② Alcohol-related characteristics Fit the OH bond absorption curve in the near-infrared range of 900-1000nm and calculate the integral area S using the trapezoidal method OH ; Synchronously extract the 350nm fluorescence background intensity I under 275nm excitation 275 , used for ethanol concentration correction.

[0041] ③Brand fingerprint characteristics Record the 365nm / 415nm double peak intensity ratio (characteristic of Wuliangye) excited at 265nm; Extract the 450nm peak shift excited by 310nm (Luzhou Laojiao characteristic).

[0042] ④Year identification characteristics The fluorescence decay curve at 430 nm excited by 275 nm was fitted exponentially: I(t) = I0·exp(-t / τ) to obtain the fluorescence lifetime τ and the decay slope dτ / dt; Measure the red shift Δλ of the ester fluorescence peak under 310nm excitation (compared with the standard peak position of new wine).

[0043] 33.Data Fusion and Model Calculation ① Weighted fusion strategy In the calculation of the amount of cyperus, the weight of the fluorescence feature was set to 60% (35% for R1 and 25% for B), and the weight of the near-infrared feature was set to 40% (ΔC3); In the alcohol content model, the near-infrared integral area S OH Weight 78%, fluorescence correction term I 275 Weight is 15% and constant term is 7%.

[0044] ②Classification and regression models Brand recognition: The normalized UV fluorescence fingerprint (365nm / 415nm / 450nm feature combination) is input into the pre-trained 1D-CNN model, which outputs the brand probability distribution (Softmax normalization). Year identification: dτ / dt and Δλ are input into a two-parameter exponential regression model to output the aging years; Authenticity determination: Based on the comprehensive spectral similarity (Maotian content > 85%), brand confidence (> 95%) and blockchain evidence comparison results, the "genuine" or "counterfeit" conclusion is output.

[0045] 4. Data output and storage Generate a structured test report, including five-dimensional indicator values, confidence levels, and key spectral feature maps; Package the feature hash value (SHA-256) with device information and timestamp and upload it to the distributed ledger through the blockchain API; The local database stores the original spectral data (retained for 15 days) and the intermediate processing results (retained for 30 days) for re-inspection.

[0046] Step 4: Five-dimensional indicator analysis model (1) Detection of tar content Calculate similarity: Similarity% = 0.6 × S (fluorescence) + 0.4 × S (near infrared); Fluorescence similarity S(fluorescence) = 0.7 × |A1 / Amao-1|⁻¹ + 0.3 × |B / Bmao-1|⁻¹ (Amao = Moutai benchmark value); Near infrared similarity S(near infrared) = ; Output: 70%-100% continuous quantization value (resolution 0.1%).

[0047] (2) Alcohol content test Near-infrared correction model: Alcohol content (vol%) = 0.78 × ∫ 900 ¹ 000 S(λ)dλ-0.15×I275+3.2; I275: 350nm fluorescence background intensity under 275nm excitation (used to eliminate impurity interference) Precision verification: The test was repeated 10 times on 53%vol Moutai Feitian, with a standard deviation of σ = 0.05vol%.

[0048] (3) Brand identity CNN model architecture is as follows Figure 4 : Database coverage: 12,000 liquor brands, training set accuracy 99.2%.

[0049] (4) Year identification Two-parameter regression: Year = exp(0.32×dτ / dt+1.75×Δλ-2.1); dτ / dt: decay rate of fluorescence lifetime at 430 nm under 275 nm excitation (unit: ns / year); Δλ: red shift of ester peak at 310 nm (relative to the new wine offset value, unit: nm); Range: 3-30 years of aging, the actual measured value of 2010 Moutai is 12.3 years (nominal 12 years).

[0050] Step 5: Data security and output Blockchain evidence storage process: Generate data fingerprint: Hash = SHA256 (device ID + timestamp + feature value); Upload to the Hyperledger Fabric blockchain network (nodes include wineries / testing agencies / trading platforms); The trading platform calls the smart contract to verify the validity of the certificate; Terminal display: The touch screen outputs a five-dimensional indicator radar chart and a blockchain QR code (scan the code to view the traceability report).

[0051] Example: 2015 Moutai Feitian Liquor Testing Sample: 53%vol Kweichow Moutai Feitian (Batch No. 201507); Key Stats: Fluorescence characteristics: A1(425nm)=2850, A2(425nm)=2930, B(485nm)=1620; Near infrared: C3 (950nm) = 0.782; Year parameters: dτ / dt=0.142ns / year, Δλ=8.3nm; Analysis results: Similarity of Maotai content: 95.7% (first in benchmark library matching); Alcohol content: 53.0±0.1vol% Brand identity: Moutai (99.8% confidence level); Year: 2015.2 (error ±0.3 years); Authenticity: Authentic (spectral fingerprint is consistent with blockchain evidence record).

[0052] Blockchain credentials: {"tx_id":"0x8a3d...f1c", "timestamp":"2025-07-11T14:22:33Z", "device_id":"AIW-9A3B", "feature_hash":"7d8f1e...a2b4c", "result":{"mao_sim":95.7,"alc":53.0,"year":2015.2} }.

[0053] This invention utilizes a three-stage detection process: ultraviolet excitation, fluorescence capture, and near-infrared verification. During the testing process, the liquor sample is first excited by a three-wavelength ultraviolet excitation light source, generating a fluorescence signal that is captured and analyzed by a three-dimensional fluorescence spectrum analysis module. Simultaneously, a four-band near-infrared detection unit performs near-infrared spectroscopy on the liquor sample. Finally, the fluorescence and near-infrared spectral data are integrated and analyzed, enabling simultaneous analysis of five indicators: Maotai content, alcohol content, authenticity, brand, and vintage within 30 seconds. This data is encrypted and transmitted to the trading platform via blockchain technology. Leveraging the immutable nature of blockchain technology, reliable spectral traceability certificates are generated, effectively addressing core issues in liquor trading, such as a lack of pricing basis, difficulty in distinguishing authenticity, and false vintage labels.

[0054] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0055] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent wine detection method based on multi-spectral fusion, characterized in that: The following steps are involved: S1. Multispectral excitation and acquisition: The wine samples were excited in sequence using a three-wavelength UV LED light source of 265 nm, 275 nm, and 310 nm in a marquee mode, with an excitation interval of 0.5 seconds. The fluorescence emission spectrum in the range of 340-1100nm and the four-band near-infrared spectrum of 850nm, 900nm, 950nm and 1000nm are collected simultaneously to generate four original spectral curves; S2, spectral data fusion processing: denoising and normalization of the original spectral curve; Extract characteristic wavelengths: To detect the amount of tar, the double peak intensity at 425nm excited by 265nm, the shoulder peak intensity at 485nm excited by 310nm, and the hydroxyl absorption intensity at 950nm need to be extracted; to detect the alcohol content, the third-order harmonic absorption slope of the O-H bond at 900-1000nm needs to be extracted; The feature data is fused using the weighted average method, and the weights are dynamically allocated according to the feature signal-to-noise ratio; S3. Parallel analysis of multiple indicators: and simultaneous output of Maotai content similarity, alcohol content, authenticity determination, brand logo and year value.

2. The intelligent wine detection method based on multispectral fusion according to claim 1 is characterized by: The detection of the amount of tartar content specifically includes: Establish a Moutai benchmark spectral library, including the 265nm / 425nm double peak area ratio, 310nm / 485nm shoulder peak offset, and 950nm absorption peak half width; Calculate similarity; \text{Similarity\%}=α·\frac{\|A_{\text{Sample}}-A_{\text{Benchmark}}\|}{A_{\text{Benchmark}}}+β·\text{SIM}(F_{\text{Sample}},F_{\text{Benchmark}}); Where α=0.4, β=0.6, SIM is the fluorescence spectrum correlation coefficient, and the output is 70%-100% quantitative results.

3. The intelligent wine detection method based on multi-spectral fusion according to claim 1 is characterized by: The alcohol content test includes: Fit the OH bond third-order harmonic absorption curve in the near-infrared band of 900-1000nm and calculate the integral area S OH ; The background fluorescence intensity of ethanol at 275 nm is 275 Correction interference: \text{Alcohol content}=k_1·S_{OH}+k_2·I_{275}+b; (k1=0.78,k2=-0.15,b=3.2), achieving ±0.1vol% accuracy.

4. The intelligent wine detection method based on multi-spectral fusion according to claim 1 is characterized by: Said brand identity includes; The pre-stored brand fingerprint database includes the 265nm / 365nm peak height ratio and 310nm / 415nm peak position difference of Wuliangye, and the 310nm / 450nm peak half width of Luzhou Laojiao; A one-dimensional convolutional neural network (CNN) model was constructed: the input layer received fused spectral data, the convolution kernel size was 5×1, the pooling layer used maximum pooling, and the fully connected layer output the brand classification probability.

5. The intelligent wine detection method based on multi-spectral fusion according to claim 1 is characterized by: Said vintage identification includes; Measure the fluorescence lifetime decay rate τ of humic acid under 275nm excitation and obtain the decay slope dτ / dt; Calculate the red shift of the fluorescence peak of ester excited at 310 nm, Δλ; Through regression model; \text{Year}=e^{a·(dτ / dt)+b·Δλ+c}; a=0.32, b=1.75, c=-2.1, 3-30 years of aging, error ±1 year The intelligent wine detection method based on multispectral fusion according to claim 1 is characterized in that: The data security processing includes: Hash the spectral feature values ​​to generate a digital fingerprint; Through encrypted transmission on the Hyperledger Fabric blockchain platform, a spectral traceability certificate containing a timestamp and device ID is generated on the transaction chain.

Images