Method and device for quickly and nondestructively detecting quality of aquatic products

Abstract

The invention provides a rapid nondestructive quality detection method and device for aquatic products, and relates to the technical field of food safety detection, and the device comprises a spectral analysis module, a biosensing module, an electrochemical detection module, a data processing module, a data output module and a portable device. According to the rapid nondestructive quality detection method and device for the aquatic products, multi-index synchronous detection of freshness, drug residues and heavy metal content of the aquatic products is realized through fusion of spectral analysis, biosensing and electrochemical detection technologies.

Description

Technical Field

[0001] The present invention relates to the technical field of food safety detection, and in particular to a method and device for rapid and non-destructive quality detection of aquatic products. Background Art

[0002] Quality testing of aquatic products during distribution is crucial for ensuring food safety. Traditional testing methods rely on large laboratory instruments (such as high-performance liquid chromatography and atomic absorption spectrometers). These methods are bulky, require long testing cycles (typically hours to days), are complex to operate, and are expensive, making them difficult to meet the demands of rapid on-site screening.

[0003] Existing portable devices using a single detection technology (such as near-infrared spectrometers or electrochemical sensors) can only detect a subset of indicators and fail to fully cover core aquatic product quality parameters (freshness, drug residues, and heavy metals). Furthermore, multi-module data fusion and analysis capabilities are insufficient, test results are susceptible to environmental interference, and the lack of automated calibration mechanisms makes data drift a common problem with long-term use.

[0004] More critically, existing devices integrating multiple detection technologies generally suffer from simplistic data fusion algorithms and inadequate feature extraction. Traditional methods often rely on simple data concatenation or linear weighted fusion, failing to fully exploit the inherent correlations and complementary information between data from different detection technologies. Furthermore, the lack of dedicated deep learning models for multi-attribute testing of aquatic products results in detection accuracy and stability that are insufficient for practical applications. Therefore, there is an urgent need to develop intelligent fusion algorithms and deep learning models for multi-source heterogeneous detection data, as well as portable devices that integrate multiple technologies and support rapid nondestructive testing, to address the limitations of traditional methods. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for rapid and non-destructive quality detection of aquatic products. Through an innovative heterogeneous data preprocessing algorithm, an adaptive feature fusion mechanism and a dedicated deep residual network model, the technical difficulties of low data fusion efficiency, insufficient feature extraction and poor model generalization ability in traditional multi-technology integrated detection are solved, and rapid detection of multiple indicators of aquatic products is achieved, breaking through the bottleneck of large size, long time consumption and low precision of traditional equipment.

[0006] To achieve the above objectives, the present invention provides a device for rapid and non-destructive quality testing of aquatic products, comprising: The spectrum analysis module is used to scan the surface reflection spectrum of aquatic products, obtain the spectrum data in the 400-2500nm band, and transmit it to the data processing unit; The biosensor module includes a microfluidic chip, an antibody-modified electrode, and an impedance measurement unit, which is used to automatically inject the sample extract into the microfluidic control chip, detect the rate of change of the electrode interface impedance, generate a drug residue detection curve, and transmit the detected drug residue detection curve data to the data processing unit; The electrochemical detection module is used to perform pre-enrichment and dissolution scanning to record the integrated area of ​​heavy metal dissolution peaks and transmit the detection data to the data processing unit; The data processing module is used to process the data transmitted by the spectral analysis module, the biosensor module and the electrochemical detection module, and output the processed data through the data output module; The data output module includes a touch screen and a wireless transmission unit. The touch screen displays the test results and quality scores in real time. The wireless transmission unit uploads the test data to the blockchain evidence storage platform through an encryption protocol and generates a test report. The self-calibration system has a built-in standard material compartment containing freshness standard samples, drug residue standard solution and heavy metal standard solution. The robotic arm automatically performs daily zero-point calibration. The calibration data is used to correct the light intensity drift of the spectral analysis module, the baseline offset of the biosensor module and the background current of the electrochemical detection module.

[0007] Preferably, the microfluidic chip drives the sample liquid flow through a micropump to trigger the antibody-antigen specific binding reaction, and the impedance measurement unit monitors the impedance change of the electrode interface in real time, generates a drug residue detection signal and transmits it to the data processing unit; The microfluidic chip is composed of a PDMS base layer bonded to a glass substrate, and contains a serpentine mixing channel, a reaction chamber, and a waste liquid pool; The antibody-modified electrode uses a nano-gold / graphene composite substrate, and chloramphenicol monoclonal antibodies are immobilized on the surface by EDC / NHS coupling method; Impedance measurement unit integrated with electrochemical workstation.

[0008] Preferably, the electrochemical detection module adopts a three-electrode system, including a working electrode, a counter electrode and a reference electrode. The surface of the working electrode is modified with a heavy metal ion selective membrane, and the heavy metal ion dissolution peak current is detected by square wave anodic stripping voltammetry.

[0009] Preferably, the steps of processing data by the data processing module and the data output module include: Baseline calibration phase: PBS buffer was injected through the micropump, the initial impedance spectrum was recorded, and the baseline equivalent circuit parameters were established; Dynamic detection stage: The sample solution flows through the antibody-modified electrode surface. The antigen-antibody binding causes the double-layer capacitance to change. The system continuously collects impedance amplitude and phase data for multiple cycles. Signal processing stage: The Savitzky-Golay filtering algorithm is used to eliminate high-frequency noise, and the real part of the impedance change at the characteristic frequency point is extracted through the sliding window method to generate a three-dimensional impedance fingerprint. Specificity verification mechanism: A control electrode modified with a nonspecific antibody is set up to eliminate matrix interference by measuring the impedance difference between the main detection site and the control site; Multi-source data receiving stage: The data processing unit synchronously receives the raw data from the three detection modules through the first communication interface, the second communication interface, and the third communication interface, and aligns the timestamps to ensure data consistency; The second communication interface adopts RS485 protocol and transmits structured data packets containing timestamp, impedance amplitude spectrum and characteristic frequency point change rate at 115200bps; The microfluidic chip is equipped with an automatic cleaning process. After the test is completed, 0.1M NaOH solution, deionized water and nitrogen are injected in sequence for three flushing cycles, and the residual rate is less than 0.5%.

[0010] Preferably, the working electrode of the three-electrode system is a glassy carbon electrode with a diameter of 3 mm, the surface of which is modified with a nano-platinum / graphene composite layer and a heavy metal ion selective membrane in sequence; The selective membrane is composed of chitosan-thioglycolic acid copolymer and Pb 2+ / Cd 2+ The ion-imprinted polymer was compounded in a mass ratio of 1:3.

[0011] Preferably, the data processing module includes a data fusion submodule and a machine learning analysis submodule. The data fusion submodule normalizes the spectral data, impedance change data, and dissolution peak current data and generates feature vectors. The machine learning analysis submodule classifies and regresses the feature vectors using a pre-trained deep neural network model and outputs a comprehensive assessment report including freshness grade, drug residue type, and heavy metal excess index. A method for rapid and non-destructive quality testing of aquatic products, comprising the following steps: S1. Place the aquatic product to be tested in the testing chamber and perform full-band spectrum scanning through the spectrum analysis module to obtain surface reflectance distribution data; S2, the biosensor module automatically injects the sample extract into the microfluidic chip, monitors the impedance change rate of the antibody-modified electrode, and generates a drug residue detection curve; S3, the electrochemical detection module performs three square wave voltammetry scans in the potential range of -1.2V to +0.5V, and records the integrated area of ​​the heavy metal dissolution peak; S4, the data processing module extracts and fuses the features of the spectral reflectance data, impedance change rate data, and dissolution peak area data, and inputs them into the deep neural network model for multi-index joint analysis; S5. Generate a test report including a freshness score, a warning of excessive drug residues, and a heavy metal content threshold based on the output results of the data processing module, and display and store it through the data output module.

[0012] Therefore, the present invention adopts the above-mentioned method and device for rapid and non-destructive quality detection of aquatic products, and the technical effects are as follows: (1) Intelligent fusion of multi-source heterogeneous data: For the first time, an adaptive fusion algorithm for spectral, biosensor, and electrochemical heterogeneous data is proposed. Through feature space mapping and dynamic weight adjustment, deep fusion of data from different physical detection principles is achieved; (2) Dedicated deep learning model: The innovative design of the branch-fusion-residual three-level network architecture improves detection accuracy by 15-20% and reduces false alarm rate by more than 60% compared to traditional machine learning methods; (3) Adaptive model optimization: Through online learning and transfer learning mechanisms, the model can automatically adjust parameters according to different aquatic product types and seasonal changes, and has strong generalization capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a schematic diagram of the modular structure of the portable device of the present invention; Figure 2 It is a flow chart of the detection method; Figure 3 This is the design diagram of the serpentine channel of the biosensor module microfluidic chip; Figure 4 This is the layout diagram of the three-electrode system of the electrochemical detection module; Figure 5 This is the deep learning model architecture diagram of the data processing unit. DETAILED DESCRIPTION

[0014] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0015] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0016] Example 1 like Figure 1 As shown, a device for rapid non-destructive quality detection of aquatic products comprises The spectrum analysis module is used to scan the surface reflection spectrum of aquatic products, obtain the spectrum data in the 400-2500nm band, and transmit it to the data processing unit; The spectral analysis module, located in the upper portion of the portable device, comprises a high-precision LED light source array, a multispectral sensor, and a reflective probe. The LED light source array consists of 32 sets of tunable wavelength LED units, covering the 400-2500nm spectral range, arranged in a circular pattern around the reflective probe at an incident angle of 45°±2°. The multispectral sensor utilizes a 2048-pixel InGaAs linear array detector equipped with a grating spectrometer system, achieving a spectral resolution of 1.5nm. The reflective probe utilizes a sapphire lens and a seven-core fiber bundle, connecting to the spectrometer via an SMA905 interface and transmitting spectral data to the data processing module via a USB 3.0 port.

[0017] like Figure 3 As shown, the biosensor module includes a microfluidic chip, an antibody-modified electrode, and an impedance measurement unit, which is used to automatically inject the sample extract into the microfluidic control chip, detect the rate of change of the electrode interface impedance, generate a drug residue detection curve, and transmit the detected drug residue detection curve data to the data processing unit; The biosensor module, located in the center of the device, consists of a microfluidic chip, antibody-modified electrodes, and an impedance measurement unit. The microfluidic chip, constructed from a PDMS substrate bonded to a glass substrate, includes an inlet, a serpentine mixing channel, a reaction chamber, an antibody-modified electrode array, a wastewater reservoir, and a micropump system. The channel is 200 μm wide and 50 μm deep. The antibody-modified electrodes utilize a gold nanoparticle / graphene composite substrate, with chloramphenicol monoclonal antibodies immobilized on the surface via EDC / NHS coupling. The impedance measurement unit integrates an electrochemical workstation, transmitting impedance change data to a data processing unit via RS485 protocol.

[0018] The basic structure of the microfluidic chip consists of a 2mm-thick PDMS substrate with microchannels fabricated using soft lithography, and a 1mm-thick glass substrate that is firmly bonded to the PDMS substrate after plasma activation. The two together form a closed microfluidic channel system, ensuring directional flow of liquid within the chip and preventing cross-contamination.

[0019] The inlet, located in the upper left corner of the chip, has a 1mm diameter and a funnel-shaped design, facilitating precise injection of sample extracts. The inlet's hydrophobic interior prevents liquid residue and ensures accurate quantitative injection. A microvalve connected to the inlet controls the precise injection timing and flow rate of the sample solution.

[0020] The serpentine mixing channel extends from the sample inlet in an S-shaped, zigzag structure. It measures 200μm wide and 50μm deep, for a total length of 35mm. Micro-mixing units are located every 1mm within the channel, forming staggered diamond-shaped micro-pillars that enhance fluid agitation and promote thorough mixing of the sample and buffer. The channel walls are hydrophilically modified to reduce surface tension and ensure smooth liquid flow.

[0021] The reaction chamber (303) is located at the end of the serpentine channel, has a diameter of 5 mm, a depth of 100 μm, and a volume of approximately 2 μL. The bottom of the chamber is a glass substrate, and the top is a transparent PDMS layer for easy optical observation. The inner wall of the chamber is blocked with BSA (bovine serum albumin) to reduce nonspecific adsorption and improve detection sensitivity.

[0022] The antibody modified electrode array (304) is fixed at the bottom of the reaction chamber and consists of three groups of parallel arranged gold electrodes. Each group of electrodes includes a working electrode (diameter 500 μm), a reference electrode and an auxiliary electrode. The working electrode is a nano-gold / graphene composite substrate, and the surface is immobilized with chloramphenicol monoclonal antibody by EDC / NHS coupling method, with an antibody loading density of 1.2×10 4 molecules / μm 2 Among them, the first group of the three groups of electrodes is the main detection site, the second group is the control electrode modified with non-specific antibodies, and the third group is the spare electrode.

[0023] The microfluidic chip workflow is as follows: Sample preparation stage: 50 μL sample liquid is extracted from the aquatic product, and injected into the injection port after simple centrifugation; Sample delivery stage: the micropump system drives the sample liquid at a constant flow rate of 0.1 μL / s from the injection port (301) into the serpentine mixing channel; Mixing reaction stage: The sample liquid is fully mixed with the pre-set PBS buffer in the serpentine channel, and the micro-mixing unit in the channel enhances the disturbance to ensure uniform distribution of the reactants; Antigen-antibody binding stage: The mixed liquid enters the reaction chamber and comes into contact with the antibody-modified electrode array. The drug residue (antigen) in the sample specifically binds to the monoclonal antibody on the electrode surface, causing the electrode interface impedance to change. Signal detection stage: The impedance measurement unit continuously collects 20 cycles of impedance amplitude and phase data at a sampling frequency of 10 Hz, while monitoring the difference in impedance changes between the main detection site and the control site to eliminate matrix interference; Data processing stage: The Savitzky-Golay filtering algorithm is used to eliminate high-frequency noise, and the real part of the impedance change at characteristic frequency points (1kHz, 10kHz, 100kHz) is extracted through the sliding window method to generate a three-dimensional impedance fingerprint. Waste liquid discharge stage: After the detection is completed, the reaction liquid flows into the waste liquid pool (305) through the drainage channel, completing one detection cycle.

[0024] After the test is completed, the microfluidic chip automatically performs a cleaning process: first, 0.1M NaOH solution is injected to flush the channels and reaction chambers. The strong alkaline environment can effectively separate the antigen-antibody complex; then deionized water is injected for neutralization and dilution to remove residual alkaline solution; finally, nitrogen gas is introduced to blow dry the channels and reaction chambers to prevent the growth of microorganisms and maintain the long-term stability of the chip.

[0025] The microfluidic chip design offers the following advantages: First, the serpentine channel structure prolongs the residence time of liquids within the chip, ensuring thorough mixing of samples and reagents. Second, the three sets of parallel electrodes enable simultaneous detection and control, effectively eliminating the effects of environmental interference and nonspecific adsorption. Finally, the integrated microfluidic structure enables integrated processing of sample injection, mixing reaction, signal detection, and waste liquid collection, significantly improving detection efficiency and accuracy. This microfluidic chip is particularly suitable for rapid screening of trace drug residues in aquatic products, with detection sensitivity and specificity meeting national standards.

[0026] like Figure 4 As shown, the electrochemical detection module is used to perform pre-enrichment and dissolution scanning to record the integrated area of ​​heavy metal dissolution peaks and transmit the detection data to the data processing unit; The electrochemical detection module is located at the bottom of the equipment and adopts a three-electrode system consisting of a working electrode, a counter electrode, a reference electrode, an electrode holder, a detection chamber and an electrochemical control unit to form a complete heavy metal detection system.

[0027] The working electrode is located at the center of the three-electrode system. It uses a 3mm diameter glassy carbon electrode, which is modified with a 50nm thick nano-platinum / graphene composite layer and a 100μm thick heavy metal ion selective membrane. The nano-platinum / graphene composite layer is prepared by electrochemical deposition to enhance the electron transfer efficiency of the electrode; the heavy metal ion selective membrane is composed of chitosan-thioglycolic acid copolymer and Pb 2+ / Cd 2+ The ion-imprinted polymer is compounded in a mass ratio of 1:3 to achieve selective recognition and enrichment of specific heavy metal ions. The working electrode surface area is 7.07mm 2 , effectively increasing the deposition area of ​​heavy metal ions.

[0028] The counter electrode is a spiral platinum wire electrode with a diameter of 0.5 mm and a length of 10 mm. It is wound in a spiral to increase surface area and reduce polarization effects. The counter electrode is located on the side of the working electrode at a 120° angle and a fixed distance of 2 mm ± 0.1 mm. The spiral design ensures sufficient surface area to support current flow to the working electrode, preventing current limitation during the reaction.

[0029] The reference electrode is an Ag / AgCl microelectrode containing 3M KCl electrolyte and has a tip diameter of 0.8 mm. It is located on the opposite side of the working electrode, also at a 120° angle to the working electrode and with a distance of 2 mm ± 0.1 mm, forming a three-electrode ring configuration. The reference electrode provides a stable reference potential, ensuring accurate and repeatable potential measurements.

[0030] The electrode holder is made of corrosion-resistant and chemically inert polytetrafluoroethylene. Precise electrode insertion holes ensure the three electrodes are precisely positioned relative to each other, preventing signal fluctuations caused by positional shifts during testing. A sealing ring at the bottom of the holder provides a tight connection to the test chamber to prevent liquid leakage.

[0031] The detection chamber is cylindrical with an inner diameter of 12 mm, a height of 10 mm, and a volume of approximately 1.1 mL. It is made of corrosion-resistant PEEK. The inner wall of the chamber is specially treated to prevent the adsorption of heavy metal ions. The bottom of the chamber is equipped with a sample injection port and a liquid discharge port. Liquid flow is controlled by a micro-solenoid valve, enabling automatic sample injection and discharge.

[0032] The electrochemical control unit, consisting of a potentiostat (±0.1mV accuracy) and a microcurrent detection unit (range 0.1nA-10mA), is responsible for controlling the electrode potential and measuring the current response. The control unit connects to the three-electrode system via a flexible cable, enabling data acquisition and potential control.

[0033] The workflow of the electrochemical detection module is as follows: Pre-test preparation stage: Before testing, the system automatically takes out the heavy metal standard solution from the standard substance compartment for calibration and establishes a standard curve between the dissolution peak current and the heavy metal concentration; Sample preparation stage: extract the liquid to be tested from the aquatic product sample, inject it into the detection cavity after simple filtration, and the liquid completely covers the three-electrode system; Pre-enrichment stage: The electrochemical control unit applies a -1.2V deposition potential to the working electrode for 300s to make the Pb 2+ 、Cd 2+ The target heavy metal ions are reduced and deposited on the electrode surface. The specific binding sites on the selective membrane at this stage help enrich the target ions and improve the detection sensitivity; Equilibrium static stage: The system pauses for 30 seconds to allow the electrode interface to reach a stable state and reduce background current interference; Dissolution scan: The system uses a square wave frequency of 25 Hz, a potential increment of 4 mV, and an amplitude of 50 mV to perform a dissolution scan within a potential range of -1.2 V to +0.5 V. As the potential gradually increases, heavy metals accumulated on the working electrode surface are oxidized and dissolved in sequence, generating characteristic dissolution peaks. Data acquisition phase: The system performs three repeated scans continuously, records the average value of the dissolution peak current, and monitors the potential position of each characteristic peak (Pb 2+ :-0.56V±0.02V, Cd 2+ :-0.76V±0.02V); Signal processing: A moving average filter algorithm is used to eliminate double-layer charging current, and wavelet denoising technology (db4 wavelet basis, decomposition level 5) is combined to remove environmental electromagnetic interference. A Gaussian fitting algorithm is used to separate overlapping dissolution peaks and accurately calculate the peak current integrated area of ​​each heavy metal ion. Concentration calculation stage: The system converts the peak current integrated area into heavy metal concentration values ​​based on the pre-established standard curve, and automatically compares it with the GB 2762-2017 national food safety standard threshold to generate an excessive warning message; Verification and correction stage: The system automatically compares the difference in dissolution peak current between the main working electrode and the reference electrode without modified ion-imprinted polymer, corrects the matrix effect, and calculates the actual heavy metal content.

[0034] After the test is completed, the system automatically performs the electrode cleaning process: First, inject 0.1MHNO3 solution and perform electrochemical cleaning at a potential of -0.2V for 30s to oxidize and dissolve the metal remaining on the electrode surface; Deionized water is then injected to rinse the electrode surface and neutralize the acidic environment; Finally, anhydrous ethanol was injected for final cleaning, and dry nitrogen was introduced to dry the electrode surface to reduce the metal residue to less than 0.1 ng / cm 2 , to ensure the accuracy of the next test.

[0035] The electrochemical detection module has the following technical advantages: First, the three-electrode annular distribution design optimizes the electric field distribution and reduces the mutual interference between electrodes; second, the chitosan-thioglycolic acid / ion-imprinted polymer composite membrane improves the selective recognition ability of heavy metal ions and reduces matrix interference; finally, the design of multiple repeated scanning combined with reference electrode comparison significantly improves the accuracy and reliability of detection. 2+ The linear detection range is 0.5-200μg / kg, and the detection limit is 0.2μg / kg (S / N=3); for Cd 2+ The linear detection range is 0.3-150μg / kg, and the detection limit is 0.1μg / kg (S / N=3), which fully meets the requirements of the national food safety standard GB 2762-2017 and provides reliable technical support for on-site rapid screening of heavy metal content in aquatic products.

[0036] like Figure 5As shown, the data processing module is used to process the data transmitted by the spectrum analysis module, the biosensor module and the electrochemical detection module, and output the processed data through the data output module; The data processing module, located in the central control area of ​​the device, includes a data fusion submodule and a machine learning analysis submodule. The data fusion submodule, consisting of a multi-channel data acquisition card, a preprocessing unit, and a feature fusion processor, is responsible for preprocessing and extracting features from the data transmitted by the three detection modules. The machine learning analysis submodule, equipped with an ARM Cortex-A72 quad-core processor and an NPU acceleration unit, runs a pretrained deep residual network model and implements multi-metric joint analysis through four residual blocks.

[0037] The core innovation of the data processing module lies in the intelligent fusion algorithm for multi-source heterogeneous data: first, for the 2048 wavelength points of the spectral data, an improved Savitzky-Golay filter (with adaptive adjustment of the window width from 5 to 25 points) is used to eliminate high-frequency noise, combined with SNV normalization to eliminate the scattering effect, and then the first 10 principal components are extracted through PCA (cumulative variance contribution rate >98%); second, for the impedance data of the biosensor, the FFT transform is used to extract the real and imaginary changes in the impedance at the three characteristic frequencies of 1kHz, 10kHz, and 100kHz, and the impedance change trend characteristics are extracted through the sliding window method (window size 20 points); third, the db4 wavelet basis 5-layer decomposition is used to denoise the electrochemical data, combined with the Gaussian fitting algorithm to separate overlapping dissolution peaks, and four types of features are extracted: peak position, peak height, peak area, and peak symmetry.

[0038] The deep learning model architecture of the data processing module consists of an input layer, a data preprocessing layer, a feature fusion layer, a deep residual network layer, a multi-task output layer, and a model optimization unit, forming a complete intelligent analysis system.

[0039] The input layer contains three parallel data interfaces, receiving reflectance data (2048 spectral points in the 400-2500nm range) from the spectral analysis module, impedance change data (including timestamp, impedance amplitude spectrum, and characteristic frequency point change rate) from the biosensor module, and dissolution peak current data (including heavy metal type, peak potential value, and peak current integrated area) from the electrochemical detection module. The input layer is equipped with a data buffer capable of temporarily storing 100 sets of historical detection data for model adaptive optimization.

[0040] The data preprocessing layer performs specific preprocessing operations for different types of input data: Spectral data is smoothed and denoised using the Savitzky-Golay filter algorithm (window width 15 points, polynomial order 3), and scattering effects are eliminated using the standard normal transformation (SNV). Impedance data are transformed into the frequency domain using the FFT algorithm, extracting the real and imaginary impedance changes at three characteristic frequencies: 1 kHz, 10 kHz, and 100 kHz. Electrochemical data are de-noised using wavelet denoising (db4 wavelet basis, decomposition level 5) to eliminate background noise, and peak positions and areas are accurately extracted using the Gaussian fitting algorithm. The preprocessing layer outputs three standardized feature data.

[0041] A three-level architecture of branch-fusion-residual is adopted: the first level is a dedicated feature extraction branch, with dedicated feature extraction networks designed for spectral, biosensor and electrochemical data respectively. The spectral branch uses a one-dimensional convolutional neural network to capture local correlations between wavelengths, the biosensor branch uses an LSTM network to process temporal impedance changes, and the electrochemical branch uses a multi-layer perceptron to process discrete peak features; the second level is an adaptive fusion layer, which dynamically calculates the weight coefficients of the three types of data through the attention mechanism, avoiding the limitations of traditional fixed weights; the third level is a deep residual analysis network, which extracts the complete feature spectrum from low-level features to high-level semantic features layer by layer through the cascade of four residual blocks.

[0042] The feature fusion layer consists of a principal component analysis (PCA) unit, a feature selector, and a fusion processor. The PCA unit performs PCA dimensionality reduction on the preprocessed spectral data, extracting the top 10 principal components, which explain 98% of the variance. The feature selector uses the Fisher discriminant ratio to screen key features for the impedance data and electrochemical data, selecting 12 and 6 features, respectively. The fusion processor performs a weighted fusion of the three features using a 0.4:0.3:0.3 ratio, combining sample type information and ambient temperature compensation parameters to generate a 128-dimensional feature vector.

[0043] The deep residual network layer adopts a four-layer residual block structure. Each residual block contains two 3×3 convolutional layers, a batch normalization layer, and a ReLU activation function, and introduces a skip connection structure to avoid the gradient vanishing problem. The first residual block mainly processes spectral features and outputs 64 feature maps; the second residual block focuses on impedance data features and outputs 128 feature maps; the third residual block focuses on electrochemical features and outputs 256 feature maps; the fourth residual block integrates the outputs of the first three blocks and generates a 512-dimensional feature representation through a global average pooling layer. The network uses a dropout (0.5) mechanism to prevent overfitting and an attention mechanism to adaptively adjust the weights of different features.

[0044] The multi-task output layer contains three parallel task branches: the freshness classification branch uses the Softmax activation function to classify the freshness of aquatic products into four levels: excellent (90-100 points), good (75-89 points), qualified (60-74 points) and unqualified (<60 points); the drug residue prediction branch uses the Sigmoid activation function to calculate the probability of containing various drug residues such as antibiotics and preservatives in the sample, and marks the risk level of exceeding the standard; the heavy metal content regression branch uses the linear output layer to predict Pb 2+ 、Cd 2+ The specific content of heavy metals such as iodine, the unit is μg / kg.

[0045] The model optimization unit includes an online learning module and an adaptive calibration module. The online learning module supports incremental learning. By storing historical test data and comparing it with laboratory verification results, it periodically updates model parameters to improve the model's generalization ability across different aquatic product types. The adaptive calibration module adjusts the model's prediction deviation in real time based on the test results of standard samples from the self-calibration system, ensuring long-term accuracy and reliability.

[0046] The data processing flow is as follows: Multi-source data receiving stage: The data processing unit synchronously receives the raw data from the three detection modules through the first communication interface, the second communication interface, and the third communication interface, and aligns the timestamps to ensure data consistency; Data preprocessing: The preprocessing layer performs noise reduction, baseline correction, and signal enhancement on various data types to improve the signal-to-noise ratio. Spectral data uses differential transformation to enhance subtle variations, impedance data uses phase analysis to eliminate capacitance interference, and electrochemical data uses baseline drift correction to improve peak accuracy. Feature extraction and fusion stage: The system extracts key feature parameters from preprocessed data, such as reflectivity at characteristic spectral wavelengths, slope of impedance curves, and symmetry of heavy metal dissolution peaks. After dimensionality reduction through principal component analysis, feature weighted fusion is performed in combination with prior knowledge to generate a unified feature vector. Deep learning analysis phase: The fused feature vector is sequentially passed through four residual blocks for deep feature extraction and transformation. Each residual block focuses on capturing feature correlations at different levels. A skip connection structure preserves the original information while extracting deep abstract features, ultimately outputting a high-dimensional feature representation. Multi-task prediction stage: Based on deep feature representation, the system performs three prediction tasks in parallel. Freshness scores are derived by comparing historical databases; drug residue detection calculates the probability of exceeding the standard based on impedance change characteristics; and heavy metal content prediction is quantitatively analyzed by combining dissolution peak characteristics and standard curves. Results Integration and Output Phase: The system integrates the three types of prediction results into a unified assessment report, comparing them with national standard thresholds to generate a risk level assessment. The final test results are displayed on the touchscreen display of the data output module and uploaded to the blockchain evidence storage platform.

[0047] Performance optimization mechanism: Periodic calibration phase: The system performs zero-point calibration every day and adjusts the model prediction deviation through standard sample testing; Parameter update phase: Verification data is aggregated monthly, and model weights are updated through transfer learning methods to adapt to the changing characteristics of aquatic products in different seasons and batches; Abnormal monitoring stage: The system continuously monitors the model prediction variance. When abnormal fluctuations occur continuously, the early warning mechanism is triggered and a full calibration is recommended.

[0048] The data processing module has the following technical advantages: First, the multi-technique fusion feature extraction method fully exploits the complementary information of spectral, biosensor and electrochemical data, improving the comprehensiveness of detection; Secondly, the deep residual network architecture effectively solves the problem of difficult deep network training and can capture complex nonlinear relationships; Finally, a multi-task learning framework enabled unified prediction of freshness, drug residues, and heavy metal content, streamlining the decision-making process. Trained and validated on 2,000 sets of real-world aquatic product samples, this data processing unit achieved 97.8% accuracy in freshness classification, 96.5% in drug residue detection, and a relative error of less than 5% in heavy metal content prediction, fully meeting the accuracy requirements for rapid on-site testing.

[0049] The data output module includes a touch screen and a wireless transmission unit. The touch screen displays the test results and quality scores in real time. The wireless transmission unit uploads the test data to the blockchain evidence storage platform through an encryption protocol and generates a test report. The data output module includes a touchscreen display and a wireless transmission unit. The touchscreen is connected to the data processing unit via an HDMI 2.0 port, displaying test results and quality scores in real time. The wireless transmission unit is connected to the data processing unit via a USB Type-C port, responsible for uploading test data to the blockchain evidence storage platform and generating a PDF test report.

[0050] The self-calibration system has a built-in standard material compartment containing freshness standard samples, drug residue standard solution and heavy metal standard solution. The robotic arm automatically performs daily zero-point calibration. The calibration data is used to correct the light intensity drift of the spectral analysis module, the baseline offset of the biosensor module and the background current of the electrochemical detection module.

[0051] like Figure 2As shown, a method for rapid non-destructive quality detection of aquatic products comprises the following steps: In the first stage (sample pretreatment), the steps are as follows: S1. Place the aquatic product to be tested in the test chamber. The equipment automatically identifies the sample type (fish, shellfish or crustacean) and adjusts the test parameters. S2. The equipment automatically performs pre-checks, including light source stability verification, electrode status detection, and system self-checks, to ensure the equipment is in optimal working condition. S3. The system calls the corresponding detection plan and sets differentiated detection thresholds and reference standards for different types of aquatic products.

[0052] In the second phase (multi-module parallel testing), each functional module is started synchronously and the following testing steps are performed: S4, the spectral analysis module, performs full-band spectral scanning (400-2500nm) to collect surface reflectance distribution data for aquatic products. Specifically, an LED light source array illuminates the sample surface at a 45° angle of incidence. A reflective probe captures the reflected light, and a grating spectrometer decomposes the optical signal into different wavelengths. The multispectral sensor then captures the complete spectral curve. S5: The biosensor module automatically injects 50 μL of sample extract into the microfluidic chip. After thorough mixing in the serpentine mixing channel, the liquid enters the reaction chamber and contacts the antibody-modified electrode. A piezoelectric micropump drives the sample flow at a flow rate of 0.1 μL / s. The impedance measurement unit monitors the change rate of the electrode interface impedance at a scan rate of 50 mV / s in differential pulse voltammetry mode to generate a drug residue detection curve. S6. In the pre-enrichment stage, the electrochemical detection module applies a constant potential of -1.2V deposition potential to the working electrode for 300s to reduce and deposit the target heavy metal ions on the electrode surface; then in the dissolution scan stage, three square wave voltammetric scans are performed in the potential range of -1.2V to +0.5V using parameters of square wave frequency of 25Hz, potential increment of 4mV, and amplitude of 50mV, and the integrated area of ​​the heavy metal dissolution peak is recorded.

[0053] In the third stage (data fusion analysis), the data processing unit integrates the three detection data and performs the following processing steps: S7, the data fusion submodule applies the Savitzky-Golay filtering algorithm to smooth and de-noise the spectral reflectance data, uses the FFT algorithm to analyze the frequency domain characteristics of the impedance change rate data, and uses the wavelet denoising technology (db4 wavelet basis, decomposition level 5) to remove environmental electromagnetic interference from the stripping peak current data; S8. The pre-processed data is reduced in dimension using a principal component analysis algorithm to extract 28-dimensional principal component features, which are then combined with ambient temperature compensation parameters and sample type features to generate a 128-dimensional feature vector; S9, the machine learning analysis submodule inputs the feature vector into the pre-trained deep residual network model, and performs a multi-index joint analysis through four residual blocks to obtain the freshness grade score, drug residue type and probability, and heavy metal content prediction value.

[0054] In the fourth stage (result output), the system generates a comprehensive evaluation report and displays the test results: S10. The report generation engine automatically generates a test report based on the model output results, including a freshness score (0-100 points), a warning of excessive drug residues (normal / slightly exceeded / severely exceeded), and a comparison of heavy metal content thresholds; S11. The system automatically compares the test results with the thresholds of GB 2762-2017 and GB 2733-2015 national food safety standards and gives a pass / fail judgment; S12. The data output module displays the test results and quality scores in real time through the touch screen. At the same time, the wireless transmission unit uploads the test data to the blockchain evidence storage platform through an encryption protocol and generates a PDF format test report containing a risk warning QR code.

[0055] After the inspection is completed, the device will automatically perform cleaning and maintenance procedures: S13, the biosensor module is flushed three times with 0.1 M NaOH solution, deionized water, and nitrogen gas in sequence to ensure that the residual rate is less than 0.5%; S14, the electrochemical detection module is injected with 0.1M HNO3 solution, deionized water and anhydrous ethanol in sequence to regenerate the electrode surface so that the metal residue is less than 0.1ng / cm 2 ; S15. The device automatically records the test data, updates the calibration parameters, and prepares for the next test.

[0056] Here's how it works: First, the aquatic product to be tested is placed in the testing chamber. The spectral analysis module activates the LED light source array, collects the surface reflection spectrum of the aquatic product through the reflective probe, and the multispectral sensor obtains spectral data in the 400-2500nm band and transmits it to the data processing unit through the USB3.0 interface. At the same time, the biosensor module automatically injects 50 μL of sample extract into the microfluidic chip. The antibody-modified electrode specifically binds to the drug residue in the sample. The impedance measurement unit monitors the impedance change at the electrode interface and transmits the data to the data processing unit via the RS485 protocol. The electrochemical detection module performs square wave voltammetry scans within the potential range of -1.2V to +0.5V. The three-electrode system detects the peak current of heavy metal dissolution and transmits the data to the data processing unit via the Modbus RTU protocol. The data fusion submodule of the data processing unit preprocesses and extracts features from the three-channel data to generate a 128-dimensional feature vector. The machine learning analysis submodule performs multi-index joint analysis using a deep residual network model. Finally, the analysis results are displayed on the data output module's touchscreen display and uploaded to the blockchain evidence storage platform via a wireless transmission unit, generating a PDF test report. The entire testing process is completed within 30 seconds, enabling rapid and non-destructive testing of aquatic product freshness, drug residues, and heavy metal content.

[0057] The solution of the present invention has been verified in actual scenarios (sampling and testing in coastal aquatic product markets), with a single test error rate of less than 3% and a consistency of 98.5% with laboratory test results. It can be widely used in aquatic product wholesale, logistics and market supervision.

[0058] Therefore, the present invention employs the aforementioned method and apparatus for rapid, nondestructive quality testing of aquatic products. First, the parallel use of multiple technologies significantly shortens the testing cycle, with a single complete test taking no more than 30 seconds. Second, multi-dimensional data fusion improves the accuracy and comprehensiveness of testing, enabling a one-stop assessment of aquatic product's freshness, drug residues, and heavy metal content. Finally, automated sample processing and data analysis reduce operational complexity, enabling even non-professionals to easily complete professional-level quality testing. This method is particularly suitable for on-site rapid screening and quality grading of aquatic product distribution, providing an efficient and reliable technical means for ensuring aquatic food safety.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A device for rapid non-destructive quality detection of aquatic products, characterized in that: include: The spectrum analysis module is used to scan the surface reflection spectrum of aquatic products, obtain the spectrum data in the 400-2500nm band, and transmit it to the data processing unit; The biosensor module includes a microfluidic chip, an antibody-modified electrode, and an impedance measurement unit, which is used to automatically inject the sample extract into the microfluidic control chip, detect the rate of change of the electrode interface impedance, generate a drug residue detection curve, and transmit the detected drug residue detection curve data to the data processing unit; The electrochemical detection module is used to perform pre-enrichment and dissolution scanning to record the integrated area of ​​heavy metal dissolution peaks and transmit the detection data to the data processing unit; The data processing module is used to process the data transmitted by the spectral analysis module, the biosensor module and the electrochemical detection module, and output the processed data through the data output module; The data output module includes a touch screen and a wireless transmission unit. The touch screen displays the test results and quality scores in real time. The wireless transmission unit uploads the test data to the blockchain evidence storage platform through an encryption protocol and generates a test report. The self-calibration system has a built-in standard material compartment containing freshness standard samples, drug residue standard solution and heavy metal standard solution. The robotic arm automatically performs daily zero-point calibration. The calibration data is used to correct the light intensity drift of the spectral analysis module, the baseline offset of the biosensor module and the background current of the electrochemical detection module.

2. The device for rapid non-destructive quality detection of aquatic products according to claim 1, characterized in that: The microfluidic chip drives the sample liquid flow through a micropump, triggering the antibody-antigen specific binding reaction. The impedance measurement unit monitors the impedance changes of the electrode interface in real time, generates a drug residue detection signal, and transmits it to the data processing unit. The microfluidic chip is composed of a PDMS base layer bonded to a glass substrate, and contains a serpentine mixing channel, a reaction chamber, and a waste liquid pool; The antibody-modified electrode uses a nano-gold / graphene composite substrate, and chloramphenicol monoclonal antibodies are immobilized on the surface by EDC / NHS coupling method; Impedance measurement unit integrated with electrochemical workstation.

3. The device for rapid non-destructive quality detection of aquatic products according to claim 1, characterized in that: The electrochemical detection module adopts a three-electrode system, including a working electrode, a counter electrode and a reference electrode. The surface of the working electrode is modified with a heavy metal ion selective membrane, and the heavy metal ion dissolution peak current is detected by square wave anodic stripping voltammetry.

4. The device for rapid non-destructive quality detection of aquatic products according to claim 1, characterized in that: The data processing module and the data output module process the data in the following steps: Baseline calibration phase: PBS buffer was injected through the micropump, the initial impedance spectrum was recorded, and the baseline equivalent circuit parameters were established; Dynamic detection stage: The sample solution flows through the antibody-modified electrode surface. The antigen-antibody binding causes the double-layer capacitance to change. The system continuously collects impedance amplitude and phase data for multiple cycles. Signal processing stage: The Savitzky-Golay filtering algorithm is used to eliminate high-frequency noise, and the real part of the impedance change at the characteristic frequency point is extracted through the sliding window method to generate a three-dimensional impedance fingerprint. Specificity verification mechanism: A control electrode modified with a nonspecific antibody is set up to eliminate matrix interference by measuring the impedance difference between the main detection site and the control site; Multi-source data receiving stage: The data processing unit synchronously receives the raw data from the three detection modules through the first communication interface, the second communication interface, and the third communication interface, and aligns the timestamps to ensure data consistency; The second communication interface adopts RS485 protocol and transmits structured data packets containing timestamp, impedance amplitude spectrum and characteristic frequency point change rate at 115200bps; The microfluidic chip is equipped with an automatic cleaning process. After the test is completed, 0.1M NaOH solution, deionized water and nitrogen are injected in sequence for three flushing cycles, and the residual rate is less than 0.5%.

5. The device for rapid non-destructive quality detection of aquatic products according to claim 1, characterized in that: The working electrode of the three-electrode system is a 3mm diameter glassy carbon electrode, the surface of which is modified with a nano-platinum / graphene composite layer and a heavy metal ion selective membrane. The selective membrane is composed of chitosan-thioglycolic acid copolymer and Pb 2+ / Cd 2+ The ion-imprinted polymer was compounded in a mass ratio of 1:

3.

6. The device for rapid non-destructive quality testing of aquatic products according to claim 1, characterized in that: The data processing module includes a data fusion submodule and a machine learning analysis submodule. The data fusion submodule normalizes the spectral data, impedance change data, and dissolution peak current data and generates feature vectors. The machine learning analysis submodule classifies and regresses the feature vectors through a pre-trained deep neural network model, and outputs a comprehensive assessment report including freshness level, drug residue type, and heavy metal exceedance index.

7. A method for rapid and non-destructive quality testing of aquatic products, characterized in that: Using the device according to any one of claims 1 to 6, comprising the following steps: S1. Place the aquatic product to be tested in the testing chamber and perform full-band spectrum scanning through the spectrum analysis module to obtain surface reflectance distribution data; S2, the biosensor module automatically injects the sample extract into the microfluidic chip, monitors the impedance change rate of the antibody-modified electrode, and generates a drug residue detection curve; S3, the electrochemical detection module performs three square wave voltammetry scans in the potential range of -1.2V to +0.5V, and records the integrated area of ​​the heavy metal dissolution peak; S4, the data processing module extracts and fuses the features of the spectral reflectance data, impedance change rate data, and dissolution peak area data, and inputs them into the deep neural network model for multi-index joint analysis; S5. Generate a test report including a freshness score, a warning of excessive drug residues, and a heavy metal content threshold based on the output results of the data processing module, and display and store it through the data output module.

Images