Intelligent detection method, system and medium for meat freshness based on fluorescence sensor array

Abstract

The present invention discloses a method, system and medium for intelligent detection of meat freshness based on a fluorescence sensor array. The method comprises: preparing a fluorescence sensor array indicator label and reacting it with meat, photographing and labeling it to obtain a labeled image data set, constructing and training a meat freshness intelligent detection model; and finally realizing the freshness detection of the meat to be tested. Based on the idea of ​​transfer learning, the present invention extracts features from the fluorescence array information image and constructs a model. The feature extraction is automatic and the training time is short. It can realize fast and accurate judgment of the degree of meat spoilage and prediction of TVB-N content. Compared with the colorimetric sensor array, the prepared fluorescence sensor array has higher sensitivity, is more sensitive to slight changes in the concentration of biogenic amine gases, has a faster response speed and a wider detection range. In addition, each indicator in the array can simultaneously indicate the freshness status of the meat. The fluorescence signal channels are richer, and the freshness of the meat can be judged from multiple dimensions.

Description

Technical Field

[0001] The present invention belongs to the technical field of non-destructive food testing, and in particular relates to a method, system and medium for intelligent meat freshness detection based on a fluorescence sensor array. Background Art

[0002] During the spoilage process of meat food, proteins decompose under the combined action of microorganisms and enzymes to produce amine compounds including ammonia, trimethylamine, dimethylamine, histamine, putrescine, cadaverine, etc. Most of these compounds are highly volatile, so they are collectively called volatile nitrogen compounds (TVB-N). The level of TVB-N can be used to indicate the freshness of meat.

[0003] In many TVB-N detection studies, optical sensor array analysis technology, by simulating the animal's olfactory function, expands the traditional detection method of obtaining single information from a single channel to the simultaneous acquisition of multi-channel and multi-dimensional information, which can significantly improve the recognition throughput and provide new ideas for the development of multi-target analysis technology in the process of meat spoilage.

[0004] According to the difference in optical signals, optical sensor arrays can be mainly divided into two categories: colorimetric sensor arrays and fluorescence sensor arrays. Among them, colorimetric sensor arrays have been widely used in the detection of meat freshness. Chinese patent CN202010746884.0 discloses "A meat freshness indicator array, its preparation method and application". This method uses 9 acid-base indicators or porphyrin compounds such as phenol red, bromothymol blue, metacresol purple, methyl red, etc. to prepare a colorimetric sensor array, and use it to detect the freshness of meat. Similarly, Chinese patent CN201910971265.9 discloses "Paper-based colorimetric sensor array label and its preparation method and application". This method configures 9 different proportions of phenol red-bromocresol green sensitive indicators to prepare a colorimetric sensor array label and use it to detect the freshness of fish meat. Although the above method can achieve a good effect, the indicators used are all toxic to a certain extent. There will be potential safety hazards in the actual use process. It is not an ideal material for preparing food smart labels. Therefore, the current research will consider more about using natural pigments instead of chemical synthetic indicators to prepare colorimetric sensor arrays for the detection of meat freshness. For example, in the Chinese patent CN202111568540.6 “A preparation method and application of a fish meat freshness indication colorimetric array”, the indicators used in the colorimetric array are anthocyanins, curcumin and beet red, and the polymers used to make the colorimetric array are all food grade, but natural pigments are more susceptible to external factors than chemical synthetic dyes, and have poor stability, low tinting power and high cost. In addition, whether chemical synthetic indicators or natural synthetic indicators are used, in the process of detecting the freshness of meat, in addition to responding to volatile amines, they also respond to CO2, hydrogen sulfide, aldehydes, alcohols, etc. in the packaging environment. It is a non-specific recognition, which reduces the sensitivity of the colorimetric sensor array and slows down the response.

[0005] Compared with colorimetric sensor arrays, fluorescence sensors have a wider range of applications and are more advantageous for the following reasons: (1) Fluorescence sensor arrays have higher sensitivity. As a photoluminescent signal, fluorescence signals are more sensitive to small changes in the concentration of the analyte than absorption signals. (2) Fluorescence signal channels are richer and can provide multiple channels such as excitation, emission spectrum, and fluorescence polarization, while colorimetric sensor arrays often have single-channel output. (3) The selection of fluorescence sensor materials is richer than that of colorimetric sensors. Not only can a variety of classic fluorophores be modified and regulated, but there are also a large number of nanofluorescent materials to choose from. Therefore, the use of fluorescence sensor arrays to detect meat freshness has great application prospects.

[0006] Currently, there are also reports on the use of fluorescence sensor arrays in meat freshness detection. For example, Tian Yanqing et al. modified curcumin and its analogs to construct a multicolor fluorescence sensor array for monitoring shrimp freshness (Sensor and Actuators B: Chemical, 2022, 367, 132153). Similarly, Bahram Hemmateenejad et al. used different blocking agents to prepare four cycloplatinized (II) complexes and eight metal clusters. The resulting fluorescence sensor paper arrays can distinguish seven amine gases with concentrations ranging from 5 to 100 ppm (Sensor and Actuators B: Chemical, 2021, 334, 129582). Although these studies have achieved good accuracy in detecting spoilage markers, the pattern recognition algorithms used in the arrays are still limited to traditional multivariate statistical analysis methods, namely clustering algorithms (hierarchical cluster analysis, principal component analysis, linear discriminant analysis, etc.) and classification algorithms (support vector machines, random forests, K-nearest neighbors, etc.). These algorithms are only effective when processing small sample data sets. When dealing with multidimensional, complex, and large datasets, traditional clustering and classification algorithms fall short, especially when dealing with multidimensional fluorescent sensor arrays with rich signal channels. Furthermore, these algorithms require manual extraction and screening of array feature information, which undoubtedly introduces human error and inconvenience. Therefore, it is necessary to explore more convenient, fast, and powerful pattern recognition algorithms to meet the requirements of fast, intelligent, and accurate fluorescent sensor arrays for meat freshness detection. Summary of the Invention

[0007] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and to provide a method, system and medium for intelligent detection of meat freshness based on a fluorescence sensor array.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] One aspect of the present invention provides a method for intelligently detecting meat freshness based on a fluorescence sensor array, characterized in that it comprises the following steps:

[0010] preparing a fluorescent sensor array indicator label;

[0011] The fluorescent sensor array indicator tag reacts with the meat, photographs it, and labels it to obtain a fluorescent sensor array label image dataset with freshness labels, and builds and trains an intelligent meat freshness detection model;

[0012] The freshness of the meat to be tested is detected using the fluorescent sensor array indicator tag and the trained meat freshness intelligent detection model.

[0013] As a preferred technical solution, the preparation of the fluorescent sensor array indicator label is specifically as follows:

[0014] Dissolving a fluorescent substance that can respond to biogenic amines in a solvent to prepare a fluorescent indicator;

[0015] Cut a hydrophobic film into multiple sensor response areas, and stick a paper-based material between two hydrophobic films to form a fluorescent array sensor tag substrate with the function of isolating the sensor units from each other;

[0016] The fluorescent indicator is dropped into the response area of ​​the fluorescent array sensor tag substrate by an artificial spotting method and then dried to obtain the fluorescent sensor array indicator tag.

[0017] As a preferred technical solution, the fluorescent substances that can respond to biogenic amine substances include pH-sensitive fluorescent substances, biogenic amine-responsive fluorescent substances, hydrogen sulfide-responsive fluorescent substances, aldehyde-responsive fluorescent substances, alcohol-responsive fluorescent substances, ether-responsive fluorescent substances, alkane-responsive fluorescent substances and CO2-responsive fluorescent substances; the hydrophobic film includes an organic film substrate, a non-woven fabric / flannel substrate, a foam substrate, a metal substrate and a composite substrate; the shape of the response area includes a circle, a square, a triangle, a diamond and a combination thereof.

[0018] As a preferred technical solution, the feasibility of detecting volatile spoilage marker gases of meat using the prepared fluorescent sensor array indicator tag is also included, specifically:

[0019] The volume required for volatile spoilage marker gases of varying concentrations was calculated using the gas diffusion formula. A fluorescent sensor array label was affixed to a sealed container and reacted with the volatile spoilage marker gases at varying concentrations. The color change of the label was captured under ultraviolet light. The color change, ΔE, of the fluorescent sensor array was calculated using the Euclidean distance formula. The magnitude of ΔE was used to distinguish between different types and concentrations of volatile spoilage markers, thereby verifying the feasibility of the prepared fluorescent sensor array for monitoring meat spoilage.

[0020] The volatile corruption marker gases include biogenic amines, hydrogen sulfide, aldehydes, alcohols, ethers, alkanes and CO2;

[0021] The gas diffusion formula is specifically:

[0022]

[0023] Among them, V μL is the volume of the reaction solution, D mg / is the density of the reaction liquid, W is the mass fraction of the reaction liquid, M g / molis the mole fraction of the reaction solution, V L is the volume of the container;

[0024] The color change adopts the Euclidean distance ΔE change of the color:

[0025]

[0026] Among them, ΔR, ΔG, and ΔB are the differences in the R, G, and B color channels before and after the reaction, respectively.

[0027] As a preferred technical solution, the fluorescent sensor array indicator label reacts with meat, photographs it, and labels it to obtain a fluorescent sensor array label image dataset with a freshness label:

[0028] A fluorescent sensor array indicator label was affixed to the top of the inside of a fresh-keeping box and stored with fresh meat in a constant temperature and humidity chamber. Digital images of the sensor array were acquired in real time under ultraviolet conditions. The TVB-N standard value of the fresh meat was measured according to the physical and chemical indicators specified in the national standard GB5009.228-2016. The image was then annotated with a freshness grade label and a TVB-N content label to obtain a labeled fluorescent sensor array image dataset.

[0029] As a preferred technical solution, the meat freshness intelligent detection model includes a meat freshness evaluation model and a meat TVB-N content prediction model, specifically:

[0030] The labeled fluorescence sensor array image dataset was preprocessed to remove noise, normalize, and set the batch size. Classifier labels were then set based on the classification category. Then, based on the idea of ​​transfer learning, the images were input into a deep convolutional neural network pre-trained on the ImageNet database. After optimizing the model parameters, a meat freshness assessment model was obtained.

[0031] Repeat the above steps to establish a meat TVB-N content prediction model based on a deep convolutional neural network. Unlike the meat freshness assessment model, this meat TVB-N content prediction model is trained using TVB-N standard values ​​annotated in fluorescence sensor array images rather than category information.

[0032] The fluorescence sensor array images to be tested are input into the meat freshness assessment model and the meat TVB-N content prediction model respectively, and the prediction results of the fresh meat freshness category and TVB-N content are output respectively.

[0033] Another aspect of the present invention provides a meat freshness intelligent detection system based on a fluorescence sensor array, which is applied to the above-mentioned meat freshness intelligent detection method based on a fluorescence sensor array, and includes a fluorescence sensor array indicator label preparation module, a model construction and training module, and a meat freshness detection module;

[0034] The label preparation module is used to prepare fluorescent sensor array indicator labels;

[0035] The model building and training module is used to react the fluorescent sensor array indicator label with meat, photograph and label it, obtain a fluorescent sensor array label image dataset with freshness labels, and build and train a meat freshness intelligent detection model;

[0036] The meat freshness detection module is used to detect the freshness of the meat to be tested using the fluorescent sensor array indicator label and the trained meat freshness intelligent detection model.

[0037] Another aspect of the present invention provides a storage medium storing a program, which, when executed by a processor, implements the above-mentioned intelligent meat freshness detection method based on the fluorescence sensor array.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] (1) Compared with the colorimetric sensor array, the fluorescent sensor array prepared by the present invention has higher sensitivity, is more sensitive to small changes in the concentration of biogenic amine gases, has a faster response speed, and a wider detection range. In addition, each indicator in the array can simultaneously indicate the freshness status of meat, and the fluorescent signal channels are richer, which can judge the freshness of meat from multiple dimensions.

[0040] (2) Based on the concept of transfer learning, this invention utilizes multiple pre-trained deep convolutional neural network algorithms to extract features from fluorescence array information images and construct a qualitative meat freshness recognition model, eliminating the need for manual feature extraction and reducing training time. The resulting models are able to quickly and accurately determine the degree of meat spoilage and predict TVB-N content, providing a new approach to the pattern recognition process of fluorescence sensor arrays. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the process of an intelligent meat freshness detection method based on a fluorescence sensor array according to an embodiment of the present invention;

[0042] Figure 2 The ultraviolet-visible absorption spectrum, fluorescence emission spectrum, and excitation spectrum of the copper nanoclusters prepared in step 1 of Example 2 of the present invention;

[0043] Figure 3This is a graph showing the response of the fluorescent sensor array tag prepared in step 2 of Example 2 of the present invention to biogenic amine gases (ammonia, trimethylamine, and dimethylamine) at different gas concentrations according to the method in step 3 of Example 2;

[0044] Figure 4 This is a graph showing changes in Euclidean distance after the fluorescence sensor array prepared in step 2 of Example 2 of the present invention responds to corruption marker gases (ammonia, trimethylamine, and dimethylamine) of different gas concentrations according to the method in step 3 of Example 2;

[0045] Figure 5 This is a graph showing the fluorescence color change of the fluorescent sensor array label prepared in step 2 of Example 2 of the present invention during the beef storage process;

[0046] Figure 6 This is a confusion matrix diagram of the prediction set based on the optimal classification model (ResNet-50) described in step 4 of Example 2 of the present invention;

[0047] Figure 7 Graphs showing the predicted value and the true value based on the optimal regression model (VGG-19) described in step 4 of Example 2 of the present invention;

[0048] Figure 8 This is a graph showing the Euclidean distance change of the fluorescence color of a single fluorescent sensor unit tag after reacting with ammonia of different concentrations in Comparative Example 1 of Example 3 of the present invention;

[0049] Figure 9 This is a graph showing the Euclidean distance change of fluorescence color after the fluorescent sensor array tag reacts with ammonia gas at different concentrations in Comparative Example 2 of Example 3 of the present invention;

[0050] Figure 10 This is the Euclidean distance change of the fluorescence color and the TVB-N change of the beef during the storage process of the fluorescent sensor array freshness indicator label prepared in Comparative Example 3 of Example 3 of the present invention.

[0051] Figure 11 1 is a schematic structural diagram of an intelligent meat freshness detection system based on a fluorescence sensor array according to an embodiment of the present invention;

[0052] Figure 12 It is a schematic structural diagram of a storage medium according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0054] In recent years, advances in computer technology, particularly artificial intelligence, have made it possible to conveniently, accurately, and efficiently detect the freshness of meat. Deep learning models have been widely used in various fields. Convolutional neural networks, due to their powerful image feature extraction capabilities, are widely used in image recognition and classification. For fluorescent sensor arrays, each sensor in the array produces a corresponding change in fluorescence signal during the meat spoilage process. This means that the collected fluorescence image information will vary significantly at different stages of spoilage. Capturing these image changes can reveal the freshness of the sample, making the selection of an appropriate pattern recognition algorithm crucial. Deep learning, with its powerful image feature extraction and analysis capabilities, offers significant advantages in pattern recognition for fluorescent sensor arrays. Furthermore, with the advent of transfer learning, when the size of a self-built dataset is small, it is possible to leverage the initial weights of a pretrained model to effectively extract common image features in a shallow network, enabling more efficient image recognition and classification. Therefore, the present invention combines deep learning technology based on transfer learning with a fluorescent array composed of a composite of fluorescent substances that respond to biogenic amines to achieve specific identification of the freshness level of meat products during the spoilage process and prediction of TVB-N content. This technology is of great significance for ensuring food safety.

[0055] Example 1

[0056] like Figure 1 As shown, this embodiment provides a method for intelligently detecting meat freshness based on a fluorescence sensor array, comprising the following steps:

[0057] S1. Prepare fluorescent sensor array indicator label;

[0058] S1.1. Prepare multiple fluorescent indicators by synthesizing or purchasing multiple fluorescent substances that respond to biogenic amines, dissolving them in a suitable solvent, and combining one or two of them together;

[0059] Furthermore, the fluorescent substances include: pH-sensitive fluorescent substances, biogenic amine-responsive fluorescent substances, hydrogen sulfide-responsive fluorescent substances, aldehyde-responsive fluorescent substances, alcohol-responsive fluorescent substances, etc.

[0060] S1.2. Cut the hydrophobic film into multiple sensor response areas of appropriate size, and then paste the paper-based material between two hydrophobic materials to form a fluorescent array sensor label substrate with the ability to isolate the sensor units from each other; use the manual spotting method to add the prepared fluorescent indicator to the response area of ​​the sensor array substrate, and place it in an oven to dry, and then you can get a fluorescent sensor array indicator label.

[0061] Furthermore, the hydrophobic film includes polytetrafluoroethylene, polyester, polyvinyl chloride, polypropylene, carbon fiber film, etc.; the shape of the response area includes circle, square, triangle, diamond and their combination; the area of ​​the array response area is 3 to 5 cm 2 ; The array response area is 8 to 25; the types of paper-based materials include filter paper, cotton paper, organic microporous filter membrane, polyamide film, etc.; the amount of fluorescent indicator added is 80 to 100 μL; the drying temperature is 30 to 50°C; and the drying time is 20 to 40 minutes.

[0062] S2. Feasibility verification of detecting volatile spoilage marker gases in meat using fluorescent sensor array indicator tags;

[0063] The volume required for volatile spoilage marker gases of different concentrations and types was calculated according to the gas diffusion formula (Formula 1). Then, the fluorescent sensor array indicator tags were placed in sealed containers to react with volatile spoilage marker gases of different concentrations and types. The color change of the tags was photographed by a smartphone under ultraviolet light excitation. The color change of the fluorescent sensor array was measured with the Euclidean distance (Formula 2) indicator to verify the feasibility of the prepared fluorescent sensor array indicator tag for detecting volatile spoilage marker gases in meat. Specifically, Figure 4 As shown in the figure, when the concentration of the corrupted gas exceeds 20 ppm, the color change (ΔE) values ​​of the fluorescence array after reacting with the three gases already differ significantly. Furthermore, as the gas concentration increases, the trends of the three ΔE curves diverge, allowing us to distinguish different concentrations and types of corrupted gases based on the ΔE values. Finally, since the ΔE value represents the degree of color change of the array, the larger the value, the more pronounced the color change. When the ΔE value exceeds 15, the color change of the array is observable to the naked eye.

[0064] Furthermore, the volatile corruption marker gas is biogenic amines, hydrogen sulfide, aldehydes, alcohols, etc.; the concentration of the volatile corruption marker gas is 5-500 ppm; the volume of the closed container is 3-5 L; and the reaction time is 15-30 min.

[0065] Furthermore, the calculation formula (1) of the volatile corruption markers at different gas concentrations is:

[0066]

[0067] Where: V μL ---Reaction liquid volume, D mg / ---Reaction liquid density, W---reaction liquid mass fraction, M g / mol ---Reaction liquid mole fraction, V L ---Container volume.

[0068] Furthermore, the color change is a change in the Euclidean distance ΔE of the color:

[0069]

[0070] Among them, ΔR, ΔG, and ΔB are the differences in the R, G, and B color channels before and after the reaction, respectively.

[0071] S3. Reacting the fluorescent sensor array indicator tag with the meat, photographing and labeling it to obtain a fluorescent sensor array label image dataset with freshness labels, and constructing and training a meat freshness intelligent detection model, wherein the meat freshness intelligent detection model includes a meat freshness assessment model and a meat TVB-N content prediction model;

[0072] S3.1. The prepared fluorescent sensor array freshness indicator label is pasted on the top of the inside of a transparent fresh-keeping box and stored in a constant temperature and humidity chamber with fresh meat. As the storage period of the meat increases, the TVB-N generated also increases with the increase in storage days, and the color of each sensor unit on the fluorescent array sensor label also changes accordingly. The digital image of the sensor array can be obtained in real time under ultraviolet conditions through a smartphone. After the TVB-N standard value of fresh meat is measured according to the physical and chemical indicators specified in the national standard GB5009.228-2016, the image is annotated with freshness category information and TVB-N content information, thereby obtaining a fluorescent sensor array label image dataset with freshness labels.

[0073] Furthermore, the temperature of the constant temperature and humidity chamber is set to 4°C, 10°C, and 28°C.

[0074] S3.2. Preprocess the collected images to remove noise and normalize them before inputting them into a pretrained deep convolutional neural network. The batch size and classifier labels are set in advance. After setting the number of model iterations, initial learning rate, and model evaluation metrics, and optimizing the model parameters, a meat freshness assessment model is obtained.

[0075] Furthermore, the deep convolutional neural network is AlexNet, VGGNet-16, VGGNet-19, ResNet-30, ResNet-50, GoogLeNet, etc.; the number of model iterations is 30 to 100 times; the initial learning rate is 0.0001 to 0.001; the learning rate optimization method is exponential decay dynamic adjustment or specified interval; the model parameter optimization is SGD optimizer or Adam gradient descent algorithm; the model evaluation indicators are the loss and accuracy of the training set, test machine and validation set, and the confusion matrix.

[0076] S3.3. Use the deep convolutional neural network model trained in step S3.3 to input the image of the fluorescent sensor array to be tested. After the trained convolutional neural network model extracts features, the decision function can output the model's prediction result on the freshness level of fresh meat.

[0077] S3.4. Repeat steps S3.1 to S3.3 to establish a meat TVB-N content prediction model based on a deep convolutional neural network. Unlike the freshness classification model, the fluorescence array image is marked with the TVB-N standard value rather than the category information.

[0078] S3.5. Use the convolutional neural network model trained in step S3.4, input the fluorescence array image with the labeled TVB-N standard value, and after the model extracts the features, output the model's prediction result for the TVB-N content in fresh meat through the decision function.

[0079] Furthermore, each image is annotated with two pieces of information: the TVB-N content and the freshness rating, which is divided into three categories based on the TVB-N content. When training the convolutional neural network models in S3.3 and S3.4, the input is the same photo, the only difference being the classifier in the final layer. The training process is similar, with different parameters and evaluation metrics. The resulting intelligent meat freshness detection model consists of two models: a TVB-N content prediction model that outputs TVB-N information, and a meat freshness assessment model that outputs category information.

[0080] S4. Use the fluorescent sensor array indicator label and the trained meat freshness assessment model to detect the freshness of the meat to be tested.

[0081] Example 2

[0082] Metal nanoclusters have the advantages of small size, good photostability, large Stokes shift, mild preparation conditions, and are non-toxic and harmless. They are a very promising material for preparing fluorescent sensors for detecting the freshness of meat. Therefore, this embodiment uses metal nanoclusters, fluorescent dyes, and a combination of the two to prepare fluorescent indicators that have different degrees of response to amine substances, and uses them to prepare fluorescent sensor arrays. Figure 1 Schematic diagram of the preparation, detection, and application process of the fluorescence sensing array.

[0083] Step 1: Preparation of fluorescent sensing indicator

[0084] Copper nanoclusters (CuNCs) were prepared by using D-penicillamine (DPA) as a template and reducing agent to reduce Cu under acidic conditions. 2+ The Au-S bond is formed. 100 μL (0.4 mol / L) of Cu(NO₃)₂·3H₂O solution was added to 40 mL (0.01 mol / L) of DPA solution. Under stirring at 35°C, 1 mol / L HCl solution was added dropwise to raise the pH of the precursor solution to 4.5. After stirring for 1 hour, the reaction solution was centrifuged at 8000 rpm for 10 minutes to remove the unreacted DPA solution. Fluorescent copper nanoclusters were obtained after freeze-drying under vacuum. Finally, 0.2 g of copper nanoclusters was dissolved in 25 mL of aqueous solution, and 0.2 g of fluorescein (fluorescein isothiocyanate, 7-hydroxycoumarin, rhodamine 6G, fluorescein, 4,4'-bis(2-sulfostyryl)dione disodium, and acriflavine) was dissolved in 100 mL of ethanol solution. The mixture was mixed and stirred evenly according to the volume ratios listed in Table 1 to obtain nine fluorescent sensing indicators.

[0085] serial number Indicator composition Indicator ratio 1 fluorescein isothiocyanate 100％ 2 7-Hydroxycoumarin 100％ 3 DPA-CuNCs 100％ 4 DPA-CuNCs@fluorescein isothiocyanate 75:1 5 DPA-CuNCs@7-hydroxycoumarin 150:1 6 DPA-CuNCs@Rhodamine 6g 375:1 7 DPA-CuNCs@4,4'-bis(2-styrylsulfonate) disodium hydroxide 600:1 8 DPA-CuNCs@fluorescein 100:1 9 DPA-CuNCs@Acriflavine 100:1

[0086] Table 1. Fluorescence sensor indicator ratio table

[0087] The results were characterized by UV-visible spectroscopy, fluorescence spectroscopy and optical properties. Figure 2 As shown, the synthesized copper nanocluster solid appears off-white under visible light and emits a strong red light under UV light. The UV-visible spectrum shows that the UV absorption curve of DPA-CuNCs lacks any surface plasmon resonance absorption peaks associated with metal particles, indicating that the synthesized copper nanoclusters do not produce byproducts such as large metal nanoparticles. The fluorescence spectrum shows that DPA-CuNCs exhibit a strong fluorescence emission peak at 667 nm. In summary, CuNCs possess excellent fluorescence properties.

[0088] Step 2: Preparation of fluorescent sensor array indicator labels

[0089] A hole puncher was used to punch out 9 (3×3) circular sensor response areas with a diameter of 8 mm on two 4 cm×4 cm black square carbon fiber hydrophobic materials. Then a square cotton paper substrate of the same size of 4 cm×4 cm was pasted between the two carbon fiber hydrophobic materials to form a fluorescent array sensor label substrate with the function of isolating the sensing units from each other. 100 μL of fluorescent indicator was added to the response area of ​​the sensor substrate using the manual spotting method, and then placed in a 40°C oven to dry for 30 minutes to obtain a fluorescent sensor array freshness indicator label.

[0090] Step 3: Verify the feasibility of the method of using the prepared fluorescent sensing array indicator tag to detect volatile spoilage marker gases in meat.

[0091] Taking the detection of biogenic ammonia gas as an example, the required volumes of 5, 20, 50, 100, 300, and 500 ppm of biogenic amines (ammonia, trimethylamine, and dimethylamine) were calculated according to the gas diffusion formula (Formula 1). The fluorescent sensor array tag was placed in a 5L sealed container and reacted with different concentrations of biogenic amines for 15 minutes. The Euclidean distance ΔE change of the tag color was captured and photographed by a smartphone under ultraviolet conditions (Formula 2). This was used to explore whether the fluorescent array smart tag can detect biogenic amines of different concentrations and types and whether it can be applied to beef freshness detection.

[0092] Depend on Figure 3 It can be seen that the fluorescent sensor array tag can respond to different types and concentrations of ammonia, trimethylamine, and dimethylamine. As the gas concentration increases, the reaction becomes more complete, that is, the color change becomes more obvious. Figure 4 It can be seen that the fluorescent sensor array tag can distinguish three different biogenic amines, and the ΔE is greater than 15, that is, the color change is obvious and can be distinguished by the naked eye.

[0093] Step 4: The fluorescent sensor array indicator tag is combined with deep learning to intelligently detect the freshness of fresh meat, including the following steps:

[0094] (1) Fresh beef purchased from the market was dried, and 35g of beef was taken as an experimental sample and placed in a 500mL transparent fresh-keeping box; the fluorescent sensor array was pasted on the top of the inner part of the packaging box; the fresh-keeping box containing beef and fluorescent sensor array was placed at a constant temperature of 28℃ for 36 hours, and the fluorescent sensor array label was photographed by a smartphone under ultraviolet excitation every 2 hours. The aspect ratio of the photographed photos was kept at 2:1, and then 3 boxes of beef were randomly selected to measure the TVB-N standard value of the beef according to the physical and chemical indicators specified in the national standard GB5009.228-2016. According to the average value of the TVB-N standard value of the 3 boxes of beef, the beef TVB-N standard value and freshness grade information were marked on the fluorescent sensor array photos (TVB-N < 15mg / 100g was considered a fresh sample, 15mg / 100g ≤ TVB-N < 20mg / 100g was considered a sub-fresh sample, TVB-N ≥ 20mg / 100g was considered a rotten sample, and the sample was considered rotten). Figure 5 It can be seen that according to this standard, the array label can effectively distinguish different freshness levels of beef. After labeling the labels, a fluorescent array sensor label image dataset with freshness labels is obtained. The image dataset is divided into training set, validation set, and test set in a ratio of 3:1:1.

[0095] (2) The collected image information (about 3,000 images) was preprocessed to remove the background blue fluorescence emitted by the packaging box under ultraviolet conditions, and the images were converted into a 448×224×3 matrix format and normalized.

[0096] Input the processed data sets into the following deep convolutional networks for training, setting parameters and selecting the method for optimizing parameters:

[0097] The network parameters were initialized using the AlexNet weights pre-trained on ImageNet. The model iteration count was 30, and the initial learning rate was set to 0.01. The learning rate was dynamically adjusted using an exponential decay strategy, and the optimizer was SGD. The resulting beef freshness assessment model based on a fluorescence sensor array was denoted as AlexNet.

[0098] The network parameters were initialized using the weights of VGG-16 and VGG-19 pre-trained on ImageNet. The model iteration count was 30, and the initial learning rate was 0.001. The initial learning rate for the feature extraction module was also set at 0.0001. A dynamic learning rate adjustment strategy with a specified interval was used to accelerate model training. The Adam gradient descent algorithm was used to optimize model parameters during training. The resulting beef freshness assessment models based on the fluorescence sensor array were denoted as VGG-16 and VGG-19.

[0099] The model parameters were initialized using RestNet-30 and ResNet-50 weights pre-trained on ImageNet. The model iteration count was 20, the initial learning rate for the weights was set to 0.01, and SGD was selected as the optimizer. The resulting beef freshness assessment models based on the fluorescence sensor array were denoted as RestNet-30 and ResNet-50.

[0100] The above models all use the accuracy and loss of the training set, test set, and validation set to measure the performance of the model, and the confusion matrix to measure the accuracy of the classification prediction.

[0101] (3) Using the AlexNet, VGG-16, VGG-19, ResNet-30, and ResNet-50 convolutional neural network models trained in step (2), the fluorescence sensor array image to be tested is input. After the trained convolutional neural network model extracts features, the decision function is used to output the model's prediction results on the freshness of beef and its category.

[0102] Depend on Figure 6 The confusion matrix shows that the best model (ResNet-50) has an accuracy of 0.985 in the test set, and can accurately identify beef of different freshness levels.

[0103] (4) Repeat steps (1) to (3) to establish a fresh meat TVB-N content prediction model based on AlexNet, VGG-16, VGG-19, and ResNet-50. Unlike the classification model, the fluorescence array images are labeled with TVB-N standard values ​​rather than category information.

[0104] The root mean square error (RMSE) of the training set, test set and validation set and the coefficient of determination (R 2 ) to measure the performance of the TVB-N prediction model.

[0105] (5) Using the convolutional neural network model trained in step (4), input the fluorescence sensor array image with the labeled TVB-N standard value. After the model extracts the features, the decision function is used to output the model's prediction results for the TVB-N content in the beef sample to be tested.

[0106] Depend on Figure 7 It can be seen that the root mean square error (RMSEP) of the best TVB-N prediction model (VGG-19) for the prediction set samples is 1.89 mg / 100 g, and the regression coefficient (R 2 ) can reach 0.9596.

[0107] Example 3

[0108] This example provides three different freshness fluorescent labels for comparison, as follows:

[0109] Comparative Example 1

[0110] Preparation of a single-sensor freshness fluorescent label: Prepare an 8g / L DPA-CuNCs aqueous solution, take 100μL and drop it onto the label substrate prepared in Example 2. Place it in a 40℃ oven and dry it for 30 minutes. This will produce a single-sensor freshness fluorescent intelligent indicator label based on DPA-CuNCs. Ammonia gas of different concentrations was detected according to the method in step 3 of Example 2. The results are as follows: Figure 8 As shown, the fluorescence color of the freshness fluorescent label based on a single sensor unit increases with the increase of ammonia concentration, and the Euclidean distance also increases with it, but the range of change is smaller than that of the fluorescent sensor array label prepared in step 3 of Example 2, and the degree of change is small, which shows that the fluorescence change of a single sensor unit cannot fully and accurately indicate the change in freshness of beef and the change in TVB-N content during the storage period. This is because the freshness fluorescent label of a single sensor unit has only one response area, rather than an array label composed of multiple response areas as in the above embodiment. Since the multiple response areas in the array respond to different substances to different degrees, and the fluorescent array has a cross-response effect, the response areas verify each other and can be applied to the detection of multiple and complex substances. In addition, the designed sensor unit does not have high requirements for the specificity of the target analyte and has a certain degree of versatility. With the continuous updating and iteration of the pattern recognition algorithm, the multi-dimensional and multi-information fluorescent array label can further improve the accuracy of predicting the freshness of meat.

[0111] Comparative Example 2

[0112] Preparation of freshness fluorescent array labels based on fluorescent dyes: Prepare 0.2g / L fluorescein (fluorescein isothiocyanate, 7-hydroxycoumarin, rhodamine 6G, 4,4'-bis(2-sulfonylphenyl)dione disodium, acriflavine, fluorescein) ethanol solution, take 100μL and add it dropwise to the label substrate prepared in step 2 of Example 2, then place it in a 40℃ oven to dry for 30 minutes to obtain a freshness fluorescent array label based on commercial fluorescein. Different concentrations of ammonia were detected according to the method in step 3 of Example 2. The results are as follows: Figure 9 As shown in the data, the fluorescence array prepared based on fluorescein can indicate the freshness of beef to a certain extent. However, with the increase of ammonia concentration, the Euclidean distance of the fluorescence array changes slowly, which means that the array cannot indicate the change of freshness in the later stage of storage. This is mainly due to the presence of several fluoresceins on the array that do not specifically bind to TVB-N or change with the pH change of the environment.

[0113] Comparative Example 3

[0114] The freshness fluorescent array smart label obtained in step 2 of Example 2 is used to detect the freshness of beef. Unlike step 4 of Example 2, this example uses the Euclidean distance change ΔE of the color of the nine fluorescent sensors in the fluorescent array during the beef storage period to perform a linear fit with the change in TVB-N content, and then predicts the change in the freshness of the beef by the Euclidean distance change of the color of the fluorescent sensor array, as shown in FIG. Figure 10 As shown in the figure, the goodness of fit between the Euclidean distance change of color and TVB is 0.75, which shows that the prediction accuracy of TVB-N by this method has decreased.

[0115] Example 4

[0116] like Figure 11 As shown, in this embodiment, a meat freshness intelligent detection system based on a fluorescence sensor array is provided, which includes a label preparation module, a model construction and training module, and a meat freshness detection module;

[0117] The label preparation module is used to prepare fluorescent sensor array indicator labels;

[0118] The model building and training module is used to react the fluorescent sensor array indicator label with meat, photograph and label it, obtain a fluorescent sensor array label image dataset with freshness labels, and build and train a meat freshness intelligent detection model;

[0119] The meat freshness detection module is used to detect the freshness of the meat to be tested using the fluorescent sensor array indicator label and the trained meat freshness intelligent detection model.

[0120] It should be noted here that the system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure can be divided into different functional modules to complete all or part of the functions described above. The system is an intelligent meat freshness detection method based on a fluorescent sensor array applied to the above embodiment.

[0121] Example 5

[0122] like Figure 12 As shown, in this embodiment, a storage medium is further provided, which stores a program. When the program is executed by the processor, the intelligent meat freshness detection method based on the fluorescence sensor array of the above embodiment is implemented, specifically:

[0123] preparing a fluorescent sensor array indicator label;

[0124] The fluorescent sensor array indicator tag reacts with the meat, photographs it, and labels it to obtain a fluorescent sensor array label image dataset with freshness labels, and builds and trains an intelligent meat freshness detection model;

[0125] The freshness of the meat to be tested is detected using the fluorescent sensor array indicator tag and the trained meat freshness intelligent detection model.

[0126] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0127] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. An intelligent meat freshness detection method based on a fluorescence sensor array, characterized in that: The steps include: Preparation of fluorescent sensor array indicator labels: Step 1: Preparation of fluorescent sensing indicator: 100 μL of Cu(NO₃)₂·3H₂O solution was added to 40 mL of D-penicillamine (DPA) solution. Under stirring at 35°C, 1 mol / L HCl solution was added dropwise to raise the pH of the precursor solution to 4.

5. After stirring for 1 hour, the reaction solution was centrifuged at 8000 rpm for 10 minutes to remove the unreacted DPA solution. The solution was then freeze-dried in a vacuum to obtain fluorescent copper nanoclusters (DPA-CuNCs). 0.2 g of copper nanoclusters was dissolved in 25 mL of aqueous solution, and 0.2 g of fluorescein was dissolved in 100 mL of ethanol solution. The mixture was then stirred evenly according to the volume ratios listed in the table below to obtain nine fluorescent sensing indicators: ； Step 2: Preparation of fluorescent sensor array indicator labels: Cut a hydrophobic film into multiple sensor response areas, and stick a paper-based material between two hydrophobic films to form a fluorescent array sensor tag substrate with the function of isolating the sensor units from each other; The above-mentioned 9 fluorescent indicators are added dropwise to the response area of ​​the fluorescent array sensor tag substrate by an artificial spotting method and then dried to obtain a fluorescent sensor array indicator tag; A fluorescent sensor array indicator label was affixed to the top of a fresh-keeping box and stored with fresh meat in a constant temperature and humidity chamber. Real-time digital images of the sensor array were acquired under ultraviolet light. The TVB-N standard value of the fresh meat was measured according to the physical and chemical indicators specified in the national standard GB5009.228-2016. The images were then annotated with freshness grade labels and TVB-N content labels to generate a labeled fluorescent sensor array image dataset. Build and train an intelligent meat freshness detection model; The meat freshness intelligent detection model includes a meat freshness evaluation model and a meat TVB-N content prediction model, specifically: The labeled fluorescence sensor array image dataset was preprocessed to remove noise, normalize, and set the batch size. Classifier labels were then set based on the classification category. Then, based on the idea of ​​transfer learning, the images were input into a deep convolutional neural network pre-trained on the ImageNet database. After optimizing the model parameters, a meat freshness assessment model was obtained. Repeat the above steps to establish a meat TVB-N content prediction model based on a deep convolutional neural network. Unlike the meat freshness assessment model, this meat TVB-N content prediction model is trained using TVB-N standard values ​​annotated in fluorescence sensor array images rather than category information. The fluorescence sensor array images to be tested are input into the meat freshness assessment model and the meat TVB-N content prediction model respectively, and the prediction results of the fresh meat freshness category and TVB-N content are output respectively; The freshness of the meat to be tested is detected using the fluorescent sensor array indicator tag and the trained meat freshness intelligent detection model.

2. The meat freshness intelligent detection method based on fluorescence sensor array according to claim 1 is characterized in that: The feasibility of using the prepared fluorescent sensor array indicator tag to detect volatile spoilage marker gases in meat is also verified, specifically: The volume of volatile corruption marker gas with different gas concentrations is calculated based on the gas diffusion formula. The fluorescent sensor array label is attached to a sealed container to react with volatile corruption marker gas with different concentrations. The color change of the label is captured under ultraviolet conditions. The color change value ∆ before and after the fluorescent sensor array is calculated based on the color Euclidean distance formula. E , and according to ∆ E The size of the volatile spoilage markers was used to distinguish different types and concentrations, thereby verifying the feasibility of the prepared fluorescence sensing array in monitoring the freshness changes of meat during spoilage. The volatile corruption marker gases include biogenic amines, hydrogen sulfide, aldehydes, alcohols, ethers, alkanes and CO2; The gas diffusion formula is specifically: ； in, V μL is the volume of the reaction solution, D mg / L is the density of the reaction solution, W is the mass fraction of the reaction solution, M g / mol is the mole fraction of the reaction solution, V L is the volume of the container; The color change is based on the Euclidean distance ∆ E change: ； where ∆ R ,∆ G ,∆ B They are the differences of R, G, and B color channels before and after the reaction.

3. Intelligent meat freshness detection system based on fluorescence sensor array, characterized by: The meat freshness intelligent detection method based on a fluorescence sensor array as claimed in any one of claims 1 to 2 comprises a fluorescence sensor array indicator label preparation module, a model building and training module, and a meat freshness detection module; The label preparation module is used to prepare fluorescent sensor array indicator labels; The model building and training module is used to react the fluorescent sensor array indicator label with meat, photograph and label it, obtain a fluorescent sensor array label image dataset with freshness labels, and build and train a meat freshness intelligent detection model; The meat freshness detection module is used to detect the freshness of the meat to be tested using the fluorescent sensor array indicator label and the trained meat freshness intelligent detection model.

4. A storage medium storing a program, characterized in that: When the program is executed by a processor, the intelligent meat freshness detection method based on a fluorescence sensor array according to any one of claims 1 to 2 is implemented.

Images