Method and system for constructing freshness detection model of low-temperature marinated beef

Abstract

The invention relates to the technical field of food quality detection, and discloses a low-temperature marinated beef freshness detection model construction method and a system thereof.The low-temperature marinated beef freshness detection model is obtained through training by combining the chromogenic reaction of a mulberry anthocyanin indicating film with a convolutional neural network image recognition technology. The accurate, rapid and nondestructive detection on the freshness of the low-temperature marinated beef is realized; the marinated beef freshness detection result can be obtained only by shooting the mulberry anthocyanin indication film picture of the to-be-detected marinated beef through the intelligent terminal, the defects that a traditional detection method is complex in operation, long in consumed time and depends on expensive professional equipment are effectively overcome, the detection cost is greatly reduced while the detection efficiency and convenience are remarkably improved, and the detection efficiency is greatly improved. And a feasible solution is provided for quality monitoring in the meat food circulation process.

Description

Technical Field

[0001] This invention relates to the field of food quality testing technology, and in particular to a method and system for constructing a freshness testing model for low-temperature braised beef. Background Technology

[0002] Braised beef, as a traditional delicacy, holds an important place in culinary culture due to its rich nutritional value and unique flavor. Braised beef is divided into high-temperature sterilized products and low-temperature sterilized products. Low-temperature sterilized products are favored by consumers for their reddish color, tender texture, and rich aroma. However, because low-temperature braised beef is rich in protein and water, it is highly susceptible to spoilage during storage and transportation due to temperature, moisture, and microbial activity. This not only increases the risk of foodborne illnesses but also causes unnecessary waste. As people increasingly value food safety, monitoring the freshness of meat products has become crucial. Common techniques for detecting the freshness of meat products include physicochemical testing, electrochemical analysis, spectroscopic techniques, chromatography, and electronic noses. However, most of these traditional methods rely on measuring specific chemical or physical signal changes produced during the spoilage process of braised beef. Their application is often limited by dependence on specialized instruments, complex sample pretreatment, or large databases. For example, chromatography requires sophisticated equipment and is time-consuming, while the accuracy of electronic noses depends on a pre-established and comprehensive odor model. Therefore, these methods generally suffer from problems such as complex operation, long detection cycle, and high cost, making it difficult to meet the needs of rapid and convenient monitoring in actual production and distribution. Summary of the Invention

[0003] This invention provides a method and system for constructing a freshness detection model for low-temperature braised beef, in order to overcome the shortcomings of existing technologies.

[0004] This invention provides a method for constructing a freshness detection model for low-temperature braised beef, comprising: Images of mulberry anthocyanin indicator membranes were obtained for low-temperature braised beef samples of different freshness. Based on the images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, a convolutional neural network was used to train a model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thus obtaining a freshness detection model for low-temperature braised beef.

[0005] According to the present invention, a method for constructing a freshness detection model for low-temperature braised beef includes obtaining images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, comprising: Preparation of mulberry anthocyanin indicator membrane; The mulberry anthocyanin indicator membrane and the low-temperature braised beef sample were placed in the same environment; The freshness of low-temperature braised beef samples was periodically tested using quality indicators of braised beef, and images of mulberry anthocyanin indicator membranes were obtained simultaneously, thus yielding images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples with different freshness levels.

[0006] According to the present invention, a method for constructing a freshness detection model for low-temperature braised beef is provided, using mulberry anthocyanin extract as an indicator and gelatin and carboxylated cellulose nanofibers as film-forming matrices to prepare a mulberry anthocyanin indicator membrane.

[0007] According to the method for constructing a freshness detection model for low-temperature braised beef provided by the present invention, the amount of mulberry anthocyanin added to the mulberry anthocyanin indicator membrane is 3%.

[0008] According to the present invention, a method for constructing a freshness detection model for low-temperature braised beef is provided. The quality indicators of braised beef include any one of the following or any combination thereof: pH, total bacterial count (TVC), and volatile basic nitrogen (TVB-N) level.

[0009] According to the present invention, a method for constructing a freshness detection model for low-temperature braised beef is provided. The freshness of low-temperature braised beef samples includes any of the following types or any combination thereof: fresh, slightly fresh, and spoiled.

[0010] According to the present invention, a method for constructing a freshness detection model for low-temperature braised beef is provided, wherein the convolutional neural network includes VGG-16, ResNet-34 and ResNet-50.

[0011] According to the present invention, a method for constructing a freshness detection model for low-temperature braised beef includes, based on images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, using a convolutional neural network to train a model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thereby obtaining a freshness detection model for low-temperature braised beef. The method comprises: Based on the images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, the VGG-16 convolutional neural network was used to train the model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thus obtaining the first low-temperature braised beef freshness detection model. Based on the images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, a ResNet-34 convolutional neural network was used to train a model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thus obtaining a second low-temperature braised beef freshness detection model. Based on the images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, a ResNet-50 convolutional neural network was used to train the model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thus obtaining the third low-temperature braised beef freshness detection model. By comparing the changes in training loss and accuracy of the first, second, and third low-temperature braised beef freshness detection models during the model training process, the final low-temperature braised beef freshness detection model is obtained.

[0012] According to the present invention, a method for constructing a freshness detection model for low-temperature braised beef is provided. The final freshness detection model for low-temperature braised beef is a model trained using a ResNet-50 convolutional neural network.

[0013] This invention also provides a method for detecting the freshness of low-temperature braised beef, comprising: Receive, from at least one terminal, a user-inputted image of the mulberry anthocyanin indicator membrane of the low-temperature braised beef to be tested; The image of the mulberry anthocyanin indicator membrane to be tested is input into the low-temperature braised beef freshness detection model obtained by the construction method of the low-temperature braised beef freshness detection model described above, and the freshness detection result of the low-temperature braised beef to be tested is obtained. The freshness test results of the low-temperature braised beef to be tested are sent to at least one terminal.

[0014] It should be noted that a terminal refers to an input / output device connected to a computer system. Depending on the function, terminals can be divided into various types: smart terminals or intelligent terminals, dumb terminals, interactive terminals or online terminals. Specifically, a terminal can be various mobile communication devices, such as mobile phones and tablets. This article aims to provide users with the function of inputting data and outputting data.

[0015] This invention also provides a system for detecting the freshness of low-temperature braised beef, comprising: The data receiving module is used to: receive images of the mulberry anthocyanin indicator membrane of the low-temperature braised beef to be tested, input by the user; The low-temperature braised beef freshness detection module is used to: obtain the freshness detection result of the low-temperature braised beef based on the mulberry anthocyanin indicator membrane image of the low-temperature braised beef to be tested, through the low-temperature braised beef freshness detection model obtained by any of the above-mentioned methods.

[0016] The present invention also provides an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the computer program to implement the method for constructing a low-temperature braised beef freshness detection model and / or the method for detecting the freshness of low-temperature braised beef as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for constructing a low-temperature braised beef freshness detection model and / or the method for detecting the freshness of low-temperature braised beef as described above.

[0018] The present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute any of the above-described methods for constructing a low-temperature braised beef freshness detection model and / or a low-temperature braised beef freshness detection method.

[0019] The present invention provides a method and system for constructing a freshness detection model for low-temperature braised beef, which can bring at least the following beneficial effects: This invention combines the colorimetric reaction of mulberry anthocyanin indicator membranes with the image recognition capabilities of convolutional neural networks, eliminating the complex sample pretreatment, instrument analysis, and data parsing processes of traditional detection methods. Users only need to take a single image of the mulberry anthocyanin indicator membrane to obtain the freshness test results for low-temperature braised beef in a short time, greatly improving detection efficiency and meeting the urgent need for real-time monitoring in production and distribution processes.

[0020] This invention relies on a pre-trained detection model and common image acquisition devices (such as mobile terminals like smartphones and tablets), eliminating the need for expensive and bulky professional analytical instruments (such as chromatographs and spectrometers). This effectively reduces hardware investment and maintenance costs, making it particularly suitable for widespread application in small and medium-sized enterprises, logistics warehouses, supermarkets, and even homes.

[0021] The entire process of detecting the freshness of low-temperature braised beef according to this invention requires no intervention from professional technicians. Ordinary users can complete image uploading, detection, and result reception via a smart terminal with simple guidance. The system encapsulates the complex model judgment process in the background, providing users with convenient "one-click" operation and completely eliminating reliance on traditional complex detection processes.

[0022] This invention utilizes the powerful feature learning and pattern recognition capabilities of convolutional neural networks. The model can accurately extract complex color and texture features that are deeply related to the freshness of beef from images of mulberry anthocyanin indicator films, effectively avoiding errors in subjective human judgment, and thus providing more objective, accurate and reliable detection results than traditional methods. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 middle, Figure 1 a shows the color of mulberry anthocyanin extract in the pH range of 2-12; Figure 1 b shows the UV-Vis spectrum of mulberry anthocyanin extract in the pH range of 2–12; Figure 1 c shows the structural changes of anthocyanins at different pH values.

[0025] Figure 2 The effect of the amount of MAE (mulberry anthocyanin) added on the color of the mulberry anthocyanin indicator film is shown.

[0026] Figure 3 The image shows the color changes (a) and RGB value differences (b) of the mulberry anthocyanin indicator membrane in phosphate buffer at pH 2–12.

[0027] Figure 4 The ammonia sensitivity analysis (a) and RGB value difference image (b) of the mulberry anthocyanin indicator membrane at different times are shown.

[0028] Figure 5 The color difference (a) and sensory score (b) of braised beef during storage are shown.

[0029] Figure 6 The changes in TVC during the storage of braised beef are shown.

[0030] Figure 7 The changes in TVB-N during the storage of braised beef are shown.

[0031] Figure 8 The pH changes during the storage of braised beef are shown.

[0032] Figure 9 The correlation between various physicochemical indicators during the storage of braised beef is shown.

[0033] Figure 10 The color changes of the indicator membrane during the storage of braised beef are shown (a), and the canonical correlation analysis between the RGB values ​​of the G / CCN / Mg2+ / MAE-3 indicator membrane and sensory scores, TVC, TVB-N, and pH is shown (be).

[0034] Figure 11The training history of three models is shown: training loss (a) and accuracy (b) of the Vgg16 model; training loss (c) and accuracy (d) of the ResNet-34 model; and training loss (e) and accuracy (f) of the ResNet-50 model.

[0035] Figure 12 The confusion matrix (a) and ROC curve (c) of the ResNet-34 model are shown; the confusion matrix (b) and ROC curve (d) of the ResNet-50 model are shown.

[0036] Figure 13 This demonstrates the development of an intelligent detection platform based on the ResNet-50 model. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0038] Figure 1 This is a flowchart illustrating a method for constructing a freshness detection model for low-temperature braised beef provided by the present invention. The execution entity of this method can be any applicable terminal-side device or network-side device, such as a device for constructing a freshness detection model for low-temperature braised beef.

[0039] See Figure 1 The present invention provides a method for constructing a freshness detection model for low-temperature braised beef, which may include: S110. Obtain images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness.

[0040] In one embodiment, S110 may include: A mulberry anthocyanin indicator membrane was prepared using mulberry anthocyanin extract (MAE) as an indicator and gelatin and carboxylated cellulose nanofibers as film-forming matrices, wherein the amount of mulberry anthocyanin added was 3%. The mulberry anthocyanin indicator membrane and the low-temperature braised beef sample were placed in the same environment; The freshness of low-temperature braised beef samples was periodically tested using quality indicators of braised beef, and images of mulberry anthocyanin indicator films were simultaneously obtained. This yielded images of mulberry anthocyanin indicator films corresponding to low-temperature braised beef samples with different freshness levels. The quality indicators of braised beef included any one of the following or any combination thereof: pH, total bacterial count (TVC), and volatile basic nitrogen (TVB-N) level. The freshness of low-temperature braised beef samples included any one of the following types or any combination thereof: fresh, slightly fresh, and spoiled.

[0041] S120. Based on the images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, a convolutional neural network is used to train a model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thereby obtaining a low-temperature braised beef freshness detection model.

[0042] In one embodiment, S120 may include: Based on the images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, the VGG-16 convolutional neural network was used to train the model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thus obtaining the first low-temperature braised beef freshness detection model. Based on the images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, a ResNet-34 convolutional neural network was used to train a model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thus obtaining a second low-temperature braised beef freshness detection model. Based on the images of mulberry anthocyanin indicator membranes corresponding to low-temperature braised beef samples of different freshness, a ResNet-50 convolutional neural network was used to train the model to learn the relationship between low-temperature braised beef of different freshness and the corresponding mulberry anthocyanin indicator membranes, thus obtaining the third low-temperature braised beef freshness detection model. By comparing the changes in training loss and accuracy of the first, second, and third low-temperature braised beef freshness detection models during the model training process, the final low-temperature braised beef freshness detection model is obtained. The final low-temperature braised beef freshness detection model is a model trained using a ResNet-50 convolutional neural network.

[0043] Based on the obtained freshness detection model for low-temperature braised beef, a method for detecting the freshness of low-temperature braised beef can be developed, including: Receive, from at least one terminal, a user-inputted image of the mulberry anthocyanin indicator membrane of the low-temperature braised beef to be tested; The image of the mulberry anthocyanin indicator membrane to be tested was input into the freshness detection model of low-temperature braised beef to obtain the freshness detection result of the low-temperature braised beef to be tested. The freshness test results of the low-temperature braised beef to be tested are sent to at least one terminal.

[0044] The present invention is described below through specific embodiments.

[0045] This embodiment utilizes a web-based platform to detect the freshness of low-temperature braised beef using a mulberry anthocyanin indicator membrane and a deep learning model. First, a pH-responsive indicator membrane (i.e., a mulberry anthocyanin indicator membrane) is prepared using mulberry anthocyanin extract (MAE) as an indicator and gelatin and carboxylated cellulose nanofibers as the film-forming matrix. The effects of anthocyanin addition on the physical properties, pH stability, and ammonia response value of the indicator membrane are studied to determine the optimal anthocyanin addition amount. Then, the mulberry anthocyanin indicator membrane is applied to the low-temperature storage of braised beef. Changes in the quality of the braised beef (TVC, pH value, TVB-N value, etc.) and the color change of the indicator membrane are measured, and correlation analysis is performed to determine three freshness levels for the braised beef: "fresh," "medium-fresh," and "spoiled." Next, the acquired mulberry anthocyanin indicator membrane was trained and tested using Vgg-16, ResNet-34, and ResNet-50 models in a convolutional neural network algorithm at an 8:2 ratio. Images of the indicator membrane were then uploaded to a web application via a mobile device in real time to form a complete image classification solution. This embodiment provides a new method for the intelligent and rapid detection of the freshness of low-temperature braised beef and also provides a theoretical basis for the application of deep learning in food testing.

[0046] Materials and reagents Braised beef was purchased from a beef company in Shanxi Province; mulberries were purchased from Jiajiali Supermarket in Taigu, Shanxi Province; acrylic sheet mold (15×15 cm) was purchased from Tomson Materials Technology Co., Ltd.; carboxylated cellulose nanofibers were purchased from Shanghai Maclean Biochemical Technology Co., Ltd.; gelatin and glycerin were purchased from Beijing Solarbio Technology Co., Ltd.; magnesium chloride, sodium hydroxide, hydrochloric acid, and anhydrous ethanol were purchased from Tianjin Zhiyuan Chemical Reagent Co., Ltd. All chemicals used in the experiment were of analytical grade.

[0047] Extraction and content determination of anthocyanins from mulberries Anthocyanins were extracted from mulberries. Fresh mulberries were washed, dried at 40°C, and pulverized. 10 g of mulberry powder was weighed and added to 100 mL of 80% ethanol solution at a solid-liquid ratio of 1:10 (g / mL). The pH of the solution was adjusted to 2.0 ± 0.1 with 1 mol / mL HCl. The mixture was sealed and allowed to stand in the dark at 4°C for 24 hours. Afterward, the solution was filtered through filter paper, and the filtrate was centrifuged at 8000 rpm for 10 minutes. The extraction process was repeated twice on the centrifuged residue. Finally, all extracts were combined, and the ethanol was removed by evaporation to obtain a concentrated mulberry anthocyanin extract (MAE). The concentrate was brought to a final volume of 80 mL and stored in the dark at 4°C for subsequent experiments.

[0048] The anthocyanin content of mulberry was determined by pH differential method. 3.98 mL of potassium chloride buffer (pH=1.0) and sodium acetate buffer (pH=4.5) were placed in test tubes, and 0.02 mL of the sample to be tested was added to each. The tubes were then incubated at room temperature in the dark for 30 min. Using distilled water as a blank, the absorbance values ​​at wavelengths of 520 nm and 700 nm were measured. The anthocyanin extraction yield was calculated according to formula (1).

[0049] (1) In the formula: X: mulberry anthocyanin extraction amount, mg / 100g; A: [(A520nm-A700nm)pH1.0-(A520nm-A700nm)pH4.5]; V: volume after dilution, mL; L: width of cuvette, 1 cm; m: sample mass, g; ε: average molar extinction coefficient of cyanidin-3-O-glucoside, 26900 L / (mol·cm); Mw: molar mass of cyanidin-3-O-glucoside, 449.38 g / mol; DF: dilution factor.

[0050] Color response and UV-Vis spectrum of anthocyanin extracts 0.25 mL of MAE concentrate was added to 15 mL of buffer solutions with pH values ​​ranging from 2 to 12. The color changes of MAE in different pH buffer solutions were observed and recorded using a smartphone. The UV-Vis spectra of anthocyanin solutions at different pH values ​​were scanned using a UV-Vis spectrophotometer, with a scanning wavelength range of 400 nm to 800 nm.

[0051] Preparation of freshness indicator film 2 g of G was soaked in 100 mL of distilled water to swell for 30 min, and then stirred at 60℃ for 30 min to obtain a 2% (w / v) G solution. 0.1 g of CCN was added to 30 mL of distilled water, along with an appropriate amount of the extracted MAE, to obtain a CCN solution containing MAE. The G solution and the CCN / MAE solution were mixed, and then 1 g of glycerol was added to adjust the pH to 4.5. The mixture was stirred for 4 h to form a film-forming solution. The film-forming solution was ultrasonicated (50 W, 10 min) to eliminate air bubbles, then poured into an acrylic mold (15 × 15 cm), dried at 35℃ for 48 h, and the membrane was removed and allowed to stand in the dark for 48 h to form a thin film.

[0052] Metal ion crosslinking: The film was immersed in a 50% ethanol-water solution containing 2% MgCl2 (pH=6, containing 20% ​​glycerol) for 4 min, then rinsed with distilled water to remove excess metal cations, and dried at 25℃ for 48 h to obtain a crosslinked film with Mg²⁺. + Cross-linked films.

[0053] Grouping: A total of 7 thin films were prepared in the experiment. Among them, films without MAE and without interaction with Mg²⁺ were prepared. + Crosslinked films are labeled G / CCN; no MAE is added but with Mg²⁺ + The cross-linked film is labeled G / CCN / Mg² + The other five types of films all contained MAE and were combined with Mg²⁺. + Crosslinking was performed, and the MAE addition amounts were 1 mL, 2 mL, 3 mL, 4 mL, and 5 mL, respectively, so they were labeled as G / CCN / Mg. 2+ / MAE-1、G / CCN / Mg 2+ / MAE-2、G / CCN / Mg 2+ / MAE-3、G / CCN / Mg 2+ / MAE-4、G / CCN / Mg 2+ / MAE-5.

[0054] Indicator film thickness and mechanical properties: The thickness (mm) of the film was measured at 10 random locations using a digital micrometer, and the average value was calculated.

[0055] The film was cut into rectangular strips of 70 mm × 10 mm, and the tensile strength (TS) and elongation at break (EB) of the film were measured using a universal testing machine (INSTRON 5544, Inström, USA). The initial clamping distance and tensile speed were 40 mm and 50 mm / min, respectively, and each sample was measured 5 times. TS and EB were calculated according to formulas (2) and (3), respectively: (2) (3) F m This represents the maximum tensile force (N) that the membrane sample can withstand when it breaks. W The width of the membrane (mm). d The thickness of the membrane is in mm. L 1 represents the maximum tensile length (mm) that the membrane can achieve when it breaks. L 0 represents the initial clamping length (mm).

[0056] Indicator membrane pH sensitivity: The membrane was cut into approximately 2 cm × 2 cm pieces and immersed in phosphate buffer solutions with different pH values ​​(2-12) for 1 min. It was then removed, laid on filter paper to dry, and photographed using a smartphone (Samsung Galaxy Z Flip 5) to record the color changes of the indicator membrane. Using Python 3.12.3 and Anaconda, pH response sensitivity analysis was performed on the acquired indicator membrane images (300 × 300 pixels). First, the RGB channel values ​​of the indicator membrane images after adjacent pH responses were extracted. Then, the pH response was analyzed according to the formula... S RGB Calculate the color response sensitivity of the indicator film, where R a , G a , B a This represents the RGB mean of the indicator membrane image under higher pH conditions (e.g., pH 3). R b , G b , B b This represents the RGB mean of the indicator membrane image under lower pH conditions (e.g., pH 2). The color response sensitivity of the indicator membrane is accurately evaluated by calculating the sum of the absolute differences in the RGB values ​​of the indicator membrane after two adjacent pH responses. The formula for calculating the pH response sensitivity of the indicator membrane is as follows: S RGB As shown in (4): (4) Indicator membrane ammonia sensitivity: The membrane sample (2 cm × 2 cm) was placed at the top of a centrifuge tube containing 10 mL of ammonia (25%). Photos were taken using a smartphone (Samsung Galaxy Z Flip 5) at 0, 5, 10, 15, 20, 40, and 60 min to record the results. The method for determining the ammonia-sensitive response sensitivity of the indicator membrane was the same as that for the pH-sensitive response sensitivity of the indicator membrane. R a, G a , B a This represents the RGB mean value of the indicator membrane image under longer time conditions (e.g., 5 minutes). R b , G b , B b This represents the RGB mean of the indicator membrane image under shorter time conditions (e.g., 1 min), with the rest remaining the same.

[0057] Application of indicator membrane in monitoring the freshness of braised beef Using PE packaging boxes as the packaging material, the boxes were first wiped with alcohol and irradiated with ultraviolet light. Then, 300±5 g of braised beef was placed inside the box, and a 2×2 cm indicator film was affixed to the top of the inside of the box. The beef was stored at 4℃ for 15 days. During storage, the appearance of the beef and the color of the indicator film were recorded daily using a smartphone (Samsung Galaxy Z Flip 5), while the color difference value of the braised beef was measured using a colorimeter (CM-5, Konica Minolta, Japan). The TVC, pH value, and TVB-N value of the beef were also measured.

[0058] Deep learning model design Indicator membrane freshness grade classification Based on a comprehensive evaluation of the measurement results, this embodiment selects the TVB-N value as the assessment index for the freshness of braised beef. A TVB-N value < 20 mg / 100 g is considered fresh, 20 mg / 100 g < TVB-N value < 30 mg / 100 g is considered sub-fresh, and a TVB-N value > 30 mg / 100 g is considered spoiled. A total of 371 indicator membrane images were collected as the dataset for the deep learning model.

[0059] Image contour detection and region extraction based on OpenCV Image contour detection and region extraction were performed using the OpenCV library in Python. The specific method is as follows: First, median filtering was used to convert the image's RGB values ​​to grayscale values ​​to reduce salt-and-pepper noise. Then, thresholding was performed, with the optimal threshold automatically calculated based on the image histogram as the initial threshold. The optimal threshold coverage was calculated using Otsu's Mothod algorithm, and contour detection was performed on the resulting binary image. Noisy contours were filtered by setting a minimum area threshold, internal sub-contours were detected, and key inflection points were preserved. Subsequently, the corresponding region was cropped from the original image, and the image's grayscale values ​​were converted back to RGB values. Finally, irregular edges in the image were further cropped to prevent them from affecting subsequent experiments.

[0060] Data preprocessing and augmentation strategies This embodiment employs two data loading modes: data loading based on directory structure (organized by subdirectories from day1 to dayN) and loading via a specified path in a CSV file. The training and test sets are divided in an 8:2 ratio, and stratified sampling is used to maintain consistency in category distribution.

[0061] Data augmentation techniques (DA) are methods to increase the size and diversity of datasets by altering and processing existing data. The aim is to improve model generalization ability, prevent overfitting, and effectively address the problem of insufficient data. Composite data augmentation schemes are as follows: (1) Geometric transformation: random horizontal / vertical flip (probability 0.5), ±20° rotation, affine transformation (translation ±10%, scaling 0.9-1.1 times); (2) Color disturbance: brightness, contrast, saturation adjustment (±5%), hue shift (±0.02); (3) Noise injection: Gaussian noise (mean 0, standard deviation 0.005, application probability 10%), salt and pepper noise (density 0.005, salt noise probability 2.5%, pepper noise probability 2.5%, application probability 10%). (4) Regularization: The center is cropped to 224×224 and ImageNet standard normalization is applied (mean [0.485, 0.456, 0.406], standard deviation [0.229, 0.224, 0.225]).

[0062] The test set only undergoes basic processing such as scaling and normalization to avoid data leakage. Images are preloaded into memory using the TestPaperDataset class, and multi-threaded acceleration using DataLoader (workers=8, prefetch_factor=4) is combined to improve training efficiency.

[0063] Model architecture and training optimization ResNet-50 was used as the base network, initialized with ImageNet pre-trained weights (ResNet50_Weights.IMAGENET1K_V2), retaining the original convolutional layer parameters, and replacing the final fully connected layers with task-related classification heads (output dimension = 3 (number of classes)). The following optimization strategies were adopted during training: (1) Loss function: cross-entropy loss combined with label smoothing (label smoothing=0.1). (2) Optimizer: Adam algorithm, initial learning rate 1e-4; (3) Learning rate scheduling: StepLR strategy, decaying by 50% every 10 epochs; (4) Regularization techniques: gradient clipping (max_norm=1.0), weight decay (L2 regularization).

[0064] Training process and evaluation indicators Model training was performed on an NVIDIA GPU hardware environment with a batch size of 128 and 100 training epochs. The loss and accuracy were recorded for each training / test epoch. After training, the best-performing model was loaded for performance evaluation.

[0065] Web Interface Design The system adopts a three-layer architecture. The front end receives user-uploaded instruction membrane images via responsive HTML templates, while the back end integrates a PyTorch-driven ResNet-50 deep neural network, forming a complete image classification solution. In terms of model construction, the fully connected layers of the pre-trained ResNet-50 model are adaptively modified, replacing the original 1000-class ImageNet classifier with a three-class freshness classifier, and domain-specific weights are loaded using transfer learning strategies. A device-adaptive model loading mechanism ensures system compatibility across different hardware environments through dynamic mapping of storage devices (CUDA / CPU), balancing GPU-accelerated inference with CPU backup support capabilities.

[0066] The core processing workflow employs a standardized image preprocessing pipeline, utilizing techniques such as spatial dimension adjustment (224×224 resolution), tensor transformation, and channel normalization (ImageNet mean and variance parameters) to eliminate input data distribution bias. The request processing module adopts a RESTful architecture with built-in multiple security verification mechanisms, including file type whitelist filtering (supporting PNG / JPEG formats), path security handling (secure_filename), and empty file defense strategies. The inference engine obtains the category probability distribution through the Softmax function, returning structured data containing Top-3 prediction results, where confidence levels are presented as percentages, effectively enhancing the interpretability of the results.

[0067] On PCs using Apple Silicon M4 chips, the system's single inference latency is kept below 50 milliseconds, meeting the requirements for real-time interaction.

[0068] Data processing All experiments were repeated three times, and the results were taken as mean ± standard deviation. Mean values ​​were calculated using Microsoft Excel 2013 software, and statistical significance was analyzed using IBM SPSS Statistix 24 software. A significance level of P < 0.05 was considered. Graphs were plotted using Origin 2025 software.

[0069] Results and Analysis Color response and visible-ultraviolet spectrum of mulberry anthocyanin extract (MAE) at different pH values The concentration of MAE was determined by pH differential method. Based on the formula and relevant absorbance values, the content of MAE extracted in this experiment was calculated to be 80 mg / 100 g. Figure 1 a represents the color change of MAE at different pH values. Figure 1 b represents the MAE UV-Vis absorption spectrum. Figure 1 c represents the possible molecular structures of MAE at different pH values. The color characteristics of MAE are significantly dependent on pH, and the changes are essentially due to the dynamic transformation of its molecular structure. Under strongly acidic conditions (pH 2–3), the solution is pink, and a characteristic absorption peak appears at 514 nm in the ultraviolet spectrum, indicating that anthocyanins mainly exist in the form of flavylium cations. When the pH rises to the weakly acidic to neutral range (4–6), the solution fades to a light pink color, and the absorption peak in the 500–600 nm range disappears (absorbance <0.5). This is because the flavylium cation is converted into a carbinol pseudobase / hemiketal through a nucleophilic addition reaction, and the breakage of its conjugated double bond leads to the loss of absorption capacity in the visible light region.

[28] Simultaneously, a dynamic equilibrium exists between the yellow cation, the quinonoidal base, and the pseudobase.

[29] The residual cations (red) and trace amounts of quinone bases (blue) combine to form a light pink color. Under weakly alkaline to alkaline conditions (pH 7–10), the solution turns purple, with a significant absorption peak at 580 nm and absorbance increasing with increasing pH (0.8→1.2). During this stage, the methanol pseudobase further deprotonates, generating a purple ionic quinone base.

[30] When the pH rises to a strongly alkaline level (pH > 10), the solution eventually turns brownish-yellow and exhibits a broad absorption peak at 380–450 nm, indicating that the quinone base undergoes an irreversible ring-opening reaction to generate a chalcone with a broken conjugated chain.

[31] .

[0070] Physical property analysis of indicator membrane The color of the indicator film: The appearance of the indicator film is a key element in forming a consumer's first impression. Figure 2 The effect of MAE addition on the color of the indicator film. For example... Figure 2 As shown, the amount of MAE added has a significant impact on the color performance of the indicator film. The film without MAE exhibits a smooth texture with high transparency, while with the addition of MAE, the indicator film gradually deepens to a purplish-red color. Notably, when the anthocyanin content exceeds 4%, the color of the indicator film gradually transitions from light purplish-red to a rich eggplant purple, and at the same time, anthocyanin aggregation occurs on the surface of the indicator film.

[0071] Indicator film thickness and mechanical properties: Table 2. Indicator film thickness, tensile strength (TS), and elongation at break (EB)

[0072] Note: Values ​​are expressed as mean ± SD (n = 3). Different lowercase letters in the same column indicate significant differences. p <0.05).

[0073] Table 2 shows the thickness and mechanical properties (TS, EB) of the films. From the table, it can be seen that Mg... 2+ The crosslinking increased the film thickness, significantly increasing it from 0.140 ± 0.002 mm to 0.151 ± 0.002 mm. p <0.05), while the addition of anthocyanins increased the membrane thickness, but the effect was not significant. p >0.05). The G / CCN membrane exhibited the highest TS value (7.32±1.0 MPa), after being treated with Mg... 2+ The cross-linking effect led to a significant decrease in the TS value to 4.27 ± 0.5 MPa. p <0.05). The addition of anthocyanins can improve the TS value of the membrane, reaching a level of 5.23~6.75 MPa. Regarding elongation at break (EB), Mg... 2+ The film elongation was significantly improved, increasing from 36.86±4.56% to 74.79±6.70%. The addition of MAE maintained the EB value between 53.32% and 86.79%. Although there was no significant correlation between the EB value and the amount of MAE added, it is evident that Mg... 2+ The synergistic effect with MAE improved the EB of the indicator membrane.

[0074] pH sensitivity of the indicator membrane: The color change of the indicator membrane in phosphate buffer solutions with pH 2-12 is as follows... Figure 3As shown in Figure a, under acidic conditions (pH 2-4), the color is mainly red. As the pH increases, the color gradually lightens to a pale red; at pH 7-10, the color gradually changes from pale red to blue. When pH > 11, the indicator membrane turns yellowish-brown. This is mainly due to the irreversible transformation of the anthocyanin structure under different pH conditions, corresponding to the aforementioned color changes of anthocyanins under different pH conditions. Figure 1 a) Regarding the amount of MAE added, when the amount of MAE added is 3% or 4%, the color change of the indicator membrane at different pH values ​​is more obvious. Figure 3 b is an image showing the RGB value differences within the pH range of 2-12, further demonstrating this phenomenon. The difference image is generated by calculating the numerical differences in the three color channels (R, G, and B) of pixels between indicator membrane images obtained under two adjacent pH conditions. It can directly reflect the degree of color difference between indicator membrane images obtained under adjacent pH conditions. In the difference image, the higher the brightness value (yellow), the greater the degree of color change.

[0075] Indicator membrane ammonia sensitivity: Figure 4 Figure a shows the ammonia sensitivity analysis of five indicator membranes at different time points. As the figure shows, all membranes underwent significant color changes with increasing exposure time in the ammonia environment, transitioning from purple to green to brownish-yellow to pale yellow within 120 minutes. A high response value to ammonia concentration within a short time is an important indicator for evaluating the ammonia sensitivity of the indicator membrane. Figure 4 The RGB value difference image of b shows that when the amount of MAE added is ≥3%, the indicator membrane has a good response value to ammonia within 5 minutes, which may help the indicator membrane to show a fast color response in actual detection.

[0076] Application of indicator membrane in monitoring the freshness of braised beef Sensory evaluation of braised beef: The sensory quality (color, aroma, taste, and texture) of braised beef during storage can directly affect consumers' acceptance and willingness to buy. Figure 5The physical pictures, color difference, and sensory evaluation of marinated beef stored at low temperature for 1 - 15 days are shown. In the early stage of storage (1 - 7 days), the beef maintained a good color, had a firm and elastic texture, a smooth and moist cut surface without exudate, accompanied by a typical aroma of marinated meat, the overall flavor was rich and harmonious, and each score was higher than 15, being judged as fresh. Entering the middle stage of storage (8 - 11 days), with the evaporation of moisture and mild decomposition of protein, the color became slightly dull, the surface was slightly dry, the aroma decreased, the elasticity decreased, but the taste was acceptable, and this stage was judged as the sub - fresh stage (edible). In the late stage of storage (12 - 15 days), the marinated beef lost its luster, tasted sour, the tissue was loose and sticky, and had an irritating putrid smell and was inedible, and this stage was judged as the spoiled stage.

[0077] Total viable count (TVC): The total viable count (TVC) is an important indicator to measure the hygienic status of meat products and is closely related to the spoilage process of products. Through in - depth research on the relationship between TVC and spoilage, it can provide a scientific basis for shelf - life assessment and food safety risk management. In cooked meat products, there is no clear national standard for TVC, but referring to the requirements of GB / T19694 - 2008, the TVC of marinated beef ≤ 80000 CFU / g (4.90 log CFU / g) is qualified. Figure 6 The change of TVC of marinated beef during storage at low temperature (4℃) is shown. As can be seen from the figure, during storage, the TVC increased from 2.50 log CFU / g on the 1st day to 3.36 log CFU / g on the 7th day, and the TVC on the 11th day reached 4.81 log CFU / g, still within the acceptable range specified by the national standard. However, when storage continued, microorganisms multiplied rapidly, the TVC on the 12th day was as high as 5.12 log CFU / g, exceeding the national standard requirements, and the TVC at the end of storage (the 15th day) reached 5.89 log CFU / g, at which time the beef had obvious spoilage and was inedible.

[0078] TVB - N value: The level of total volatile basic nitrogen (TVB - N) is an important indicator for evaluating the freshness of fresh meat. In cooked meat products, there is no clear national standard for the TVB - N value, but cooked meat products with a TVB - N value < 20 mg / 100g are generally considered fresh. Figure 7The figure shows the changes in TVB-N value during the storage of braised beef. As can be seen from the graph, the TVB-N value showed an increasing trend throughout the storage period. From day 0 to 7 of storage, the increase in TVB-N value was relatively slow, and on day 7, the TVB-N value was <20 mg / 100g, indicating that the beef was fresh at this stage. From day 8 to 11 of storage, the increase in TVB-N value remained relatively slow, and on day 11, the TVB-N value was 28.63 mg / 100g, exceeding 20 mg / 100g but below 30 mg / 100g. From day 12 to 15 of storage, the TVB-N value increased sharply and exceeded 30 mg / 100g, at which point the braised beef had spoiled.

[0079] Monitoring pH changes is often used as an effective means of assessing the freshness and spoilage level of meat products. pH changes are not only related to microbial changes during the spoilage process of braised meat, but also affect the flavor and safety of meat products. Figure 8 The figure shows the pH changes of braised beef during low-temperature (4℃) storage. As can be seen from the figure, the pH showed a decreasing trend (6.09-6.00) during the first 3 days of storage. After 3 days of storage, the pH significantly increased from 6.07 (day 4) to 7.09 (day 15). This is because in the early stages of storage, microorganisms such as lactic acid bacteria fermented residual sugars to produce acidic substances, leading to a decrease in pH.

[45] As storage time increases, protein decomposition and putrefactive bacteria activity intensify, leading to the gradual accumulation of alkaline substances such as ammonia and amines, which in turn causes the pH value to rise.

[0080] Correlation analysis among various physicochemical indicators of braised beef during storage: During the storage of braised beef, there were significant correlations among TVC, TVB-N, pH value and sensory quality. Figure 9 A heatmap showing the correlation between various physicochemical indicators of braised beef during low-temperature (4℃) storage reflects the spoilage process of braised beef. As shown in the figure, storage time is positively correlated with TVC, TVB-N, and pH value. With the extension of storage time, the rapid growth and reproduction of microorganisms accelerates the decomposition of protein in the meat, leading to an increase in TVB-N. In the later stages of storage, alkaline substances such as ammonia and amines produced by decomposition accumulate, and pH value increases significantly. However, storage time is significantly negatively correlated with sensory score and a* value of braised beef. This is because in the early stage of storage (1-7 days), lower TVC, TVB-N, and pH value correspond to higher sensory scores (>20 points), and the braised beef has a reddish color, indicating freshness. In the middle stage of storage (8-11 days), the indicators are close to the threshold, the braised beef has a slightly dull color, and the sensory score drops to near freshness. In the later stage of storage (12-15 days), higher TVC, TVB-N, and pH value correspond to lower sensory scores (<15 points, indicating spoilage), the braised beef loses its luster, the a* value decreases, and the braised beef has spoiled and is no longer edible. The relationship between the color change of the indicator film during storage and the freshness of braised beef: Figure 10 Figure a shows the color change of the indicator film during the storage of braised beef. As can be seen from the figure, the indicator films were all purplish-red on day 1 of storage, turning purple on day 2. This is likely due to moisture escaping from the braised beef entering the indicator film, causing swelling and resulting in color change. From day 3 to 7 of storage, the indicator film was light pink. From day 8 to 11 of storage, a significant color change occurred, transitioning from light pink to blue. After day 12 of storage, the color gradually changed from blue to brownish-yellow. A comparison of the five types of films (indicator film G / CCN / Mg) was made. 2+ The color change of / MAE-3 is the most obvious, especially during storage days 8 to 11, when the color difference of the indicator film is the greatest.

[0081] Figure 10 be is the indicator membrane G / CCN / Mg 2+ A scatter plot of canonical correlation (CCA) analysis of the RGB values ​​of / MAE-3 images with beef quality evaluation indicators (sensory evaluation, TVC, pH, TVB-N). The plot shows that sensory scores, TVC, pH, and TVB-N all exhibit a significant linear relationship with RGB values, with correlation coefficients of 0.99, 0.96, 0.95, and 0.97, respectively, indicating a significant correlation between the indicators.

[0082] Based on a comprehensive analysis of the sensory scores, TVC, TVB-N, pH, and other indicators measured above, this embodiment classifies the quality grade of braised beef stored from day 0 to 7 as "fresh," from day 8 to 11 as "sub-fresh," and from day 12 to 15 as "spoiled." The indicator film G / CCN / Mg... 2+ / MAE-3 not only exhibits different color changes at different stages ( Figure 10 a), and its RGB value is highly correlated with sensory ratings, TVC, TVB-N, pH, and other indicators. Figure 10 bcde), therefore G / CCN / Mg can be used. 2 + The color response of / MAE-3 is used to evaluate the freshness of braised beef.

[0083] Performance evaluation of different models Table 3 Classification Reports of Three Different Models

[0084] Note: In this embodiment, the freshness grade of braised meat is divided into "fresh", "less fresh" and "rotten". In the classification report, "0" represents fresh, "1" represents less fresh and "2" represents rotten.

[0085] In this work, this embodiment evaluates the performance of three models: VGG-16, ResNet-34, and ResNet-50. After training, the classification reports for the three models are shown in Table 3. The ResNet-34 and ResNet-50 models both achieved training accuracy exceeding 95%, while VGG-16 only achieved 93%. Furthermore, ResNet-34 and ResNet-50 both achieved higher recall and F1-scores than VGG-16. This is due to significant differences in network structure. ResNet (Residual Network) addresses the vanishing gradient problem in deep networks by introducing residual connections, allowing for deeper networks and improved feature extraction capabilities. Additionally, ResNet employs the Bottleneck structure, reducing computation through 1×1 convolutional layers and further improving efficiency. In contrast, VGG-16 constructs a deep network by stacking 3×3 convolutional layers and 2×2 pooling layers. The weighted average and macro average of the ResNet model are improved compared to VGG-16. This is because ResNet improves the robustness of the model through regularization techniques (such as Batch Normalization), enabling it to maintain stable performance when faced with imbalanced data.

[0086] The training loss and accuracy of the three models, Vgg-16, ResNet-34, and ResNet-50, are as follows: Figure 11 As shown, changes in training loss and accuracy during training dynamically reflect the model's convergence characteristics and overfitting risk. The "train_acc" and "val_acc" metrics represent the accuracy on the training and test sets, respectively. Similarly, the "train_loss" and "val_loss" metrics represent the loss values ​​on the training and test sets, respectively. As the number of epochs increases from 0 to 100, both "train_acc" and "val_acc" show an upward trend, while both "train_loss" and "val_loss" show a downward trend. After more than 20 epochs, the loss function and accuracy of the ResNet model on both the training and test sets tend to stabilize. However, the Vgg-16 model exhibits significant fluctuations in both loss curve and accuracy throughout the training process, failing to converge even after 100 epochs. These results indicate that the accuracy of the Vgg-16 model is lower than that of the ResNet model.

[0087] In practical applications, the quality of image acquisition for the indicator membrane cannot be completely consistent, resulting in dataset bias. To compare the accuracy of ResNet-34 and ResNet-50 models in a three-class classification task, Figure 12ab displays the confusion matrices of the two models. The confusion matrix reflects the relationship between the actual situation of the test dataset and the predicted values. A total of 75 samples were detected. In the ResNet-34 model, 3 "nearly fresh" samples were classified as "corrupted," and one "corrupted" sample was classified as "fresh." In the ResNet-50 confusion matrix, one "fresh" sample and one "nearly fresh" sample were incorrectly classified as "corrupted," and one "corrupted" sample was classified as "nearly fresh." The accuracy of ResNet-50 (96%) is higher than that of ResNet-34 (95%), indicating that the ResNet-50 model is superior to ResNet-34.

[0088] Figure 12 cd represents the ROC curves of the ResNet-50 and ResNet-34 algorithm models. The x-axis of the ROC curve represents the false positive rate, while the y-axis represents the true positive rate. The area under the curve (AUC) is defined as the region bounded by the ROC curve and the coordinate axes. The ROC curve and AUC value are effective indicators for evaluating the performance of diagnostic tests. Regarding the ROC curve, the closer to the top left corner, the higher the specificity and AUC value. The closer the AUC value is to 1, the better the diagnostic performance. In this case, the AUC values ​​for the three classification tasks of the ResNet-50 algorithm are 1.000, 0.995, and 0.993, respectively, while the AUC values ​​for the three classification tasks of the ResNet-34 algorithm are 1.000, 0.995, and 0.990, respectively. Although the difference between the two algorithms is not significant, the ResNet-50 algorithm is superior to the ResNet-34 algorithm. Therefore, the ResNet-50 model is used to monitor the freshness of braised beef.

[0089] An intelligent detection platform based on the ResNet-50 model was built. To quickly determine food freshness, after comparing three models—ResNet-50, ResNet-34, and Vgg-16—this embodiment selected ResNet-50 as the deep neural network model to develop a web-based detection system and achieve local deployment. Figure 13 As shown, the system adopts a three-layer architecture. The front end receives user-uploaded indicator film images via a responsive HTML template, while the back end integrates a PyTorch-driven ResNet-50 deep neural network to form a complete image classification solution. When using this detection platform, users can take photos in real-time via their mobile phones or select existing images from their albums and upload them. The system then connects directly to the local server via a local area network to quickly determine the freshness of the food. Notably, using this system, the freshness level can be determined within one second of uploading an indicator film image.

[0090] In this embodiment, a gelatin nanofiber indicator film based on mulberry anthocyanins was first prepared by a casting method. Studies have shown that when the amount of mulberry anthocyanins added is 3%, and it is combined with Mg... 2+ Cross-linking resulted in nanofiber membranes exhibiting excellent physical properties, pH sensitivity, and ammonia sensitivity. The indicator membranes were then applied to monitor the freshness of braised beef. The color changes of the indicator membranes showed a high degree of correlation with meat quality indicators (sensory quality, pH, TVC, TVB-N), classifying braised beef into three different grades: "fresh," "medium-fresh," and "spoiled." The obtained indicator membranes were used to train a deep learning model at an 8:2 ratio. The ResNet-50 model achieved an accuracy of 96%. A web-based detection system developed using this deep neural network model can quickly determine freshness within one second of image upload. This monitoring system is characterized by real-time operation, visibility, reliability, and automatic, non-destructive operation.

[0091] The following describes the low-temperature braised beef freshness detection system provided by the present invention. The low-temperature braised beef freshness detection system described below and the construction method of the low-temperature braised beef freshness detection model described above can be referred to and correspond to each other.

[0092] The present invention provides a system for detecting the freshness of low-temperature braised beef, which may include: The data receiving module is used to: receive images of the mulberry anthocyanin indicator membrane of the low-temperature braised beef to be tested, input by the user; The low-temperature braised beef freshness detection module is used to: obtain the freshness detection result of the low-temperature braised beef based on the mulberry anthocyanin indicator membrane image of the low-temperature braised beef to be tested, through the low-temperature braised beef freshness detection model obtained by any of the above-mentioned methods.

[0093] The present invention provides an electronic device that may include: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus. The processor can call logical instructions in the memory to execute the steps of the method for constructing a freshness detection model for low-temperature braised beef and / or the method for detecting the freshness of low-temperature braised beef.

[0094] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0095] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to perform the steps of the method for constructing a freshness detection model for low-temperature braised beef and / or the method for detecting the freshness of low-temperature braised beef.

[0096] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of constructing a method for detecting the freshness of low-temperature braised beef and / or a method for detecting the freshness of low-temperature braised beef.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0098] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a low-temperature halogenated beef freshness detection model, characterized by, The method comprises the following steps: acquiring pictures of mulberry cyanine indicator films corresponding to low-temperature marinated beef samples with different freshness; training a model to learn the relationship between low-temperature marinated beef with different freshness and corresponding mulberry cyanine indicator films by using a convolutional neural network according to the pictures of mulberry cyanine indicator films corresponding to low-temperature marinated beef samples with different freshness, and obtaining a low-temperature marinated beef freshness detection model.

2. The method of claim 1, wherein the low-temperature halogenated beef freshness detection model is constructed by using a machine learning algorithm. The method comprises the following steps: preparing a mulberry cyanine indicator film; placing the mulberry cyanine indicator film and the low-temperature marinated beef sample in the same environment; detecting the freshness of the low-temperature marinated beef sample by using a marinated beef quality indicator, and simultaneously obtaining a picture of the mulberry cyanine indicator film, thereby obtaining pictures of mulberry cyanine indicator films corresponding to low-temperature marinated beef samples with different freshness.

3. The method of claim 2, wherein the low-temperature halogenated beef freshness detection model is constructed by using a machine learning algorithm. With mulberry anthocyanin extract as an indicator, gelatin, carboxylated cellulose nanofiber as a film-forming matrix, a mulberry anthocyanin indicating film is prepared, wherein in the preparation of the mulberry anthocyanin indicating film, Mg 2+ Cross-linking enhances the mechanical properties of the mulberry anthocyanin indicating film.

4. The method of claim 3, wherein the low-temperature halogenated beef freshness detection model is constructed by using a machine learning algorithm. The addition amount of mulberry cyanine in the mulberry cyanine indicator film is 3%.

5. The method according to any one of claims 1 to 4, wherein The marinated beef quality indicator comprises any one or any combination of the following: pH, total bacterial count, and volatile basic nitrogen level; the freshness of the low-temperature marinated beef sample comprises any one or any combination of the following: fresh, sub-fresh, and spoiled.

6. The method of claim 1 to 4, wherein, The method comprises the following steps: training a model to learn the relationship between low-temperature marinated beef with different freshness and corresponding mulberry cyanine indicator films by using a VGG-16 convolutional neural network according to the pictures of mulberry cyanine indicator films corresponding to low-temperature marinated beef samples with different freshness, and obtaining a first low-temperature marinated beef freshness detection model; training a model to learn the relationship between low-temperature marinated beef with different freshness and corresponding mulberry cyanine indicator films by using a ResNet-34 convolutional neural network according to the pictures of mulberry cyanine indicator films corresponding to low-temperature marinated beef samples with different freshness, and obtaining a second low-temperature marinated beef freshness detection model; training a model to learn the relationship between low-temperature marinated beef with different freshness and corresponding mulberry cyanine indicator films by using a ResNet-50 convolutional neural network according to the pictures of mulberry cyanine indicator films corresponding to low-temperature marinated beef samples with different freshness, and obtaining a third low-temperature marinated beef freshness detection model; comparing the training loss and accuracy of the first low-temperature marinated beef freshness detection model, the second low-temperature marinated beef freshness detection model, and the third low-temperature marinated beef freshness detection model during the model training process, and obtaining a final low-temperature marinated beef freshness detection model.

7. A low-temperature marinated beef freshness detection method, comprising the following steps: receiving a mulberry cyanine indicator film picture of a low-temperature marinated beef to be detected input by a user from at least one terminal; inputting the mulberry cyanine indicator film picture to be detected into a low-temperature marinated beef freshness detection model obtained by the method for constructing a low-temperature marinated beef freshness detection model according to any one of claims 1-6, and obtaining a freshness detection result of the low-temperature marinated beef to be detected. The freshness detection result of the low-temperature marinated beef to be tested is sent to at least one terminal.

8. A system for constructing a freshness detection model for low-temperature cured beef, characterized by comprising: The application relates to a low-temperature marinated beef freshness detection method and a low-temperature marinated beef freshness detection model. The data receiving module is used for receiving a mulberry cyanine indicator film picture of low-temperature marinated beef to be tested input by a user. The low-temperature marinated beef freshness detection module is used for obtaining a freshness detection result of the low-temperature marinated beef to be tested according to the mulberry cyanine indicator film picture of the low-temperature marinated beef to be tested and the low-temperature marinated beef freshness detection model obtained by the construction method of the low-temperature marinated beef freshness detection model according to any one of claims 1-6.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the construction method of the low-temperature marinated beef freshness detection model according to any one of claims 1-6 and / or the low-temperature marinated beef freshness detection method according to claim 7. 10.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the construction method of the low-temperature marinated beef freshness detection model according to any one of claims 1-6 and / or the low-temperature marinated beef freshness detection method according to claim 7.

Images