System and method for rapidly detecting freshness of raw meat by using deep learning and colorimetric fluorescent array

Abstract

The invention belongs to the technical field of fresh food detection, and particularly relates to a system and a method for rapidly detecting freshness of raw meat by using deep learning and a colorimetric fluorescent array, and the change of the freshness of chilled meat is intuitively reflected in a colorimetric / fluorescent dual mode by using the specific response of alizarin and FITC to biogenic amine. The structural change of the two pigments in different acid-base environments causes the change of color and fluorescence, so that the indication label sensitively responds to amine substances generated by the decay of chilled fresh meat. By optimizing pigment extraction and concentration screening, the detection sensitivity and accuracy are improved. And in combination with a deep learning model, the limitation of distinguishing color differences by human eyes is overcome, a large amount of image data can be quickly processed, and the freshness grade is accurately judged. And the food pollution risk is reduced by using natural pigments. According to the method, the defects of the existing chilled meat freshness detection technology are effectively overcome, rapid, accurate and nondestructive detection is realized, food safety is guaranteed, food waste is reduced, and the method has remarkable application value in the field of food detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of fresh food detection, and particularly relates to a system and method for rapidly detecting the freshness of raw meat by using deep learning and a colorimetric fluorescence array. Background Art

[0002] There are different freshness levels of fresh meat. Moreover, most chilled fresh meat is prone to spoilage due to microbial contamination and enzymatic reactions during storage, transportation, and sales. At present, the methods for detecting the freshness of chilled fresh meat are mainly divided into destructive detection and non-destructive detection. Destructive detection methods, such as traditional physical and chemical analysis methods, although having high accuracy, require destructive treatment of samples, are complex and time-consuming to operate, and are not suitable for on-site rapid detection. Non-destructive detection methods, such as spectroscopy analysis, electronic nose and other technologies, although capable of realizing non-destructive detection, have problems such as expensive equipment and the need for professional personnel to operate. At the same time, as an emerging detection means, the intelligent indicator label for chilled fresh meat can, to a certain extent, intuitively reflect the freshness of meat, but the existing indicator labels have problems such as low sensitivity, narrow detection range, and slow response speed, and it is difficult to meet the actual needs.

[0003] There are many deficiencies in the existing technologies for detecting the freshness of chilled fresh meat. Traditional detection methods cannot quickly and accurately detect the freshness without destroying the samples, and cannot meet the requirements for rapid detection of chilled fresh meat in the food supply chain. The existing intelligent indicator labels need to be improved in terms of detection sensitivity and accuracy, are not sensitive enough to biogenic amines at different concentrations, and cannot accurately distinguish. The cold indicator labels can only be detected through a single mode (such as colorimetry or fluorescence), and are easily interfered by environmental factors, resulting in inaccurate detection results. In addition, the indicator labels using chemically synthesized pigments in the existing technologies may cause pollution to food and affect food safety. Summary of the Invention

[0004] The purpose of the present invention is to provide a system and method for rapidly detecting the freshness of raw meat by using deep learning and a colorimetric fluorescence array, so as to solve the technical problems of inaccurate traditional detection results and easy pollution to food.

[0005] To solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0006] In some embodiments of the present application, there is provided a system and method for rapidly detecting the freshness of raw meat by using deep learning and a colorimetric fluorescence array, including the following steps:

[0007] 1) Prepare a pigment screening label for CFA;

[0008] 2) Measure the pigment screening label for CFA;

[0009] 3) The freshness of the pigment screening label of the CFA after the measurement in step 2 is measured in raw meat.

[0010] In some embodiments of the present application, step 1 includes:

[0011] 1.1) Pigment extraction: The passion fruit peel and pitaya peel are dried in an oven at 55 °C for 5 hours, crushed into powder, and then added with 60% ethanol aqueous solution with a solid-liquid ratio of 1:10, and stirred at 50 °C for 2 hours. The ethanol in the filtrate is removed by a rotary evaporator under dark conditions at 35 °C, and finally the solution is vacuum freeze-dried to obtain passion fruit peel anthocyanin extract powder and pitaya peel anthocyanin extract, which are respectively filled into brown sample bottles and stored in an environment at 4 °C.

[0012] 1.2) Pigment screening: Alizarin, passion fruit peel anthocyanin, and pitaya peel anthocyanin are respectively prepared into solutions with a concentration of 5 mg / mL. 200 μL of each solution is taken into a 96-well plate. After taking the initial photo, 10, 20, 50, 100, 200, and 500 mg / mL ammonia water solutions are respectively added, and photos are taken after a 10-minute response. The RGB values of the solution before and after the response are extracted using PS software, and the ED value of the solution amine response is calculated according to the formula. Pigments for subsequent experiments are screened according to the ED value.

[0013] 1.3) Concentration screening of alizarin and FITC: The alizarin solution is serially diluted from 20 mg / mL to 0.16 mg / mL, and the FITC solution is diluted from 0.8 mg / mL to 0.1 mg / mL. Using the checkerboard mixing method, 36 different alizarin / FITC concentration combinations are generated, and the solutions are loaded onto PVDF membranes (diameter 5 mm) by capillary action. The membranes are exposed to an ammonia gas environment for 25 minutes, and photos are taken under natural light and ultraviolet light conditions. The RGB values are extracted to calculate the ED value, and nine pigment combinations are selected according to the comprehensive ED value for the colorimetric / fluorescent array.

[0014] 1.4) The selected alizarin and FITC are accurately loaded onto the PVDF membrane as a pigment solution according to the selected concentration combination using a capillary to prepare a colorimetric / fluorescent dual-mode freshness indicator label.

[0015] In some embodiments of the present application, the measurement of the pigment screening label of the CFA in step 2 includes:

[0016] 2.1) Microstructure and EDS measurement: The surfaces of each unit A1-A9 of the indicator label are sputter-coated with gold, fixed on the sample stage, the voltage is set to 10 kV, the scanning speed is set to medium, the surface morphology is photographed and area scanning is performed to generate an elemental distribution map.

[0017] 2.2) Water contact angle measurement: Fix the sample on the sample stage to ensure the surface is horizontal. Use a 2 μL deionized water droplet for measurement. Set the droplet injection speed to 0.5 - 2 μL / s, record the change in contact angle within 15 s, and repeat the measurement 3 times at different positions.

[0018] 2.3) Performance test of colorimetric / fluorescent dual-mode freshness indicator labels, recycling characteristics: Cover the indicator label on a 24-well plate containing an ammonia aqueous solution with a concentration of 2500 ppm. After 20 minutes, place it in an acidic environment to reset the color. Repeat the experiment, record the color change, use PS to extract the RGB values and calculate the ED value. The formula is, where R0, G0, and B0 are the initial gray values of each unit in the array, and Rn, Gn, and Bn are the gray values of each unit of the indicator label after exposure to amine gas.

[0019] Amine response behavior: Fix the indicator label in the test chamber and expose it to ammonia, dimethylamine, and trimethylamine atmospheres with different concentrations (10, 50, 100, 200, 250, 2500 ppm). Take pictures at different time points (10, 20, 30, 120, 300, 600 s) and use PS to record the RGB values.

[0020] Discrimination of different types of amines: Place the indicator label in different amine atmospheres with a concentration of 1000 ppm respectively for 10 min, take pictures to record the RGB values, and calculate the ED value.

[0021] In some embodiments of the present application, the freshness determination of the pigment screening label of CFA in step 3 for raw meat includes:

[0022] 3.1) Determination of TVB-N: After pretreating the pork sample, add HCl titration solution and calculate the content of volatile amines according to the titration result;

[0023] 3.2) Determination of pH: After pretreating the pork sample from which fat and connective tissue have been removed, obtain the supernatant and measure the pH value of the supernatant with a pH meter;

[0024] 3.3) Determination of TBARS: Based on the reaction of thiobarbituric acid with lipid oxidation products to generate a pink compound, measure the absorbance by colorimetry and calculate the TBARS value;

[0025] 3.4) Determination of TVC: Evaluate the degree of contamination of the pork sample by obtaining the number of viable bacteria in the pork sample.

[0026] In some embodiments of the present application, the pretreatment in step 3.1 includes: treating the pork sample in a sterile environment at room temperature, homogenizing the minced meat sample in distilled water; after 30 minutes, filtering the mixture to obtain a liquid sample; adding boric acid absorption solution and pH indicator into the inner cavity of a Conway dish; sequentially adding saturated potassium carbonate solution and sample solution into the outer chamber; sealing the dish and culturing it at 37°C for 2 hours, then titrating the solution with 0.1 mol / L HCl, and calculating the content of volatile amines according to the titration result.

[0027] In some embodiments of the present application, the calculation of the TBARS value in step 3.3 includes: taking 10 g of pork sample, removing visible fat and connective tissue, chopping or homogenizing the sample; extracting lipids with a chloroform-methanol mixture; centrifuging and separating, collecting the lower chloroform phase, evaporating the solvent to obtain a lipid extract; mixing the lipid extract with TBA reagent, heating at 90 - 100°C for 30 - 60 min; centrifuging after cooling, taking the supernatant; measuring the absorbance at a wavelength of 530 nm using a spectrophotometer, and calculating the TBARS value according to the malondialdehyde standard curve.

[0028] In some embodiments of the present application, obtaining the number of viable bacteria in the pork sample in step 3.4 includes:

[0029] Taking an appropriate amount of pork sample and putting it into a sterile homogenization bag; adding sterile physiological saline, homogenizing with a homogenizer for 1 - 2 minutes to prepare a 1:10 sample dilution, taking 1 mL of the 1:10 dilution, adding it into sterile physiological saline and mixing evenly to prepare a 1:100 dilution; repeating the above steps to sequentially prepare 10-3, 10-4, 10-5 gradient dilutions; respectively taking 1 mL of each gradient dilution and injecting it into a sterile petri dish; pouring about 15 - 20 mL of sterilized and cooled plate count agar at 45 - 50°C, shaking well, waiting for the agar to solidify; inverting the petri dish and placing it in a constant temperature incubator, culturing at 37°C for 48 h; after the culturing is completed, taking out the petri dish, selecting the petri dish with the number of colonies between 30 - 300, and counting the number of colonies with a colony counter.

[0030] In some embodiments of the present application, a system for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescent arrays is disclosed. A deep learning model is constructed through the training processes of ResNet-152 and Vision Transformer models and the construction of a dataset, and is applied to systems in smartphones and computer terminals. Images of the indicator label in the environment of chilled fresh pork with different freshness levels are collected to construct a dataset containing a large amount of image data. The ResNet-152 model and Vision Transformer model in the convolutional neural network (CNN) are used for training, and the model parameters are optimized through 10-fold cross-validation. A mobile APP named porkcheck is developed using the trained model to analyze the array images, realizing rapid and accurate judgment of the freshness of chilled fresh pork.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention utilizes the specific responses of alizarin and FITC to biogenic amines to visually reflect the change in the freshness of chilled meat through colorimetric / fluorescent dual modes. The structural changes of the two pigments in different acid-base environments lead to color and fluorescence changes, enabling the indicator label to sensitively respond to amine substances produced by the spoilage of chilled meat. By optimizing pigment extraction and concentration screening, the detection sensitivity and accuracy are improved. Combining with a deep learning model, the limitation of the human eye in distinguishing color differences is overcome, enabling rapid processing of a large amount of image data and accurate judgment of the freshness level. Moreover, the use of natural pigments reduces the risk of food contamination. The present invention effectively solves the deficiencies of the existing chilled meat freshness detection technology, realizes rapid, accurate, and non-destructive detection, ensures food safety, reduces food waste, and has significant application value in the field of food detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of illustrating the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0033] Figure 1 It is a schematic diagram of pigment screening provided by an embodiment of the present invention;

[0034] Figure 2 It is a schematic diagram of the process of alizarin and FITC for preparing CFA provided by an embodiment of the present invention;

[0035] Figure 3 It is a schematic diagram of the response picture and ultraviolet-visible spectrum of alizarin at different pH values provided by an embodiment of the present invention;

[0036] Figure 4 It is a schematic diagram of the FTIR spectrum of alizarin provided by an embodiment of the present invention;

[0037] Figure 5 Schematic diagram of XPS of alizarin before and after amine response provided by an embodiment of the present invention;

[0038] Figure 6 Schematic diagram of the ultraviolet-visible spectrum of FITC provided by an embodiment of the present invention;

[0039] Figure 7 Schematic diagram of the fluorescence spectrum of FITC provided by an embodiment of the present invention;

[0040] Figure 8 Schematic diagram of the preparation process of CFA and the training of the model provided by an embodiment of the present invention;

[0041] Figure 9 Schematic diagram of the architecture of the ResNet-152 model and the actual interface of the developed mobile phone App provided by an embodiment of the present invention. Detailed implementation manners

[0042] The following further describes in detail the specific implementation manners of the present invention in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0043] To better understand the purpose, structure and function of the present invention, the following further describes the present invention in detail in conjunction with the accompanying drawings.

[0044] Embodiment 1:

[0045] The pigment screening for preparing CFA includes the following steps:

[0046] (1) Accurately prepare pigment solutions and ammonia water solutions, react according to the specified volume and time, ensure consistent lighting conditions when taking pictures, and accurately extract RGB values and calculate ED values using PS software.

[0047] (2) Concentration screening of alizarin and FITC: Accurately prepare alizarin and FITC solutions with different concentrations, and adopt a checkerboard mixing method to ensure the accuracy of concentration combinations. During the exposure to the ammonia environment and the process of taking pictures, strictly control the environmental conditions and time to ensure the reliability of the data.

[0048] (3) Preparation of CFA: Use a capillary to accurately load the pigment solution onto the PVDF membrane, ensure uniform loading, and properly store the prepared labels before use to avoid contamination and light exposure.

[0049] Embodiment 2:

[0050] The characterization of CFA includes the following contents:

[0051] (1) Microstructure and EDS measurement: When sputtering gold on the surface of the indicator label, ensure that the sputtering is uniform. During the operation of the scanning electron microscope, strictly set the voltage and scanning speed according to the equipment operation procedures to accurately obtain the surface morphology and element distribution information.

[0052] (2) Water contact angle measurement: Ensure the surface is horizontal when fixing the sample. When measuring with deionized water droplets, control the injection speed and accurately record the change in the contact angle within the specified time. Repeat the measurement 3 times for each sample and take the average value.

[0053] (3) Recycling characteristics of CFA: The reusability of the colorimetric / fluorescent dual-mode indicator label was studied. The colorimetric / fluorescent dual-mode indicator label was covered on a 24-well plate containing an ammonia aqueous solution with a concentration of 2500 ppm. After 20 minutes, it was placed in an acidic environment to reset the color of the indicator label. When the color of the film returned to the color before reacting with ammonia, repeat the above experimental steps. Use a camera to record the color change of the indicator label, and use PS to extract the RGB values of the indicator label and calculate the ED value. The ED value of the indicator label is calculated according to the following formula:

[0054]

[0055] where, R 0 , G 0 and B 0 are the initial gray values of each unit in the array, and R n , G n and B n are the gray values of each unit of the indicator label after exposure to amine gas.

[0056] (4) Response of CFA at different concentrations and different times in an ammonia atmosphere

[0057] Keep the environmental temperature at room temperature. Fix the colorimetric / fluorescent dual-mode freshness indicator label in the test chamber to ensure that the surface of the indicator label is exposed to the ammonia atmosphere. Expose the sample to ammonia atmospheres with different concentrations (10, 50, 100, 200, 250, 2500 ppm), and use a camera to take photos of the indicator label at different time points (10, 20, 30, 120, 300, 600 s). Use PS to record the RGB values at each concentration and time point.

[0058] (5) Response of CFA at different concentrations and different times in a dimethylamine atmosphere

[0059] Keep the environmental temperature at room temperature. Fix the colorimetric / fluorescent dual-mode freshness indicator label in the test chamber, ensuring that the surface of the indicator label is exposed to the ammonia atmosphere. Expose the sample to dimethylamine atmospheres with different concentrations (10, 50, 100, 200, 250, 2500 ppm), and use a camera to take photos of the indicator label at different time points (10, 20, 30, 120, 300, 600 s). Use PS to record the RGB values at each concentration and time point.

[0060] (6) Response of CFA at different concentrations and different times in trimethylamine atmosphere

[0061] Keep the environmental temperature at room temperature. Fix the colorimetric / fluorescent dual-mode freshness indicator label in the test chamber, ensuring that the surface of the indicator label is exposed to the ammonia atmosphere. Expose the sample to trimethylamine atmospheres with different concentrations (10, 50, 100, 200, 250, 2500 ppm), and use a camera to take photos of the indicator label at different time points (10, 20, 30, 120, 300, 600 s). Use PS to record the RGB values at each concentration and time point.

[0062] Discrimination of different types of amines by CFA

[0063] Place the indicator label in different amine atmospheres with a concentration of 1000 ppm for 10 min, then use a camera to take photos, use PS to record the RGB values of the indicator label, and calculate the ED value of the indicator label using the formula.

[0064] Example 3:

[0065] The application of CFA in the freshness detection of chilled fresh pork includes the following:

[0066] (1) Determination of TVB-N: The pork sample is processed in a sterile environment at room temperature. Homogenize 10 g of minced meat sample in 90 mL of distilled water. After 30 minutes, filter the mixture to obtain a liquid sample. Then, add 3 mL of boric acid absorbent solution (20 g / L) and 50 μL of pH indicator (the ratio of methyl red to bromocresol green is 1:5, v / v) to the inner cavity of the Conway dish. Add 3 mL of saturated potassium carbonate solution and 1 mL of sample solution to the outer chamber in sequence. Seal the dish and incubate it at 37 °C for 2 hours to allow the volatile amines to be absorbed by the boric acid absorbent. Then titrate the solution with 0.1 mol / L HCl, and calculate the content of volatile amines according to the titration result. The measurement experiment is repeated six times. The TVB-N value is determined by the following formula:

[0067]

[0068] Where: X is the TVB-N content in the sample (mg / 100 g); V 1is the volume (mL) of 0.1 mol / L HCl in the sample; V 2 is the volume (mL) of 0.1 mol / L hydrochloric acid in the control group; C is the concentration of hydrochloric acid (mol / L); 14 is the molecular weight of nitrogen; m is the weight (g) of the meat sample.

[0069] (2) Determination of pH: Take an appropriate amount of pork sample, remove fat and connective tissue, and cut it into small pieces. Weigh 10 g of the sample, add 90 mL of distilled water, and homogenize it with a homogenizer or mortar. Filter the homogenate and reserve the supernatant. Turn on the pH meter and preheat it for 15 - 30 minutes. Calibrate the pH meter using standard buffer solutions (pH 4.01, 7.00, 10.01). Rinse the pH electrode with distilled water and dry it with filter paper. Immerse the electrode into the sample supernatant, stir gently, and record the pH value after the reading stabilizes. Each sample is measured 3 times.

[0070] (3) Determination of TBARS: The determination of thiobarbituric acid reactive substances (TBARS) is a common method for evaluating the degree of lipid oxidation in meats such as pork during storage. The higher the TBARS value, the more severe the lipid oxidation, which may lead to poor meat quality. The TBARS determination is based on the reaction of thiobarbituric acid (TBA) with lipid oxidation products (such as malondialdehyde) to form a pink compound, and the absorbance is measured by colorimetry to calculate the TBARS value. Take 10 g of pork sample and remove visible fat and connective tissue. Chop or homogenize the sample. Extract lipids using a chloroform - methanol mixture (2:1, v / v). Centrifuge and collect the lower chloroform phase, evaporate the solvent to obtain a lipid extract. Mix the lipid extract with TBA reagent and heat (90 - 100 °C, 30 - 60 min). After cooling, centrifuge and take the supernatant. Measure the absorbance at a wavelength of 530 nm using a spectrophotometer. Calculate the TBARS value based on the malondialdehyde standard curve.

[0071] (4)Determination of TVC: Total Viable Count (TVC) is an important indicator for evaluating the degree of microbial contamination of pork during storage. By measuring TVC, the hygienic condition of pork and the microbial changes during storage can be understood. Take 10 g of pork sample and put it into a sterile homogenization bag. Add 90 mL of sterile normal saline and homogenize with a homogenizer for 1 - 2 minutes to prepare a 1:10 sample dilution. Take 1 mL of the 1:10 dilution and add it to 9 mL of sterile normal saline, mix well to prepare a 1:100 dilution. Repeat the above steps to successively prepare 10-3, 10-4, and 10-5 gradient dilutions. Take 1 mL of each gradient dilution and inject it into a sterile petri dish. Pour about 15 - 20 mL of sterilized and cooled plate count agar at 45 - 50 °C, shake gently, and wait for the agar to solidify. Invert the petri dish and place it in a constant temperature incubator and incubate at 37 °C for 48 h. After the incubation, take out the petri dish and count the number of colonies with a colony counter. Select the petri dishes with the number of colonies between 30 - 300 for counting. Calculate TVC according to the following formula:

[0072]

[0073] Where: N: Number of colonies in the petri dish, D: Dilution factor, W: Sample weight (g), and the result is expressed in CFU / g.

[0074] Example 4:

[0075] The construction of the deep learning model and the development of the App include the following:

[0076] (1) Collection of the dataset for model training and validation.

[0077] The CFA is pasted on the widely used transparent polyvinyl chloride (PVC) meat packaging film, with its position facing inward and not in direct contact with the meat sample. The flexibility and transparency of the CFA enable seamless integration with existing polystyrene and polyvinyl chloride packaging materials. The packaged meat samples are stored at 25 °C, and high-quality images of the CFA are captured at different time intervals using a high-definition camera. These images are classified into three categories according to the TVB-N content of the pork, which is measured using the Conway dish method, an international standard for determining meat freshness. Here, the TVB-N test uses independent meat packages prepared under the same conditions to avoid any inference with image acquisition. The meat is classified according to the following TVB-N critical values: ≤15 mg / 100 g is considered fresh, 15 - 20 mg / 100 g indicates that the meat is edible but not very fresh, and ≥20 mg / 100 g indicates that the meat has spoiled and is inedible. Then, the captured images are classified into three freshness categories: fresh, not very fresh, and spoiled. Since the TVB-N thresholds for different types and sources of meat may vary, to generalize this classification method to other meats, it is necessary to thoroughly study the specific TVB-N values associated with each type of meat for accurate classification.

[0078] According to these classification principles, two datasets were created under natural light and ultraviolet illumination conditions to evaluate freshness. Given that a larger dataset has been the main driving force for predicting meat freshness using DL, each dataset consists of 4,085 images, representing 74 pork groups and their corresponding CFAs. This scale is the largest dataset in this field to date. These images were captured from different angles and at different times to ensure data diversity.

[0079] (2) Neural network selection

[0080] Neural networks are widely used in computer vision tasks, and the ability to accurately represent and evaluate freshness is crucial in applications such as food quality monitoring and environmental sensing. In this study, we demonstrated the effectiveness of the CFA in freshness evaluation using two state-of-the-art DL architectures: CNN and Vision Transformer. For the CNN, we selected the deep network ResNet-152, which consists of 152 layers. Its architecture includes Res Blocks, fully connected layers (FC), Max Pool, and Average Pool. Res Blocks mainly include convolutional layers (Conv) and short connection layers (SC). ResNet-152 is well-known for its strong performance in different computer vision tasks. The main architecture of ResNet-152 is shown in the figure.

[0081] In terms of vision transformers, we selected CMT-B, which has components such as CMT Stem, Patch Embedding Layers, CMT Blocks, FC, and Avg Pool. CMT Stem and Patch Embedding Layers are mainly composed of Conv, while CMT Blocks integrate Multi-head Self-attention, Conv, and SC. As one of the state-of-the-art vision transformer models, CMT-B is the backbone for multiple computer vision tasks. The main architecture of CMT-B is shown in the figure.

[0082] (3) Evaluation Metrics

[0083] We used accuracy (Acc), precision (Pr), and recall (Re) as evaluation metrics, which are the basic elements for evaluating model performance. Acc represents the prediction accuracy of the most likely class for each input sample55. Pr for a given class is the ratio of correctly predicted positive observations to the total number of predicted positive observations in the same class. Re for a certain class is the ratio of correctly predicted positive results to the total number of actual positive results in that class. The final Pr and Re are calculated based on the average of Pr and Re across all classes.

[0084] (4) Implementation Details

[0085] To improve the model performance through transfer learning, we first pre-trained the ResNet-152 and CMT-B models on the ImageNet-1k dataset and then performed fine-tuning. The fine-tuning experiments were conducted using the PyTorch software library (version 1.10.1) on the Ubuntu 22.04 operating system. All experiments were carried out on a server equipped with dual Intel Xeon E5-2678v3 CPUs and four Nvidia GeForce RTX 2080Ti GPUs.

[0086] Example 5:

[0087] The screening of pigments is as Figure 1As shown in the figure. Among them, as the ammonia concentration gradually increases from 10 mg / mL to 500 mg / mL for the pitaya peel pigment, its ED value shows a gradually increasing trend. At low ammonia concentrations, the ED value is relatively low and the growth is relatively slow; when the ammonia concentration reaches 100 mg / mL and above, the growth rate of the ED value increases. This indicates that the pitaya peel pigment has a certain response to the change in ammonia concentration, and its response is more obvious at higher ammonia concentrations. The passion fruit peel pigment also generally shows an upward trend in the ED value with the increase in ammonia concentration. Compared with the pitaya peel pigment, at the same ammonia concentration, the ED value of the passion fruit peel pigment is generally lower. In the low ammonia concentration stage, its ED value increases relatively smoothly; when the ammonia concentration reaches 200 - 500 mg / mL, the growth rate slightly accelerates. In addition, the ED value of alizarin at each ammonia concentration is higher than that of the pitaya peel pigment and the passion fruit peel pigment. And with the increase in ammonia concentration, its ED value shows an obvious growth trend. Especially when the ammonia concentration reaches above 200 mg / mL, the ED value increases rapidly and reaches the highest at an ammonia concentration of 500 mg / mL. The change range of the ED value of alizarin at different ammonia concentrations is relatively large, indicating that it is more sensitive to the change in ammonia concentration and can more obviously reflect the change in ammonia concentration. In contrast, the sensitivities of the pitaya peel pigment and the passion fruit peel pigment are relatively low. The ED value of alizarin still has a large growth space at high ammonia concentration (500 mg / mL), which means it has an advantage in the detection of high ammonia concentration. While the ED values of the pitaya peel pigment and the passion fruit peel pigment are relatively low and the growth amplitude is limited at high ammonia concentration, and the detection range may be relatively narrow. In summary, considering the responses of the three pigments at different ammonia concentrations, alizarin is selected for the preparation of the subsequent freshness indicating label.

[0088] Example 6:

[0089] As Figure 2 shown, alizarin and FITC are mixed to screen out a suitable pigment combination. By mixing them in different ratios, the influence of their interaction on the performance is explored. The mixed pigment is loaded onto a polyvinylidene fluoride (PVDF) membrane to fix the pigment, and then the PVDF membrane loaded with the pigment is used to obtain the RGB value of each unit through PS, and the ED is calculated to evaluate the performance of different combinations of alizarin and FITC concentrations.

[0090] Example 7:

[0091] The ultraviolet-visible spectrum of alizarin is as Figure 3As shown. Figure A shows the color change of alizarin solution at different pH values. It is yellow in acidic conditions. As the pH increases, the color changes from yellow to orange, then to red, and is purple in alkaline conditions. The color of the solution is determined by the absorption and reflection of light. In acidic conditions, alizarin mainly absorbs short-wavelength light and reflects yellow light. As the pH increases, the absorption peak redshifts, the absorption of long-wavelength light increases, and the color of the reflected light changes. As can be seen from Figure B, in the wavelength range of 300 - 700 nm, as the pH value increases from 2 to 13, the absorption peak of alizarin gradually shifts towards the long-wavelength direction, that is, redshifts. Under acidic conditions (pH = 2 - 5), the absorption peak is located in the relatively short-wavelength region; under alkaline conditions (pH = 10 - 13), the absorption peak shifts to the longer-wavelength region. This is because alizarin contains groups such as phenolic hydroxyl groups. Under alkaline conditions, the phenolic hydroxyl group dissociates into a phenoxide anion, increasing the molecular conjugation degree, reducing the energy of electron transition, and causing the absorption peak to redshift. In addition, the absorption intensity changes with the pH value. The absorption intensity is relatively low in acidic conditions and gradually increases as the pH increases, reaching a relatively high value under alkaline conditions (pH = 12 - 13). This may be due to the formation of a more stable conjugated structure of alizarin under alkaline conditions, enhancing the light absorption ability, or it may also be related to the change in the molecular aggregation state. The CIE coordinate diagram is used to accurately describe color. In theory, the spectral data of alizarin at different pH values can be converted into CIE coordinates to determine its position in the color space. Combining with the absorption spectrum of alizarin, its colors at different pH values correspond to different regions in the CIE diagram. The yellow color in acidic conditions corresponds to a specific region in the diagram, and the purple color in alkaline conditions corresponds to another region. It can also be seen from the figure that information such as the color temperature corresponding to different color coordinates is helpful for better applying alizarin in freshness detection applications. Finally, the change in the position and intensity of the absorption peak in the ultraviolet-visible spectrum determines the absorption of alizarin to different wavelengths of light, thereby affecting its color, which is reflected as the change of color coordinates in the CIE coordinate diagram. By comparing the ultraviolet-visible spectra and CIE coordinates of alizarin at different pH values, the relationship between its optical properties and color change can be deeply understood, providing theoretical support for the application of alizarin in the detection of alkaline amine gases.

[0092] Example 8:

[0093] The Fourier transform infrared spectra of alizarin amine before and after response are as Figure 4As shown. Before the alizarin amine reaction, there is an absorption peak at 3305 cm-1, which corresponds to the stretching vibration of the hydroxyl group (OH) in the alizarin molecule. The presence of this peak indicates that the alizarin molecule contains a hydroxyl structure. After the amine reaction, in the infrared spectrum of alizarin, the shape of the absorption peak in this region changes partially, and the intensity of the absorption peak weakens. This may be because after the reaction of amine with alizarin, the hydroxyl group participates in the reaction, the number decreases or its chemical environment changes, affecting the stretching vibration characteristics of the hydroxyl group. Before the alizarin response, the absorption peak at 2960 cm-1 corresponds to the stretching vibration of carbon-hydrogen (CH) in the alizarin molecule, reflecting the presence of saturated or unsaturated carbon-hydrogen groups in the molecule. After the amine reaction, there may be slight changes in the absorption peak in this region, which may be because the reaction of amine with alizarin changes the carbon-hydrogen skeleton structure or electron cloud distribution of the molecule, thereby affecting the vibration characteristics of the carbon-hydrogen bond. Before the amine reaction, the absorption peak in the region of about 1620 cm-1 corresponds to the stretching vibration of the carbonyl group (C=O) in the alizarin molecule, indicating the presence of a carbonyl structure in the molecule. After the amine reaction, the intensity of the absorption peak in this region changes slightly. It may be that the amine reacts with the carbonyl group, and the vibration characteristics of the carbonyl group will change, resulting in the shift or intensity change of the absorption peak. Before the amine reaction, the absorption peak at 1532 cm-1 corresponds to the stretching vibration of the carbon-carbon double bond (C=C) in the alizarin molecule, indicating that the molecule contains a carbon-carbon double bond structure. After the amine reaction, there may be changes in the absorption peak in this region.

[0094] Example 9:

[0095] In the C1s spectrum of alizarin, the binding energy of the C-C bond is 284.8 eV, and the binding energy of the C-O bond is 286.2 eV, corresponding to the carbon in the C-C single bond and the C-O bond in the molecule, respectively. After the reaction of alizarin with ammonia, the binding energy of the C-C bond remains 284.8 eV without obvious change, indicating that the effect of ammonia has little influence on the electron cloud density around the C-C bond; the binding energy of the C-O bond remains at 286.2 eV without significant change, suggesting that the chemical environment of the C-O bond is relatively stable. In the O1s spectrum of alizarin, there are peaks of C-O (533.3 eV) and N-O (531.6 eV), corresponding to different oxygen-containing compound structures in the molecule. After the reaction of alizarin with ammonia, the binding energy of the C-O bond decreases to 532.5 eV, and the binding energy of the N-O bond decreases to 531.0 eV, indicating that after the reaction of ammonia with alizarin, the electron cloud distribution of the oxygen-containing compound structure has changed, resulting in a decrease in the binding energy of O1s, and a new oxygen-containing compound structure may be formed or the chemical environment of the original structure has been changed. In the N1s spectrum of alizarin, there are peaks of N-C- (399.6 eV) and N-H (400.4 eV), corresponding to specific nitrogen-containing structures in the molecule. After the reaction of alizarin with ammonia, the binding energy of N-C- decreases to 399.1 eV, and the binding energy of N-H remains at 400.4 eV, and the intensity and shape of the peaks also change, indicating that the reaction of ammonia with alizarin has an impact on the nitrogen-containing structure, and new nitrogen-containing chemical bonds may be formed or the electron cloud distribution of the original nitrogen-containing structure has been changed. Generally speaking, by comparing the XPS spectra of alizarin and alizarin after the reaction with ammonia, it can be seen that ammonia reacts with alizarin, resulting in changes in the chemical environment and chemical bond states of carbon, oxygen, and nitrogen elements in the molecule.

[0096] Example 10:

[0097] Such as Figure 6As shown, when comparing the control group with the pH = 13 group: regarding the absorption intensities of the two curves, it is found that when pH = 13, the absorption intensity of FITC at the absorption peak is different from that of the control group. The absorption intensity increases at some wavelengths while decreases at others. This further indicates that the alkaline environment has an impact on the molecular structure and electronic state of FITC. It may be that a new conjugated structure is formed or the original conjugated structure is destroyed under alkaline conditions, causing the absorption ability of the molecule to different wavelengths of light to change, thus affecting the absorption intensity. Overall shape change of the spectrum: The overall shapes of the spectra of the control group and the pH = 13 group are significantly different. In addition to the changes in the position and intensity of the absorption peak, the trend of the entire spectral curve is also different. This shows that the pH value has a significant impact on the ultraviolet-visible absorption characteristics of FITC. The molecular structure of FITC has different stabilities and electronic characteristics in different acid-base environments, which in turn affects its absorption of ultraviolet-visible light. The change in the environmental pH value can be reflected by detecting the change in its ultraviolet-visible absorption spectrum. In summary, through the analysis of the ultraviolet-visible spectra of FITC under different pH conditions, the relationship between its molecular structure and light absorption characteristics, as well as the influence of the pH value on its light absorption properties, can be explored.

[0098] Example 11:

[0099] As Figure 7 shows the influence of pH on the fluorescence intensity. From the fluorescence spectra of FITC under different pH values (7 - 13), it can be seen that as the pH value increases, the fluorescence intensity of FITC changes. When the pH value is relatively low (such as pH = 7), the fluorescence intensity is relatively weak; when the pH value rises to a certain extent (such as pH = 13), the fluorescence intensity increases significantly. This indicates that the fluorescence property of FITC is relatively sensitive to the pH value. It may be that the dissociation state of the phenolic hydroxyl group in its molecular structure under different pH conditions affects the conjugated structure and electron cloud distribution of the molecule, thereby affecting fluorescence emission. The emission peaks of the fluorescence spectra under each pH condition are all around 500 - 550 nm, indicating that the fluorescence emission wavelength of FITC is basically not affected by the change in the pH value and has relative stability. This characteristic enables FITC to maintain the emission light color basically in the green light region in different acid-base environments, although the fluorescence intensity will change, showing certain advantages of FITC in fluorescence detection applications.

[0100] In the description of this application, it should be understood that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing this application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation to this application.

[0101] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0102] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0103] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0104] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array, characterized in that: The following steps are involved: 1) Preparation of CFA pigment screening tags; 2) Determination of the pigment screening tag of CFA; 3) The CFA pigment screening label determined in step 2 is used to measure the freshness of raw meat.

2. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array according to claim 1, characterized in that: The step 1 includes: 1.1) Pigment extraction: after pre-treating passion fruit peel and pitaya peel, anthocyanin extracts from passion fruit peel and pitaya peel were obtained; 1.2) Pigment screening: passion fruit peel anthocyanin extract, pitaya peel anthocyanin extract and alizarin are added according to a ratio to screen out experimental pigments, and the experimental pigments are prepared into FITC solution; 1.3) Alizarin and FITC concentration screening: preparing alizarin and FITC solutions with different solubility and performing concentration screening; 1.4) The selected alizarin and FITC are combined in selected concentrations, and the pigment solution is accurately loaded onto the PVDF membrane using a capillary to prepare a colorimetric / fluorescent dual-mode freshness indicator label.

3. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array according to claim 1, characterized in that: The pretreatment in step 1.1 includes drying the passion fruit peel and the pitaya peel in an oven at 55°C for 5 hours, crushing them into powder, adding an ethanol aqueous solution with a solid-liquid ratio of 1:10, and stirring at 50°C for 2 hours; removing the ethanol in the filtrate with a rotary evaporator under dark conditions at 35°C, and finally vacuum freeze-drying the solution to obtain passion fruit peel anthocyanin extract powder and pitaya peel anthocyanin extract, which are respectively put into brown sample bottles and stored at 4°C.

4. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array according to claim 1, characterized in that: In the step 1.2, alizarin, passion fruit peel anthocyanin, and pitaya peel anthocyanin are respectively prepared into 5 mg / mL solutions, 200 μL of each is taken into a 96-well plate, and after taking an initial photo, 10, 20, 50, 100, 200, and 500 mg / mL of ammonia solution are added respectively, and photos are taken after a response of 10 minutes; Use PS software to extract the RGB values ​​before and after the solution response, and calculate the ED value of the solution amine response according to the formula, and screen out the pigments for subsequent experiments based on the ED value.

5. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array according to claim 1, characterized in that: The step 2 of determining the pigment screening label of CFA includes: 2.1) Microstructure and EDS determination: Spray gold on the surface of each unit A1-A9 of the indicator tag, fix it on the sample stage, set the voltage to 10kV, set the scanning speed to medium, take pictures of the surface morphology and perform surface scanning to generate an element distribution map; 2.2) Water contact angle measurement: fix the sample on the sample stage to ensure the surface is level, use a 2 μL deionized water droplet for measurement, set the droplet injection speed to 0.5-2 μL / s, record the contact angle change within 15 s, and repeat the measurement 3 times at different positions.

6. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array according to claim 1, characterized in that: In step 3, the CFA pigment screening label is used to determine the freshness of raw meat, including: 3.1) Determination of TVB-N: After pretreatment of pork samples, HCl titration solution was added, and the content of volatile amines was calculated based on the titration results; 3.2) pH determination: after pre-treating the pork sample from which fat and connective tissue are removed, the supernatant is obtained, and the pH value of the supernatant is obtained by a pH meter; 3.3) TBARS determination, based on the reaction of thiobarbituric acid with lipid oxidation products to generate a pink compound, the absorbance was measured by colorimetry, and the TBARS value was calculated; 3.4) TVC determination: by obtaining the number of live bacteria in pork samples, the degree of contamination of pork samples can be assessed.

7. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array according to claim 6, characterized in that: The pretreatment in step 3.1 includes: treating the pork sample in a sterile environment at room temperature, homogenizing the minced meat sample in distilled water; after 30 minutes, filtering the mixture to obtain a liquid sample; adding a boric acid absorption solution and a pH indicator to the inner cavity of the Conway culture dish; adding a saturated potassium carbonate solution and a sample solution to the outer chamber in sequence; sealing the culture dish and culturing it at 37° C. for 2 hours, titrating the solution with 0.1 mol / L HCl, and calculating the content of volatile amines based on the titration result.

8. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array according to claim 6, characterized in that: The calculation of the TBARS value in step 3.3 includes: taking 10 g of pork sample, removing visible fat and connective tissue, and mincing or homogenizing the sample; extracting lipids using a chloroform-methanol mixture; centrifuging, collecting the lower chloroform phase, evaporating the solvent, and obtaining a lipid extract; mixing the lipid extract with a TBA reagent, heating at 90-100° C. for 30-60 min; cooling and centrifuging, and taking the supernatant; using a spectrophotometer to measure the absorbance at a wavelength of 530 nm, and calculating the TBARS value using a malondialdehyde standard curve.

9. A method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array according to claim 6, characterized in that: The step 3.4 of obtaining the number of live bacteria in the pork sample includes: Take an appropriate amount of pork sample and put it into a sterile homogenizing bag; add sterile saline and homogenize with a homogenizer for 1-2 minutes to prepare a 1:10 sample dilution; take 1 mL of the 1:10 dilution and add it to the sterile saline, mix well to prepare a 1:100 dilution; repeat the above steps to prepare 10-3, 10-4, and 10-5 gradient dilutions in sequence; take 1 mL of each gradient dilution and inject it into a sterile culture dish; pour about 15-20 mL of plate count agar that has been sterilized and cooled to 45-50°C and shake well, and wait for the agar to solidify; turn the culture dish upside down, put it in a constant temperature incubator, and culture it at 37°C for 48 hours; after the culture is completed, take out the culture dish, select the culture dish with a colony count between 30 and 300, and count the colony count with a colony counter.

10. A system for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array, using the method for rapidly detecting the freshness of raw meat using deep learning and colorimetric / fluorescence array in claims 1-9, characterized in that: Through the training process of ResNet-152 and Vision Transformer models, deep learning models are constructed based on data sets and applied to systems in smartphones and computers.

Images