Method for preparing dark plum extract based on supercritical extraction method

Abstract

The invention discloses a method for preparing a dark plum extract based on a supercritical extraction method. The method comprises the steps of obtaining a first classification result of smoked plums based on the morphology of the smoked plums; the smoked plums are smashed into powder, the characterization value of the smoked plum powder is obtained through an electronic nose method, the characterization value is used for characterizing the relative strength of polar substances in the smoked plum powder, and a second classification result is obtained based on the relative strength; selecting the type and content of an entrainer based on the first classification result and the second classification result, and preparing the dark plum extract by utilizing a supercritical extraction method based on the selected type and content of the entrainer. The method can improve the extraction efficiency of the dark plum extract.

Description

Technical Field

[0001] The invention relates to the technical field of supercritical extraction, and in particular to a method for preparing a black plum extract based on supercritical extraction. Background Art

[0002] Supercritical CO 2 Extraction technology has a strong solubility for non-polar or weakly polar substances, while black plums contain more polar substances such as organic acids, which require additional entrainers to improve the extraction efficiency. This increases the complexity and cost of the process. Strongly polar substances usually refer to compounds with uneven charge distribution in the molecules. In chemistry, "polarity" refers to the property that the positive and negative charge centers in a molecule do not overlap. This uneven charge distribution causes the molecule to have polarity, that is, one end has a partial positive charge and the other end has a partial negative charge. In supercritical CO 2 During the extraction process, due to CO 2 As a non-polar substance, it has limited solubility for substances with higher polarity or molecular weight. To improve this situation, supercritical CO 2 A small amount of other substances are added to the supercritical CO. These substances are called entrainers. Entrainers can improve the target substance in supercritical CO. 2 For solutes with high polarity, the solubility in supercritical CO 2 The solvent is poorly soluble in water, and adding an entrainer can change the activity of the solvent and improve the extraction efficiency. Commonly used entrainers include methanol, chloroform, etc. The amount added is generally determined by experiments.

[0003] The polar substances such as organic acids in black plums include the following categories: Citric acid: The molecular formula is C 6 H 8 O 7 , is one of the main organic acids in black plum. Malic acid: The molecular formula is C 4 H 6 O 5 , and is also an important organic acid component in black plum. Tartaric acid: The molecular formula is C 4 H 6 O 6 , present in black plum. Succinic acid: molecular formula is C 4 H 6 O 4 , is one of the organic acid components in black plum. Acetic acid: The molecular formula is CH 3 COOH, this organic acid is also found in black plums. Neochlorogenic acid: The molecular formula is C 16H 18 O 9 , is the organic acid component in black plum. Quinic acid: The molecular formula is C 7 H 12 O 6 , and is also one of the organic acids in black plum. Propionic acid: The molecular formula is CH 3 CH 2 COOH, found in black plum. Chlorogenic acid: molecular formula is C 16 H 18 O 9 Fumaric acid: The molecular formula is C 4 H 4 O 4 , and is also one of the organic acids in black plum. Oxalic acid: The molecular formula is H 2 C 2 O 4 , present in black plum. Vanillic acid: molecular formula is C 8 H 8 O 4 , is the organic acid component in black plum. Gentisic acid (Gallicacid): The molecular formula is C 7 H 6 O 4 , present in black plum. Ferulic acid: molecular formula is C 10 H 10 O 4 , which is also one of the organic acids in black plum. Caffeic acid: The molecular formula is C 9 H 8 O 4 , which are present in black plum. These organic acids are one of the main active ingredients of black plum, mostly present in aliphatic and aromatic organic acids, and have multiple pharmacological effects such as antioxidant and anti-inflammatory.

[0004] However, in experiments, we often find that there may be differences between different batches of black plum extracts. The cultivation areas of black plums are all over the country, and the varieties are numerous and complex. Black plums from different origins may have different chemical component contents due to different environmental factors such as climate and soil. For example, Yunnan black plums have the highest content of acid components, among which the content of citric acid is as high as more than 35%. There are obvious differences in the active ingredients and contents in black plums with different processing methods. For example, the content of citric acid and malic acid in the raw product decreased significantly after processing, and the content of gallic acid in the charcoal product increased significantly. Black plum samples vary in size and the proportion of fruit cores is different. Studies on the effect of purification on the quality of black plums show that the transfer and transformation laws of other chemical components that may occur during the purification process will affect the quality of the final extract. Different extraction process conditions, such as liquid-to-solid ratio, soaking time and extraction time, will affect the quality and yield of black plum extracts. For example, the optimal extraction process conditions for total organic acids in black plum are a liquid-solid ratio of 14:1 mL / g, a soaking time of 70 min, an extraction time of 1.5 h, and 2 extractions. Under these conditions, the extraction rate of total organic acids in black plum is 35.21%. Influence of processing methods: Different processing methods will affect the intrinsic quality of black plums, thereby affecting the chemical composition of the extracts. Studies have shown that compared with raw black plums, the total content of 16 components in black plum pulp increased, while the content in charcoal decreased.

[0005] Therefore, we found that optimizing supercritical CO 2 The application of extraction technology to black plum extract still faces some difficulties. The biggest problem is how to control the ratio of entrainer to black plum from different batches and origins to increase the yield of black plum extract under the same production equipment. 2 Analysis of volatile components in extracts by gas chromatography-mass spectrometry[Lin Wenjin, Xu Rongqing, Zhang Yamin. Supercritical CO 2 2 Analysis of volatile components in extracts by gas chromatography-mass spectrometry [J]. Modern Chinese Medicine Research and Practice, 2011, 25(06): 27-29.] mentioned: Supercritical CO 2 The results of volatile substances in the black plum extract extracted by extraction technology are quite different from those of GC-MS analysis by existing technology. The main reason may be related to the different extraction methods. 2 The extraction of black plum extract by extraction technology still faces the problem of extraction efficiency. For some components, supercritical CO 2 The extraction is not as efficient as the traditional extraction method, especially when dealing with black plums with different "polarity" (different origins, different batches), the difference will be more obvious. Summary of the invention

[0006] In view of this, the purpose of the present invention is to propose a method for preparing black plum extract based on supercritical extraction, which can improve the extraction efficiency of black plum extract.

[0007] According to one aspect of the present invention, a method for preparing a black plum extract based on supercritical extraction is provided, comprising:

[0008] Based on the morphology of black plum, the first classification result of black plum is obtained;

[0009] The black plum is crushed into powder, and an electronic nose method is used to obtain a characterization value of the black plum powder, where the characterization value is used to characterize the relative intensity of polar substances in the black plum powder, and a second classification result is obtained based on the relative intensity;

[0010] Based on the first classification results and the second classification results, the type and content of the entrainer are selected, and based on the selected type and content of the entrainer, the black plum extract is prepared by supercritical extraction.

[0011] In the above technical scheme, this study integrates the visual image of black plum with its medicinal properties and constructs a mapping relationship based on historical data. The model can accurately associate the image data of black plum with its entrainer and its content information. This method optimizes the adaptation process of entrainers and their contents by combining image analysis and medicinal properties, thereby avoiding a large number of tedious measurement steps in traditional experiments and significantly improving production efficiency. The innovation of this study is to introduce electronic nose technology to capture the medicinal properties of black plum. Thanks to the rich organic acid components in black plum, such as malic acid and succinic acid, these key components can be effectively identified by electronic nose technology. In the supercritical extraction process, polar substances are a major obstacle to yield. By qualitatively and quantitatively analyzing these organic acids and combining historical data, we can accurately adapt different entrainers and contents to optimize the extraction process.

[0012] In some embodiments, based on the morphology of dark plums, a first classification result of dark plums is obtained, specifically:

[0013] Obtain a black plum image, input the image into a pre-trained image classification neural network, predict the black plum type through the neural network, and use the output result as the first classification result.

[0014] In the above technical solution, the powerful feature extraction and classification capabilities of deep learning can be used to quickly and accurately identify the types of black plums, while avoiding the cumbersome feature engineering and classifier design steps in traditional methods, thereby improving classification efficiency and accuracy.

[0015] In some embodiments, the black plum is crushed into powder, and the electronic nose method is used to obtain the characterization value of the black plum powder, specifically:

[0016] Put 100g-300g of crushed black plum powder into a beaker, seal it and let it stand for 2-10min;

[0017] The gas in the beaker is introduced into the sensor channel of the electronic nose through the collection tube. When the response value of the sensor tends to be stable, the average value of the data in the last 1 / 3 of the measurement time is taken as the characterization value.

[0018] In the above technical solution, the electronic nose technology can quickly and accurately identify the odor characteristics of black plum powder with high sensitivity and good repeatability. By collecting gas response values ​​and taking the average value, random errors can be reduced and the stability and reliability of the measurement can be improved. In addition, the electronic nose technology does not require complex sample pretreatment and is a green detection technology. It is easy to operate and can achieve real-time online monitoring, greatly reducing material and labor costs. At the same time, the electronic nose technology has a wide range of applications and can be applied to a variety of different application scenarios, with good adaptability and flexibility.

[0019] In some embodiments, the black plum is crushed into powder, and the electronic nose method is used to obtain the characterization value of the black plum powder, specifically:

[0020] Put 100g-300g of crushed black plum powder into a beaker, seal it and let it stand for 2-10min;

[0021] The gas in the beaker is introduced into the sensor channel of the electronic nose through the collection tube. When the response value of the sensor tends to be stable, the response curve area of ​​the electronic nose intensity and time is calculated and the area is used as the characterization value.

[0022] In the above technical scheme, the electronic nose method is used to obtain the characterization value of the black plum powder, and the processing method of calculating the response curve area as the characterization value has the following advantages: First, this method can fully capture the volatile compound characteristics of the black plum powder, because the response curve area integrates the overall response of the sensor to the odor, thereby providing the overall information of the volatile components in the sample. Secondly, this method can improve the accuracy and reliability of the data, because it can reflect the overall information of the sampling data by analyzing the dynamic characteristics and mainstream characteristic information of the sensor response. Combined with the characteristics of bionic breath sampling, its response differential characteristics can be used to verify the accuracy of the sampling data. In addition, this method also enhances the practicality of the electronic nose, reduces erroneous samples, and objectively improves the recognition rate of the electronic nose, making the application range of the electronic nose wider and more practical. Finally, this method can provide a quantitative odor intensity measurement through the calculation of the response curve area, which is helpful for further data analysis and comparison.

[0023] In some embodiments, based on the first classification result and the second classification result, the type and content of the entrainer are selected, specifically:

[0024] Through a limited number of experiments, a first mapping relationship between different polarity grades and entrainers and entrainer content categories is constructed;

[0025] Through limited experiments, a second mapping relationship between different categories in each grade and entrainers and entrainer content categories is constructed;

[0026] A database is constructed based on the first mapping relationship and the second mapping relationship, and the mapping relationship database is used to subsequently select the type and content of the entrainer based on the first classification result and the second classification result.

[0027] In the above technical solution, the processing method of the type and content of the entrainer is selected by constructing a first mapping relationship and a second mapping relationship database, which has the advantage of being able to systematize and standardize the entrainer selection process. This method can reduce the number of experiments and improve experimental efficiency, and through the application of the database, it can quickly retrieve and apply existing experimental results, thereby accelerating the design and execution of new experiments. In addition, this method can provide a more accurate and reliable basis for the selection of entrainers because it is based on experimental data and statistical analysis rather than relying on experience or intuition. Such a systematic method helps to improve the efficiency and effectiveness of supercritical extraction, especially when dealing with black plums of different polarities and categories, it can ensure the consistency and repeatability of the extraction process.

[0028] In some embodiments, based on the first classification result and the second classification result, the type and content of the entrainer are selected, specifically:

[0029] Obtaining black plum images and electronic nose data, and respectively extracting feature vectors of black plum images using convolutional neural networks (CNNs), and extracting feature vectors of electronic nose data using principal component analysis (PCA) or autoencoders; connecting or fusing the image feature vectors and the electronic nose feature vectors in series, and jointly inputting them into a pre-trained first classification neural network, predicting the black plum category through the neural network, and selecting the type and content of the entrainer based on matching the output results with the database;

[0030] The first classification neural network is trained in the following way: image data and electronic nose data of black plums from different origins and categories are obtained, each pair of image data and corresponding electronic nose data are paired to form a data pair, and the origin and category of the black plum are labeled for each pair of data; supervised learning methods are used, with image feature vectors and electronic nose feature vectors as inputs, and the origin and category of the black plum as labels, and the neural network is trained through the cross entropy loss function.

[0031] In the above technical solution, the multimodal information of image and electronic nose data can be comprehensively utilized to improve the accuracy and robustness of the classification of black plum categories. By extracting image feature vectors through CNN, visual features such as the morphology and texture of black plums can be captured; and by extracting feature vectors of electronic nose data through PCA or autoencoder, the odor characteristics of black plums can be obtained. After fusing these two feature vectors and inputting them into the pre-trained neural network, the model can not only rely on a single data source when predicting the black plum category, but also comprehensively consider visual and olfactory information, thereby improving the accuracy of classification. In addition, the use of cross entropy loss function for supervised learning training can effectively optimize the model parameters, so that the model can better distinguish different categories of black plums when predicting. The implementation of this method not only improves the efficiency of classification, but also enhances the model's ability to recognize different categories of black plums, especially when dealing with black plum categories with similar appearance or smell.

[0032] In some embodiments, based on the first classification result and the second classification result, the type and content of the entrainer are selected, specifically:

[0033] Obtaining black plum images and electronic nose data, and respectively extracting feature vectors of black plum images using convolutional neural networks (CNNs), and extracting feature vectors of electronic nose data using principal component analysis (PCA) or autoencoders; connecting or fusing the image feature vectors and the electronic nose feature vectors in series, and jointly inputting them into a pre-trained second classification neural network, and predicting the type and content of entraining agents through the neural network;

[0034] The second classification neural network is trained as follows:

[0035] The image data and electronic nose data of black plums from different origins and categories were obtained, and each pair of image data and the corresponding electronic nose data were paired to form a data pair. The origin, category, type and content of the black plum entrainers were annotated for each pair of data. The supervised learning method was used, with the image feature vector and the electronic nose feature vector as input, and the origin, category, type and content of the black plum entrainers as labels, and the neural network was trained using the cross entropy loss function.

[0036] In the above technical solution, accurate prediction of the types and contents of black plums and entrainers can be achieved, thereby optimizing the supercritical extraction process. By combining the image feature vector extracted by CNN and the electronic nose feature vector extracted by PCA or autoencoder, the model can integrate visual and olfactory information to improve the accuracy and robustness of the prediction. In addition, the use of supervised learning methods and cross-entropy loss functions to train the neural network can ensure that the model can effectively learn and distinguish the characteristics of black plums of different origins and categories, as well as the corresponding optimal entrainer types and contents. This method not only improves the accuracy of the prediction, but also enhances the generalization ability of the model, enabling it to quickly and accurately guide the selection of entrainers in practical applications, thereby improving extraction efficiency and product quality.

[0037] In some embodiments, based on the first classification result and the second classification result, selecting the type and content of the entrainer further includes:

[0038] Obtaining black plum images and electronic nose data, and extracting feature vectors of black plum images using convolutional neural networks (CNNs), and extracting feature vectors of electronic nose data using principal component analysis (PCA) or autoencoders; connecting or fusing the image feature vectors and the electronic nose feature vectors in series, and inputting them into a pre-trained multi-task neural network, and predicting the type and content of entrainers and corresponding process parameters of black plums through the neural network;

[0039] The multi-task neural network is trained as follows:

[0040] The image data and electronic nose data of black plums from different origins and categories, as well as the corresponding optimized extraction process parameters and supercritical extraction rates were obtained to construct data pairs of image data, electronic nose data and process parameters. The origin, category, type and content of entrainers and supercritical extraction rate of black plums were labeled for each pair of data. The supervised learning method was used with the image feature vector, electronic nose feature vector and process parameter feature vector as input, and the origin and category of black plums, type and content of entrainers and supercritical extraction rate as labels, and the neural network was trained using the cross entropy loss function and the mean square error loss function.

[0041] In the above technical solution, multi-task learning can be achieved, that is, the black plum category, entrainer type and content and corresponding process parameters can be predicted simultaneously in a unified model. By combining the image feature vector extracted by CNN and the electronic nose feature vector extracted by PCA or autoencoder, the model can integrate visual and olfactory information to improve the accuracy and robustness of the prediction. In addition, through supervised learning, the neural network is trained using the cross entropy loss function and the mean square error loss function. The model can simultaneously learn classification and regression tasks to optimize the prediction results. This method not only improves the accuracy of the prediction, but also enhances the generalization ability of the model, so that it can quickly and accurately guide the selection of entrainers and the adjustment of process parameters in practical applications, thereby improving the extraction efficiency and product quality, and realizing a more intelligent and automated supercritical extraction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 It is a schematic flow chart of an embodiment of a method for preparing black plum extract based on supercritical extraction method of the present invention;

[0044] Figure 2 The present invention is a neural network reasoning process diagram of an embodiment of a method for preparing black plum extract based on supercritical extraction. DETAILED DESCRIPTION

[0045] The present invention will be further described in detail below in conjunction with the accompanying drawings and examples. It is particularly noted that the following examples are only used to illustrate the present invention, but are not intended to limit the scope of the present invention. Similarly, the following examples are only partial embodiments of the present invention rather than all embodiments, and all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0046] The invention provides a method for preparing a black plum extract based on a supercritical extraction method, which can improve the extraction efficiency of the black plum extract.

[0047] See also Figure 1 , the method comprises the following steps:

[0048] S1. Obtain the first classification result of black plum based on the black plum morphology;

[0049] S2, crushing the ebony into powder, and using an electronic nose method to obtain a characterization value of the ebony powder, the characterization value is used to characterize the relative intensity of polar substances in the ebony powder, and obtaining a second classification result based on the relative intensity;

[0050] S3. Based on the first classification result and the second classification result, the type and content of the entrainer are selected, and based on the selected type and content of the entrainer, the black plum extract is prepared by supercritical extraction.

[0051] From the background technology, we know that polar substances such as organic acids in black plums specifically include the following categories: citric acid, malic acid, tartaric acid, succinic acid, acetic acid, neochlorogenic acid, quinic acid, propionic acid, chlorogenic acid, fumaric acid, oxalic acid, vanillic acid, gentisic acid, ferulic acid, caffeic acid, etc. The main ingredients in black plum extract are as follows:

[0052] (1) Organic acids: Black plums are rich in organic acids, such as citric acid and malic acid. These organic acids are one of the main medicinal ingredients of black plums and are often used as quality evaluation indicators.

[0053] (2) Flavonoids: including quercetin and kaempferol. Quercetin plays an important role in the prevention and treatment of chronic diseases such as cardiovascular disease and cancer, while kaempferol shows anti-obesity and anti-diabetic potential. (3) Terpenes and sterols: These ingredients have anti-tumor effects.

[0054] (4) Volatile components: have antioxidant and anti-inflammatory effects.

[0055] (5) Amino acids and sugars: These components have also been detected in black plums and have certain pharmacological effects. (6) Lipids: Black plums also contain lipid components.

[0056] (7) Alkaloids: They also have anti-tumor effects.

[0057] (8) Polysaccharide extract: Black plum extract also contains polysaccharides, which have certain antioxidant effects.

[0058] As can be seen from the above, citric acid and malic acid involved in organic acids are the main causes of low extraction efficiency, but they are also one of the important components of black plum extract. Therefore, for black plum extract, the main difficulty lies in improving the extraction efficiency of the main component organic acid. To this end, the following work was done in this case, as follows:

[0059] (1) Issues with different batches of black plums

[0060] As described in the background technology, there may be differences between different batches of black plum extracts. However, if it is necessary to use traditional extraction methods (such as water decoction, ultrasonic extraction, heating reflux extraction, etc.) to extract black plum extracts every time, and then further measure the specific differences and then adapt different contents of entrainers, it is not as efficient as the traditional solution. For this reason, we adopt the following solution for the batch problem:

[0061] 1) Image classification,

[0062] In this case, a neural network is used to distinguish the types of black plums in the image:

[0063] ① Data collection and preprocessing: Collect image data of different types of black plums. This data needs to include the image path and the corresponding label. Preprocess the image, including resizing the image, converting it to Tensor, etc. For example, you can use torchvision.transforms to preprocess the image, resize the image to a uniform size, such as 256x256 pixels, and convert it to Tensor. (For details, please refer to the image processing steps of the neural network, and the specific details will not be explained in detail.)

[0064] ②Build a neural network model: Build a convolutional neural network (CNN) model. CNN is widely used due to its excellent performance in image processing tasks. The model contains multiple convolutional layers, pooling layers, and fully connected layers. Define the model structure, including convolutional layers, activation functions, pooling layers, and fully connected layers. You can use deep learning frameworks such as TensorFlow and PyTorch to implement the model. (For details, please refer to the image processing steps of the neural network. The specific details will not be explained in detail.)

[0065] ③Training model: Divide the data set into a training set and a test set. Use the training set to train the model, and optimize the model parameters through the back propagation algorithm and gradient descent. During the training process, monitor the loss value and accuracy to evaluate the model performance.

[0066] ④Evaluate the model: Use the test set to evaluate the performance of the model. Calculate the accuracy of the model to determine the classification effect of the model on different types of black plums.

[0067] ⑤Model optimization: Adjust and optimize the model based on the evaluation results, including adjusting the network structure, changing the optimizer, adjusting the learning rate, etc.

[0068] ⑥Model deployment: After training is completed, the model is deployed to actual applications to classify new black plum images.

[0069] Through the above steps, a neural network model can be constructed and trained to distinguish different types of black plums in the image. This method can use the powerful feature extraction and classification capabilities of deep learning to quickly and accurately identify the types of black plums, while avoiding the cumbersome feature engineering and classifier design steps in traditional methods, and improving classification efficiency and accuracy. Through the neural network, we can preliminarily classify the specific varieties of black plums to achieve the purpose of preliminary division. The reason for the so-called preliminary division is that we found in the test process that the results of the neural network are not particularly ideal. If we want to further improve the accuracy of the classification, we need to solve it from the medicinal properties of the black plum itself. Because the appearance characteristics of black plums are relatively similar, especially between certain varieties, subtle differences in morphology, color and texture may make it difficult for neural networks to recognize and learn. If the neural network model is further optimized, the cost of deployment will be further increased. Therefore, this case further considers the classification problem from the medicinal properties of black plums themselves. At the same time, it should be noted that the specific neural network data can be collected by itself to adapt to the actual scenarios required by each. The data in this case uses the experimental results of this laboratory in the past as experimental data, which will not be explained here. Those skilled in the art can set it according to actual needs.

[0070] 2) Odor classification

[0071] Based on existing technologies, electronic nose technology has the advantages of short response time and fast detection speed. It can detect fruit odor non-destructively, and the measurement results are close to consumers' sense of smell, which has more direct and practical value.

[0072] In further research, we found from existing research papers that malic acid and succinic acid can be identified by electronic noses [Arvinte, A.; Rotariu, L.; Bala, C. Amperometric Low-Potential Detection of Malic Acid Using Single-Wall Carbon Nanotubes Based Electrodes. Sensors 2008, 8, 1497-1507. https: / / doi.org / 10.3390 / s8031497and Jiang Hao. Study on the baking process of charred hawthorn based on electronic nose and HPLC multi-component content analysis. Tianjin Traditional Chinese Medicine, 2022, 39(5): 668-674.]. Furthermore, Liu Zhiyang et al. [Liu Zhiyang, Liu Yan. Study on the identification of apple storage period by electronic sensory combined with chemometrics [J]. Northern Horticulture, 2017, (22): 137-141.] adopted electronic nose technology and applied partial least squares regression (PLSR) algorithm to construct a mathematical model to describe the changes in organic acid content and soluble solids content (SSC) content of apples during storage. Through a detailed statistical analysis of the adopted regression analysis method, this study aims to develop an efficient detection technology to quickly evaluate the storage life, total acid content and SSC of apples. The results of the study revealed that when electronic nose technology distinguishes apples in different storage stages, the combined contribution rate of each sensor unit reached 90.62%, indicating that the technology has significant effectiveness in distinguishing the storage quality of apples. Based on this, we determined that there is a strong linear relationship between the detection data of the electronic nose and the content of organic acids in the fruit, which can be used for preliminary analysis. Similarly, it can be seen that malic acid and succinic acid in the organic acids of black plums can be identified by the electronic nose. In addition, the non-destructive detection advantage of the electronic nose means that it does not impose a large burden on the overall process. To this end, the following plan was made in this case:

[0073] When performing supercritical extraction, it is often necessary to crush and sieve the black plums and put them into a beaker for standby use. After crushing, put them into the extraction tank for the next step. Therefore, in this case, before placing it into the extraction tank, put the black plum powder (it should be noted that different extraction equipment has different requirements for the particle size of the powder, which is not specifically limited here) into a beaker, and seal the beaker with a sealing film to ensure that the gas does not leak, and let it stand in the beaker for 2-5 minutes. Subsequently, the gas is introduced into the sensor channel of the electronic nose through the collection tube, and when the response value of the sensor tends to stabilize, the average value of the data in the last 1 / 3 of the measurement time is taken for analysis, or the response diagram corresponding to the electronic nose intensity and time is used to calculate the response area to obtain the characterization value. The whole process takes about ten minutes, and since the electronic nose does not need too much measurement sample, generally about 100-300g is enough, it can be measured while crushing, saving measurement time.

[0074] However, electronic noses can generally only perform qualitative analysis. Electronic noses convert the response of gas to electrical signals through gas sensor arrays, thereby realizing gas monitoring and analysis. Electronic nose technology can identify polar substances such as organic acids in black plums and estimate their concentrations to a certain extent. However, this estimate is usually relative. For this reason, we consider using the average value of the data in the last 1 / 3 of the measurement time as the characterization value of the electronic nose data feature (this consideration is based on stability considerations, but different electronic noses have different accuracies, and the characterization value can also be calculated by calculating the response curve area) as the evaluation score of malic acid and succinic acid in this batch of black plum powder, and use the sensor value as the evaluation indicator of the polarity of black plums. The reason for this setting is that malic acid, succinic acid, etc. can be recognized by electronic noses and are important organic acid components in black plums. The data value of the electronic nose can evaluate the content of organic acids to a certain extent. Of course, the complex pretreatment method in the above paper can also be used to accurately measure the specific content. But the purpose of this case is to improve efficiency, and for the grades with similar evaluation indicators, the same entrainer content can be referred to, and further accurate measurement values ​​are obtained, which have limited impact on the final output. Further, after obtaining the data value of the electronic nose, it is roughly graded according to the data value. Since the results of different batches of measurements are different, the measurement results of the electronic nose can only obtain a relative value. Specifically, the grading method can be set by those skilled in the art according to actual needs, and will not be repeated here. In this way, a mapping relationship can be constructed between the entrainer and the entrainer content category according to the grading method. This is because the data value of the electronic nose in the data collection can reflect the approximate level of the polar substances (malic acid and succinic acid) of the black plum, and a corresponding relationship is established through a limited number of experiments after grading. For example, in the first grade, the entrainer and the content should be a mapping relationship of A, thereby simplifying the optimization steps of the entrainer and the entrainer content in the subsequent order. The details are as follows: based on the image classification results, the black plum category is obtained, and based on the electronic nose classification results, the polarity grade is obtained, the type and content of the entrainer are selected, and through a limited number of experiments, a first mapping relationship between different polarity grades and the entrainer and the entrainer content category is constructed; through a limited number of experiments, a second mapping relationship between different categories in each grade and the entrainer and the entrainer content category is constructed; based on the first mapping relationship and the second mapping relationship, a database is constructed, and the mapping relationship database is used for the subsequent selection of the type and content of the entrainer based on the first classification results and the second classification results.

[0075] Based on this indicator, as an optional example, we further made the following improvements to the above neural network:

[0076] See also Figure 2As an example of this embodiment, this case uses the most basic CNN neural network for explanation. However, it should be noted that there are many types of neural networks, and the setting methods are also different. This case only shows a simple solution by way of example. Those skilled in the art can choose a neural network according to actual needs, and will not be repeated here.

[0077] 1) Image data input and processing: The image data is first input into the convolutional neural network (CNN). CNN extracts the features of the image through multiple convolutional layers and pooling layers, and finally outputs a feature vector 1, which represents the key features of the image data.

[0078] 2) Electronic nose data processing and feature extraction: The electronic nose data is preprocessed, such as normalization or PCA dimension reduction. The preprocessed electronic nose data can be used to extract features through structures such as fully connected layers or autoencoders to form feature vectors 2, which represent the key features of the electronic nose data.

[0079] 3) Feature vector merging (early fusion): Feature vector 1 (from CNN) and feature vector 2 (from electronic nose data) are merged in the merging layer. The merging can be a simple concatenation, that is, concatenating the two feature vectors to form a longer feature vector.

[0080] 4) Fully connected layer: The merged feature vector is input into one or more fully connected layers (also called dense layers) for further feature integration.

[0081] 5) Classification layer: The neural network outputs the classification results through the classification layer. For multi-class classification problems, the classification layer usually uses the Softmax function for output, and each node represents the probability of a category.

[0082] 6) Training, the classification neural network is trained in the following way: obtaining image data and electronic nose data of black plums from different origins and categories, pairing each pair of image data with the corresponding electronic nose data to form a data pair, and labeling the origin and category of the black plum for each pair of data; using the supervised learning method, taking the image feature vector and the electronic nose feature vector as input,

[0083] This scheme is a typical early fusion neural network model, which combines the image data and electronic nose data processed by CNN, further processes them through the fully connected layer, and uses the classification layer to classify black plums. Among them, the specific model training and other methods of constructing the data set can refer to the existing technology, which will not be repeated here. Finally, based on the classification results, the appropriate entrainer and the content of the entrainer are selected according to the historical data. The historical data mentioned here refers to the experimental data of the entrainer and the entrainer content set for different black plums in the past. Compared with simple grading, thanks to the neural network technology, the specific content of organic acids can be predicted more accurately, and then the specific entrainer and content can be controlled to further improve production efficiency. The advantage of this scheme is that it can comprehensively utilize the multimodal information of image and electronic nose data to improve the accuracy and robustness of black plum category classification. By extracting image feature vectors through CNN, the visual features such as morphology and texture of black plums can be captured; and by extracting feature vectors of electronic nose data through PCA or autoencoder, the odor characteristics of black plums can be obtained. After fusing these two feature vectors and inputting them into the pre-trained neural network, the model can not only rely on a single data source when predicting the category of black plums, but also comprehensively consider visual and olfactory information, thereby improving the accuracy of classification. In addition, using the cross entropy loss function for supervised learning training can effectively optimize the model parameters, so that the model can better distinguish different categories of black plums when predicting. The implementation of this method not only improves the efficiency of classification, but also enhances the model's ability to recognize different categories of black plums, especially when dealing with black plum categories with similar appearance or smell.

[0084] As an optional example, the neural network architecture and training set can also be modified so that the final prediction result of the model is the entrainer and entrainer content data. This part can be adjusted by adjusting the data label when constructing the data set, so as to achieve the mapping relationship between the learning input data (images and electronic nose data) and the entrainer and entrainer content categories, and the network output is the category probability. The specific mode can refer to the prior art, and this embodiment gives the following example:

[0085] Obtaining black plum images and electronic nose data, and respectively extracting feature vectors of black plum images using convolutional neural networks (CNNs), and extracting feature vectors of electronic nose data using principal component analysis (PCA) or autoencoders; connecting or merging the image feature vectors and the electronic nose feature vectors in series, and jointly inputting them into a pre-trained classification neural network, and predicting the type and content of entraining agents through the neural network;

[0086] The classification neural network is trained as follows:

[0087] The image data and electronic nose data of black plums from different origins and categories were obtained, and each pair of image data and the corresponding electronic nose data were paired to form a data pair. The origin, category, type and content of the black plum entrainers were annotated for each pair of data. The supervised learning method was used, with the image feature vector and the electronic nose feature vector as input, and the origin, category, type and content of the black plum entrainers as labels, and the neural network was trained using the cross entropy loss function.

[0088] The advantages of this solution are that it can accurately predict the types of black plums and the types and contents of entrainers, thereby optimizing the supercritical extraction process. By combining the image feature vectors extracted by CNN and the electronic nose feature vectors extracted by PCA or autoencoders, the model can integrate visual and olfactory information to improve the accuracy and robustness of the prediction. In addition, the use of supervised learning methods and cross-entropy loss functions to train the neural network can ensure that the model can effectively learn and distinguish the characteristics of black plums from different origins and categories, as well as the corresponding optimal types and contents of entrainers. This method not only improves the accuracy of the prediction, but also enhances the generalization ability of the model, enabling it to quickly and accurately guide the selection of entrainers in practical applications, thereby improving extraction efficiency and product quality.

[0089] It should be noted that the production process involves factors such as extraction pressure, extraction temperature, extraction time, and particle size, and includes them in the training range, and the model is trained specifically to adapt to the corresponding process parameters to further improve the accuracy of the model. It should be noted that such processing requires a more complex model to be provided, and reference can be made to existing neural networks. Those skilled in the art can select a neural network according to actual needs and make corresponding adjustments. The specific model can refer to the prior art, and this embodiment provides the following examples:

[0090] Obtaining black plum images and electronic nose data, and extracting feature vectors of black plum images using convolutional neural networks (CNNs), and extracting feature vectors of electronic nose data using principal component analysis (PCA) or autoencoders; connecting or fusing the image feature vectors and the electronic nose feature vectors in series, and inputting them into a pre-trained multi-task neural network, and predicting the type and content of entrainers and corresponding process parameters of black plums through the neural network;

[0091] The multi-task neural network is trained as follows:

[0092] The image data and electronic nose data of black plums from different origins and categories, as well as the corresponding optimized extraction process parameters and supercritical extraction rates were obtained to construct data pairs of image data, electronic nose data and process parameters. The origin, category, type and content of entrainers and supercritical extraction rate of black plums were labeled for each pair of data. The supervised learning method was used with the image feature vector, electronic nose feature vector and process parameter feature vector as input, and the origin and category of black plums, type and content of entrainers and supercritical extraction rate as labels, and the neural network was trained using the cross entropy loss function and the mean square error loss function.

[0093] The advantage of this solution is that it can achieve multi-task learning, that is, to simultaneously predict the black plum category, entrainer type and content, and corresponding process parameters in a unified model. By combining the image feature vector extracted by CNN and the electronic nose feature vector extracted by PCA or autoencoder, the model can integrate visual and olfactory information to improve the accuracy and robustness of the prediction. In addition, through supervised learning, the neural network is trained using the cross entropy loss function and the mean square error loss function. The model can simultaneously learn classification and regression tasks to optimize the prediction results. This method not only improves the accuracy of the prediction, but also enhances the generalization ability of the model, enabling it to quickly and accurately guide the selection of entrainers and the adjustment of process parameters in practical applications, thereby improving extraction efficiency and product quality, and realizing a more intelligent and automated supercritical extraction process.

[0094] The advantage of this case is that this study combines the visual image of black plum with its medicinal properties and constructs a mapping model based on historical data. The model can accurately associate the image data of black plum with its entrainer and its content information. This method optimizes the adaptation process of entrainers and their contents by combining image analysis and medicinal properties, thereby avoiding a large number of tedious measurement steps in traditional experiments and significantly improving production efficiency. The innovation of this study is to introduce electronic nose technology to capture the medicinal properties of black plum. Thanks to the rich organic acid components in black plum, such as malic acid and succinic acid, these key components can be effectively identified by electronic nose technology. In the supercritical extraction process, polar substances are a major obstacle to yield. By qualitatively and quantitatively analyzing these organic acids and combining historical data, we can accurately adapt different entrainers and contents to optimize the extraction process. In the data acquisition stage, the data values ​​provided by the electronic nose can reflect the relative content of polar substances (malic acid and succinic acid) in black plum. After grading, a limited number of experiments are used to establish an accurate correspondence. For example, in the first gear, the entrainer and content should correspond to a specific mapping relationship, which simplifies the subsequent optimization steps of the entrainer and its content. In addition, considering the application potential of neural networks in prediction and classification, this study proposes a neural network-based recognition scheme, which can further improve the classification accuracy and is of great significance to improving production efficiency.

[0095] The above descriptions are only some embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for preparing black plum extract based on supercritical extraction, characterized in that: The method includes: Based on the morphology of black plum, the first classification result of black plum is obtained; The black plum is crushed into powder, and an electronic nose method is used to obtain a characterization value of the black plum powder, where the characterization value is used to characterize the relative intensity of polar substances in the black plum powder, and a second classification result is obtained based on the relative intensity; Based on the first classification results and the second classification results, the type and content of the entrainer are selected, and based on the selected type and content of the entrainer, the black plum extract is prepared by supercritical extraction.

2. The method for preparing black plum extract based on supercritical extraction as claimed in claim 1, characterized in that: Based on the morphology of black plum, the first classification result of black plum is obtained, specifically: Obtain a black plum image, input the image into a pre-trained image classification neural network, predict the black plum type through the neural network, and use the output result as the first classification result.

3. The method for preparing black plum extract based on supercritical extraction as claimed in claim 1, characterized in that: The black plums were crushed into powder, and the electronic nose method was used to obtain the characterization values ​​of the black plum powder, specifically: Put 100g-300g of crushed black plum powder into a beaker, seal it and let it stand for 2-10min; The gas in the beaker is introduced into the sensor channel of the electronic nose through the collection tube. When the response value of the sensor tends to be stable, the average value of the data in the last 1 / 3 of the measurement time is taken as the characterization value.

4. The method for preparing black plum extract based on supercritical extraction as claimed in claim 1, characterized in that: The black plums were crushed into powder, and the electronic nose method was used to obtain the characterization values ​​of the black plum powder, specifically: Put 100g-300g of crushed black plum powder into a beaker, seal it and let it stand for 2-10min; The gas in the beaker is introduced into the sensor channel of the electronic nose through the collection tube. When the response value of the sensor tends to be stable, the response curve area of ​​the electronic nose intensity and time is calculated and the area is used as the characterization value.

5. The method for preparing black plum extract based on supercritical extraction as claimed in claim 1, characterized in that: Based on the results of the first and second classifications, the type and content of the entrainer are selected, specifically: Through a limited number of experiments, a first mapping relationship between different polarity grades and entrainers and entrainer content categories is constructed; Through limited experiments, a second mapping relationship between different categories in each grade and entrainers and entrainer content categories is constructed; A database is constructed based on the first mapping relationship and the second mapping relationship, and the mapping relationship database is used to subsequently select the type and content of the entrainer based on the first classification result and the second classification result.

6. The method for preparing black plum extract based on supercritical extraction as claimed in claim 5, characterized in that: Based on the results of the first and second classifications, the type and content of the entrainer are selected, specifically: Obtaining black plum images and electronic nose data, and respectively extracting feature vectors of black plum images using convolutional neural networks (CNNs), and extracting feature vectors of electronic nose data using principal component analysis (PCA) or autoencoders; connecting or fusing the image feature vectors and the electronic nose feature vectors in series, and jointly inputting them into a pre-trained first classification neural network, predicting the black plum category through the neural network, and selecting the type and content of the entrainer based on matching the output results with the database; The first classification neural network is trained in the following way: image data and electronic nose data of black plums from different origins and categories are obtained, each pair of image data and corresponding electronic nose data are paired to form a data pair, and the origin and category of the black plum are labeled for each pair of data; supervised learning methods are used, with image feature vectors and electronic nose feature vectors as inputs, and the origin and category of the black plum as labels, and the neural network is trained through the cross entropy loss function.

7. The method for preparing black plum extract based on supercritical extraction as claimed in claim 5, characterized in that: Based on the results of the first and second classifications, the type and content of the entrainer are selected, specifically: Obtaining black plum images and electronic nose data, and respectively extracting feature vectors of black plum images using convolutional neural networks (CNNs), and extracting feature vectors of electronic nose data using principal component analysis (PCA) or autoencoders; connecting or fusing the image feature vectors and the electronic nose feature vectors in series, and jointly inputting them into a pre-trained second classification neural network, and predicting the type and content of entraining agents through the neural network; The second classification neural network is trained as follows: The image data and electronic nose data of black plums from different origins and categories were obtained, and each pair of image data and the corresponding electronic nose data were paired to form a data pair. The origin, category, type and content of the black plum entrainers were annotated for each pair of data. The supervised learning method was used, with the image feature vector and the electronic nose feature vector as input, and the origin, category, type and content of the black plum entrainers as labels, and the neural network was trained using the cross entropy loss function.

8. The method for preparing black plum extract based on supercritical extraction as claimed in claim 5, characterized in that: Based on the results of the first and second classifications, the type and content of the entrainer are selected, which also includes: Obtaining black plum images and electronic nose data, and extracting feature vectors of black plum images using convolutional neural networks (CNNs), and extracting feature vectors of electronic nose data using principal component analysis (PCA) or autoencoders; connecting or fusing the image feature vectors and the electronic nose feature vectors in series, and inputting them into a pre-trained multi-task neural network, and predicting the type and content of entrainers and corresponding process parameters of black plums through the neural network; The multi-task neural network is trained as follows: The image data and electronic nose data of black plums from different origins and categories, as well as the corresponding optimized extraction process parameters and supercritical extraction rates were obtained to construct data pairs of image data, electronic nose data and process parameters. The origin, category, type and content of entrainers and supercritical extraction rate of black plums were labeled for each pair of data. The supervised learning method was used with the image feature vector, electronic nose feature vector and process parameter feature vector as input, and the origin and category of black plums, type and content of entrainers and supercritical extraction rate as labels, and the neural network was trained using the cross entropy loss function and the mean square error loss function.

Images