Bear gall powder processing method and system

Abstract

The invention discloses a bear gall powder processing method and system. The method comprises the following steps: firstly, standing and clarifying bear bile to obtain supernate, then, carrying out centrifugal impurity removal treatment on the supernate to obtain water-soluble bear bile, and then, carrying out drying, crushing and sieving treatment on the water-soluble bear bile to obtain bear bile powder. Therefore, the bear gall powder with consistent product quality can be obtained.

Description

Technical Field

[0001] The present disclosure relates to the field of bear bile powder processing, and more specifically, to a bear bile powder processing method and a system thereof. Background Art

[0002] Bear bile is the dried bile in the gallbladder of black bears, Northeastern horse bears or brown bears. Bear bile can clear away heat and detoxify, improve eyesight, and has antispasmodic, anticonvulsant and choleretic effects. In addition, it has inhibitory effects on Haemophilus influenzae, Escherichia coli, Pneumococcus, etc. It plays a pharmacological and health-care role in the treatment of gallstones, acute icteric hepatitis and chronic hepatitis B.

[0003] Currently, the bear bile powder on the market is expensive and easy to absorb moisture. The content of tauroursodeoxycholic acid in bear bile powder is low, which greatly limits its application in medicine and health care. Therefore, an optimized bear bile powder processing scheme is expected. Summary of the invention

[0004] In view of this, the present disclosure proposes a bear bile powder processing method and a system thereof, which can obtain bear bile powder with consistent product quality.

[0005] According to one aspect of the present disclosure, a method for processing bear bile powder is provided, comprising:

[0006] The bear bile is allowed to stand and clarify to obtain a supernatant;

[0007] The supernatant is subjected to a centrifugal treatment to remove impurities to obtain water-soluble bear bile; and

[0008] The water-soluble bear bile is dried, crushed and sieved to obtain bear bile powder.

[0009] According to another aspect of the present disclosure, a bear bile powder processing system is provided, comprising:

[0010] A static clarification module, used for statically clarifying the bear bile to obtain a supernatant;

[0011] A centrifugal impurity removal module, used for performing a centrifugal impurity removal treatment on the supernatant to obtain water-soluble bear bile; and

[0012] The drying, crushing and sieving module is used to dry, crush and sieving the water-soluble bear bile to obtain bear bile powder.

[0013] According to an embodiment of the present disclosure, the bear bile is first allowed to stand for clarification to obtain a supernatant, then the supernatant is centrifuged to remove impurities to obtain water-soluble bear bile, and then the water-soluble bear bile is dried, crushed and sieved to obtain bear bile powder. In this way, bear bile powder with consistent product quality can be obtained.

[0014] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0016] Figure 1 A flow chart showing a method for processing bear bile powder according to an embodiment of the present disclosure is shown.

[0017] Figure 2 A flowchart of sub-step S120 of the bear bile powder processing method according to an embodiment of the present disclosure is shown.

[0018] Figure 3 A schematic diagram of the architecture of sub-step S120 of the bear bile powder processing method according to an embodiment of the present disclosure is shown.

[0019] Figure 4 A flowchart showing sub-step S122 of the bear bile powder processing method according to an embodiment of the present disclosure is shown.

[0020] Figure 5 A block diagram of a bear bile powder processing system according to an embodiment of the present disclosure is shown.

[0021] Figure 6 A diagram showing an application scenario of a bear bile powder processing method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present disclosure will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present disclosure, rather than all embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present disclosure.

[0023] As shown in the present disclosure and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0024] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0025] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0026] The present invention provides a method for processing bear bile powder. Figure 1 A flow chart of a method for processing bear bile powder according to an embodiment of the present disclosure is shown. Figure 1 As shown, the bear bile powder processing method according to the embodiment of the present disclosure includes the steps of: S110, allowing the bear bile to stand and clarify to obtain a supernatant; S120, centrifuging the supernatant to remove impurities to obtain water-soluble bear bile; and, S130, drying, crushing and sieving the water-soluble bear bile to obtain bear bile powder.

[0027] In the above processing method, when to stop the centrifugation operation is a technical issue worth considering. Usually, the basis for judging whether to stop the centrifugation operation is mainly to observe the bear bile in the centrifuge tube, which requires the operator to make a subjective judgment. However, different operators may have different observation standards, resulting in subjectivity and uncertainty in the results. Therefore, a solution is expected.

[0028] With the development of artificial intelligence technology, it has broad development space in the field of bear bile powder processing. For example, artificial intelligence technology can be applied to data analysis and prediction in bear bile powder processing. By collecting and analyzing a large amount of production data, quality data, process parameters and other information, AI can help identify potential optimization and improvement points. For the technical problem of when to stop the centrifugal operation, artificial intelligence technology provides a new solution.

[0029] Specifically, the technical concept of the present disclosure is to collect images of the centrifuged supernatant and combine it with artificial intelligence technology to determine when to stop the centrifugation operation.

[0030] Based on this, Figure 2 A flowchart of sub-step S120 of the bear bile powder processing method according to an embodiment of the present disclosure is shown. Figure 3 FIG. 1 is a schematic diagram showing a sub-step S120 of the bear bile powder processing method according to an embodiment of the present disclosure. Figure 2 and Figure 3As shown, according to the bear bile powder processing method of the embodiment of the present disclosure, the supernatant is centrifuged to remove impurities to obtain water-soluble bear bile, including the steps of: S121, acquiring a real-time state image of the centrifuged supernatant captured by a camera; S122, performing image feature extraction on the real-time state image to obtain a semantically combined supernatant state feature map; and, S123, determining whether to stop the centrifugation operation based on the semantically combined supernatant state feature map.

[0031] Accordingly, in the technical solution disclosed in the present invention, first, a real-time state image of the centrifuged supernatant collected by a camera is obtained. Then, the real-time state image is passed through a supernatant image feature extractor based on a pyramid network to obtain a shallow feature map of the supernatant and a deep feature map of the supernatant. That is, while capturing the deep implicit features of the real-time state image, the shallow texture feature information is retained. Specifically, the shallow texture features of the supernatant can reflect the distribution of particles in the supernatant. In the processing of bear bile powder, the purpose of the centrifugal operation is to separate the solid particles and impurities in the bear bile. By extracting the shallow texture features of the supernatant, the distribution of particles in the supernatant can be understood. If the particles are evenly distributed and sparse, it means that the centrifugal operation has fully separated most of the particles, and centrifugation can be considered to be stopped. If the particles are aggregated or unevenly distributed, it means that there are still many particles that need to be separated.

[0032] It is worth mentioning that the pyramid network is mainly used to solve the multi-scale problem in target detection. By simply changing the network connection, it can make independent predictions on different feature layers without basically increasing the computational complexity of the original model, which greatly improves the performance of small target detection. Therefore, the use of the pyramid network here can extract the multi-scale features of the real-time state image and feature information of different depths.

[0033] Then, the supernatant shallow feature map and the supernatant deep feature map are fused through the joint semantic propagation module to obtain a semantic joint supernatant state feature map. The joint semantic propagation module can extract the global semantic information and local semantic information of the deep features, propagate the semantics of the deep features to the shallow features, narrow the semantic gap between the two, and enrich the feature expression of the semantic joint supernatant state feature map.

[0034] In a specific example of the present disclosure, the encoding process of fusing the shallow feature map of the supernatant and the deep feature map of the supernatant to obtain a semantically joint supernatant state feature map through a joint semantic propagation module includes: first upsampling the deep feature map of the supernatant to obtain a resolution reconstruction feature map; then, performing point convolution, batch normalization and ReLU-based inactivation function operations on the global mean feature vector obtained by global mean pooling of the resolution reconstruction feature map to obtain a global semantic vector; then performing point convolution, batch normalization and ReLU-based inactivation function operations on the resolution reconstruction feature map to obtain a local semantic vector; then, performing point addition on the global semantic vector and the local semantic vector to obtain a semantic weight vector; then, using the semantic weight vector as the weight vector, weighting the shallow feature map of the steel plate surface state to obtain a semantically joint feature map; finally, fusing the shallow feature map of the supernatant with the semantically joint feature map to obtain the semantically joint supernatant state feature map.

[0035] Accordingly, if Figure 4 As shown, the real-time state image is subjected to image feature extraction to obtain a semantically-joint supernatant state feature map, including: S1221, the real-time state image is subjected to a supernatant image feature extractor based on a pyramid network to obtain a supernatant shallow feature map and a supernatant deep feature map; and, S1222, the supernatant shallow feature map and the supernatant deep feature map are fused through a joint semantic propagation module to obtain the semantically-joint supernatant state feature map. It should be understood that step S1221 is used for supernatant image feature extraction. In this step, the real-time state image is processed using a supernatant image feature extractor based on a pyramid network to obtain a shallow feature map and a deep feature map of the supernatant. The purpose of this step is to extract feature information related to the supernatant from the input real-time state image. By generating a shallow feature map and a deep feature map, supernatant features at different levels can be captured. Step S1222 is used for the joint semantic propagation module. In this step, the joint semantic propagation module is used to fuse the shallow feature map and the deep feature map of the supernatant to obtain a semantically joint supernatant state feature map. The function of this module is to fuse the shallow feature map and the deep feature map, and interact and transmit the information between the two through semantic propagation, so as to obtain a more comprehensive and accurate supernatant state feature map. The purpose of this step is to improve the understanding and representation ability of the supernatant state and provide more useful feature representations for subsequent tasks and analysis. In summary, step S1221 is used to extract the shallow feature map and the deep feature map of the supernatant, and step S1222 fuses and semantically propagates the two to obtain a semantically joint supernatant state feature map. These feature maps can be used for further data analysis, task processing or decision making.

[0036] It is worth mentioning that the Pyramid Network is a deep neural network structure for image processing and computer vision tasks. Its design is inspired by the concept of pyramid image processing, in which the image is decomposed into sub-images of different scales. The main purpose of the pyramid network is to process multi-scale information. By introducing multiple branches or multiple convolutional layers in the network, each branch or convolutional layer processes different scales of the input image, thereby capturing features at different levels. This multi-scale processing helps the network perceive and understand objects or features of different scales. The advantages of the pyramid network include: 1. Multi-scale feature extraction: The pyramid network can process information of different scales at the same time, thereby more comprehensively capturing the details and contextual information in the image. 2. Scale invariance: The pyramid network can extract features at different scales, making the network robust to the scale changes of the target. 3. Feature fusion: By extracting and fusing features at different scales, the pyramid network can generate richer and more comprehensive feature representations and improve the performance of the task. The pyramid network can effectively process multi-scale image information, improve the perception and expression capabilities of the model, and thus improve the accuracy and robustness of the task.

[0037] More specifically, in step S1222, the supernatant shallow feature map and the supernatant deep feature map are fused through a joint semantic propagation module to obtain the semantic joint supernatant state feature map, including: upsampling the supernatant deep feature map to obtain a resolution reconstruction feature map; performing point convolution, batch normalization and ReLU-based inactivation function operations on the global mean feature vector obtained by global mean pooling of the resolution reconstruction feature map to obtain a global semantic vector; performing point convolution, batch normalization and ReLU-based inactivation function operations on the resolution reconstruction feature map to obtain a local semantic vector; performing point addition on the global semantic vector and the local semantic vector to obtain a semantic weight vector; using the semantic weight vector as the weight vector, weighting the shallow feature map of the steel plate surface state to obtain a semantic joint feature map; and, fusing the supernatant shallow feature map with the semantic joint feature map to obtain the semantic joint supernatant state feature map.

[0038] It is worth mentioning that upsampling, also known as deconvolution or unpooling, is an image processing technique used to restore low-resolution feature maps to the resolution of the original input image. In deep neural networks, after multiple downsampling (pooling or convolution operations), the size of the feature map will gradually decrease, but the semantic information of the feature is enhanced. However, some tasks require prediction or analysis at the size of the original input image, so the feature map needs to be restored to its original size. Upsampling expands the pixels in the feature map to a larger size by filling or copying them. Common upsampling methods include deconvolution, bilinear interpolation, and unpooling. These methods can increase the size of the feature map according to specific interpolation rules or deconvolution operations, thereby achieving resolution reconstruction. In the generation of the semantic joint supernatant state feature map, upsampling is used to restore the supernatant deep feature map to the same resolution as the supernatant shallow feature map. This is to maintain a consistent size when fusing the two feature maps so that feature element-by-element addition or other operations can be performed. By upsampling, the detailed information of the deep feature map can be restored to its original size, so as to better fuse the shallow and deep features and obtain a more accurate and comprehensive semantic joint supernatant state feature map. In short, upsampling is used in image processing to restore the low-resolution feature map to the resolution of the original input image for subsequent analysis, prediction or task processing.

[0039] Pointwise convolution is a convolution operation that uses a 1x1 convolution kernel to perform convolution operations on each position of the input. Point convolution can be regarded as an independent linear transformation of each channel of the input feature map without changing the spatial size of the feature map. Batch normalization is a regularization technique used in neural networks. It normalizes the input of each small batch during the training process so that the input of each layer of the network has zero mean and unit variance. Batch normalization can accelerate the training process, reduce the dependence on initial parameters, and help prevent the problem of gradient vanishing or gradient exploding. The ReLU-based nonlinear activation function (RectifiedLinear Unit) is a commonly used activation function that sets negative input values ​​to zero and keeps positive input values ​​unchanged. The ReLU function has a simple calculation form and good nonlinear characteristics, which can enhance the expression ability of the neural network. Through the point convolution operation, each position of the feature map can be independently linearly transformed to extract local features in the feature map. Point convolution can adapt to the extraction requirements of different tasks or features by learning the weights of the convolution kernel. The batch normalization operation can normalize the feature map during the training process so that the input of each layer of the network has zero mean and unit variance, which helps to speed up the training process, improve the convergence speed and stability of the model, and reduce the dependence on the initial parameters. By using a nonlinear activation function based on ReLU, nonlinear transformations can be introduced to enhance the expressive power of the network. The ReLU function can retain the information of positive input values ​​and set negative input values ​​to zero, effectively activating neurons in the network. The combination of these operations can extract, normalize, and nonlinearly activate local features of the feature map, thereby obtaining global semantic vectors and local semantic vectors. The global semantic vector and the local semantic vector can be used to represent the global and local semantic information of the feature map, providing useful feature representations for subsequent tasks or analysis.

[0040] Element-wise Addition is an element-wise addition operation that adds the corresponding elements of two tensors (such as vectors, matrices, or feature maps) of the same size to generate a new tensor in which the value of each element is equal to the sum of the elements at the corresponding positions of the two input tensors at the corresponding positions. In the process of generating the semantic joint supernatant state feature map, the point addition operation is used to add the global semantic vector and the local semantic vector to obtain a semantic weight vector. This semantic weight vector can be used to weight the shallow feature map of the surface state of the steel plate to obtain a semantic joint feature map. The role of point addition is to fuse global and local semantic information. The global semantic vector contains the overall semantic information, while the local semantic vector contains the semantic information for a specific area. By adding them element-wise, the global and local semantic information can complement each other to obtain a more comprehensive and accurate semantic weight vector. The semantic weight vector can be regarded as a weight distribution to guide the weighted processing of the shallow feature map of the surface state of the steel plate. By element-wise multiplication or weighted addition of the semantic weight vector and the shallow feature map, the regions with higher semantic weights can be highlighted, the feature representation of these regions can be enhanced, and the influence of low-weight regions can be suppressed. In this way, a semantic joint feature map with more semantic information can be obtained. Finally, the semantic joint feature map is fused with the shallow feature map of the supernatant to obtain a semantic joint supernatant state feature map, where the fusion process can be element-wise addition, multiplication, or other appropriate operations, depending on the task requirements and design choices. This feature map will comprehensively consider the surface state and semantic information of the supernatant to provide a more comprehensive and accurate feature representation.

[0041] Furthermore, the semantically combined supernatant state feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether to stop the centrifugation operation.

[0042] Accordingly, based on the semantically combined supernatant state feature map, determining whether to stop the centrifugation operation includes: passing the semantically combined supernatant state feature map through a classifier to obtain a classification result, and the classification result is used to indicate whether to stop the centrifugation operation. It should be understood that the classifier is used to determine whether to stop the centrifugation operation, and by inputting the semantically combined supernatant state feature map into the classifier, a classification result can be obtained to indicate whether to stop the centrifugation operation. The classifier can be various machine learning algorithms or deep learning models, such as support vector machines (SVM), random forests (Random Forest), convolutional neural networks (CNN), etc. The classifier can determine whether the current supernatant state has reached the condition for stopping centrifugation according to the characteristics of the feature map by learning the mapping relationship from the feature map to the classification result. Through the combination of S1231 and S1232 steps, the semantically combined supernatant state feature map can be used to determine whether to stop the centrifugation operation through a classifier. In this way, based on the semantic information of the feature map and the discrimination ability of the classifier, automatic control and decision-making of the centrifugation operation can be achieved, and the accuracy and efficiency of the operation can be improved.

[0043] That is, in the technical solution of the present disclosure, the labels of the classifier include stopping the centrifugation operation (first label), and not stopping the centrifugation operation (second label), wherein the classifier determines which classification label the semantic joint supernatant state feature map belongs to through a soft maximum function. It is worth noting that the first label p1 and the second label p2 here do not contain artificially set concepts. In fact, during the training process, the computer model does not have the concept of "whether to stop the centrifugation operation". It only has two classification labels and the probability of the output feature under these two classification labels, that is, the sum of p1 and p2 is one. Therefore, the classification result of whether to stop the centrifugation operation is actually converted into a binary class probability distribution that conforms to the laws of nature through the classification label. In essence, the physical meaning of the natural probability distribution of the label is used, rather than the linguistic text meaning of "whether to stop the centrifugation operation".

[0044] It should be understood that the role of the classifier is to use the given categories and known training data to learn classification rules and classifiers, and then classify (or predict) unknown data. Logistic regression and SVM are often used to solve binary classification problems. For multi-class classification problems, logistic regression or SVM can also be used, but multiple binary classifications are required to form a multi-classification. However, this is prone to errors and is not efficient. Commonly used multi-classification methods include the Softmax classification function.

[0045] It is worth mentioning that fully connected encoding is a process of encoding the input vector through a fully connected layer. In classification tasks, fully connected encoding is often used to map the input feature vector to the hidden layer of the classifier to extract higher-level feature representations. The semantic joint supernatant state feature map is unfolded into an optimized classification feature vector, that is, each row or column of the feature map is regarded as a vector, so that the spatial structure of the feature map can be converted into a one-dimensional feature vector, which is convenient for subsequent fully connected encoding and classification operations. Through fully connected encoding, the optimized classification feature vector is input into the fully connected layer of the classifier. Each neuron in the fully connected layer is connected to all neurons in the previous layer, and each connection has a weight for weighted summation of the input features. The role of the fully connected layer is to map the input feature vector to a higher-dimensional feature space by learning appropriate weights, so as to better distinguish the feature differences between different categories. Finally, after the encoded classification feature vector passes through the fully connected layer, it will be input into the Softmax classification function of the classifier. The Softmax function maps each element of the input vector to a probability value, indicating the probability that the input belongs to each category. By comparing these probabilities, the most likely category can be determined, thereby obtaining a classification result, which is used to indicate whether to stop the centrifugal operation. The role of fully connected encoding is to perform nonlinear mapping on the input feature vector by learning appropriate weights to extract more discriminative feature representations. This can enhance the classifier's ability to distinguish between different categories and improve classification accuracy and generalization ability.

[0046] It is worth mentioning that those skilled in the art should be aware that before applying a deep neural network model for inference, the deep neural network model must be trained so that the deep neural network can implement specific functional capabilities.

[0047] Specifically, in an embodiment of the present application, it also includes: a training step for training the supernatant image feature extractor based on the pyramid network, the joint semantic propagation module and the classifier; the training step includes: obtaining training data, the training data including: a training real-time state image of the centrifuged supernatant; passing the training real-time state image through the supernatant image feature extractor based on the pyramid network to obtain a training supernatant shallow feature map and a training supernatant deep feature map; fusing the training supernatant shallow feature map and the training supernatant deep feature map through the joint semantic propagation module to obtain a training semantic joint supernatant state feature map; passing the training semantic joint supernatant state feature map through the classifier to obtain a classification loss function value; based on the classification loss function value and through gradient descent back propagation, the supernatant image feature extractor based on the pyramid network, the joint semantic propagation module and the classifier are trained, wherein, at each model iteration, the training semantic joint supernatant state feature map is optimized.

[0048] In the technical solution of the present application, when the training supernatant shallow feature map and the training supernatant deep feature map are fused through a joint semantic propagation module to obtain a training semantic joint supernatant state feature map, the training supernatant shallow feature map is weighted by the global-local semantic feature representation of the training supernatant deep feature map to obtain the training semantic joint supernatant state feature map, thereby, the training semantic joint supernatant state feature map corresponds to the shallow image semantic features of the training real-time state image expressed by each feature matrix of the training supernatant shallow feature map. Therefore, on the basis of the correspondence of the spatial distribution of image semantic features belonging to the feature matrix dimension of the training semantic joint supernatant state feature map and the training supernatant deep feature map, if the local distribution differences of the training semantic joint supernatant state feature map in different dimensions can be suppressed, the classification iteration effect can be improved.

[0049] Based on this, the present application optimizes the training semantic joint supernatant state feature map at each model iteration, including the steps of: expanding the training semantic joint supernatant state feature map into a training semantic joint supernatant state feature vector; determining a class probability value obtained by passing the training semantic joint supernatant state feature map through a classifier and a probability difference value minus the class probability value; calculating a relative class probability value obtained by dividing the class probability value by the probability difference value, and calculating a class probability synergy value obtained by multiplying the class probability value by the probability difference value; determining a first eigenvalue and a second eigenvalue at any two different positions of the training semantic joint supernatant state feature vector; subtracting the class probability from the first eigenvalue and the second eigenvalue, respectively. The relative values ​​are multiplied to obtain a first intermediate eigenvalue, and the first eigenvalue and the second eigenvalue are added and then divided by the class probability synergy value to obtain a second intermediate eigenvalue; the first intermediate eigenvalue and the second intermediate eigenvalue are added and used as the matrix values ​​at the coordinates corresponding to different positions of the first eigenvalue and the second eigenvalue to obtain a correction matrix; the matrix product of the correction matrix and the training semantic joint supernatant state feature vector is calculated to obtain an optimized training semantic joint supernatant state feature vector, wherein the training semantic joint supernatant state feature vector is a column vector; the optimized training semantic joint supernatant state feature vector is restored to an optimized training semantic joint supernatant state feature map.

[0050] That is, by taking the feature value of the training semantic joint supernatant state feature graph as a unit, for the relative class probability distribution form and the collaborative class probability distribution form of the class probability value and the probability difference value obtained by the classifier with the training semantic joint supernatant state feature graph as a feature set, a relativity-based joint representation and a synergy-based response representation are respectively performed on the feature value pairs of the training semantic joint supernatant state feature graph, so as to avoid the class induction bias caused by the corresponding local distribution differences between the feature values ​​of the training semantic joint supernatant state feature graph, so as to establish a robust understanding paradigm for class cognition of the training semantic joint supernatant state feature graph as a whole as a feature set, thereby improving the iterative effect of the training semantic joint supernatant state feature graph as a feature set in the classification process, that is, improving the speed of classification training and the accuracy of classification results when the training semantic joint supernatant state feature graph is classified by the classifier.

[0051] In summary, based on the bear bile powder processing method of the embodiment of the present disclosure, it can determine when to stop the centrifugation operation by capturing images of the centrifuged supernatant and combining artificial intelligence technology.

[0052] Figure 5 FIG. 1 is a block diagram of a bear bile powder processing system 100 according to an embodiment of the present disclosure. Figure 5As shown, the bear bile powder processing system 100 according to the embodiment of the present disclosure includes: a standing clarification module 110, used for standing and clarifying the bear bile to obtain a supernatant; a centrifugal impurity removal module 120, used for centrifugal impurity removal of the supernatant to obtain water-soluble bear bile; and a drying, crushing and screening module 130, used for drying, crushing and screening the water-soluble bear bile to obtain bear bile powder.

[0053] In a possible implementation, the centrifugal impurity removal module 120 includes: a camera acquisition unit, used to obtain a real-time state image of the centrifuged supernatant captured by a camera; an image feature extraction unit, used to extract image features from the real-time state image to obtain a semantically combined supernatant state feature map; and a centrifugal control unit, used to determine whether to stop the centrifugation operation based on the semantically combined supernatant state feature map.

[0054] In a possible implementation, the image feature extraction unit includes: a supernatant image feature extraction subunit, used to pass the real-time state image through a supernatant image feature extractor based on a pyramid network to obtain a supernatant shallow feature map and a supernatant deep feature map; and a fusion subunit, used to fuse the supernatant shallow feature map and the supernatant deep feature map through a joint semantic propagation module to obtain the semantic joint supernatant state feature map.

[0055] In a possible implementation, the fusion subunit is used to: upsample the supernatant deep feature map to obtain a resolution reconstruction feature map; perform point convolution, batch normalization and ReLU-based inactivation function operations on the global mean feature vector obtained by global mean pooling of the resolution reconstruction feature map to obtain a global semantic vector; perform point convolution, batch normalization and ReLU-based inactivation function operations on the resolution reconstruction feature map to obtain a local semantic vector; perform point addition on the global semantic vector and the local semantic vector to obtain a semantic weight vector; perform weighted processing on the shallow feature map of the steel plate surface state with the semantic weight vector as the weight vector to obtain a semantic joint feature map; and, fuse the supernatant shallow feature map with the semantic joint feature map to obtain the semantic joint supernatant state feature map.

[0056] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the bear bile powder processing system 100 have been described in detail above. Figures 1 to 5 The bear bile powder processing method has been described in detail in the description, and therefore, its repeated description will be omitted.

[0057] As described above, the bear bile powder processing system 100 according to the embodiment of the present disclosure can be implemented in various wireless terminals, such as a server with a bear bile powder processing algorithm. In one possible implementation, the bear bile powder processing system 100 according to the embodiment of the present disclosure can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the bear bile powder processing system 100 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the bear bile powder processing system 100 can also be one of the many hardware modules of the wireless terminal.

[0058] Alternatively, in another example, the bear bile powder processing system 100 and the wireless terminal may also be separate devices, and the bear bile powder processing system 100 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0059] Figure 6 The following is a diagram showing an application scenario of the bear bile powder processing method according to an embodiment of the present disclosure. Figure 6 As shown, in this application scenario, first, a real-time state image of the centrifuged supernatant collected by a camera is obtained (for example, Figure 6 D), then, the real-time status image is input into a server where a bear bile powder processing algorithm is deployed (for example, Figure 6 S) as shown in , wherein the server is capable of using the bear bile powder processing algorithm to process the real-time status image to obtain a classification result indicating whether to stop the centrifugal operation.

[0060] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0061] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for processing bear bile powder, characterized in that: include: The bear bile is allowed to stand and clarify to obtain a supernatant; The supernatant is subjected to a centrifugal treatment to remove impurities so as to obtain water-soluble bear bile; as well as The water-soluble bear bile is dried, crushed and sieved to obtain bear bile powder.

2. The method for processing bear bile powder according to claim 1, characterized in that: The supernatant is centrifuged and impurity-removed to obtain water-soluble bear bile, including: Acquiring a real-time state image of the centrifuged supernatant collected by a camera; Performing image feature extraction on the real-time state image to obtain a semantically-joint supernatant state feature map; and Based on the semantic combined supernatant state characteristic diagram, determine whether to stop the centrifugation operation.

3. The method for processing bear bile powder according to claim 2, characterized in that: Performing image feature extraction on the real-time state image to obtain a semantically combined supernatant state feature map includes: Passing the real-time state image through a supernatant image feature extractor based on a pyramid network to obtain a supernatant shallow layer feature map and a supernatant deep layer feature map; and The supernatant shallow feature map and the supernatant deep feature map are fused through a joint semantic propagation module to obtain the semantic joint supernatant state feature map.

4. The method for processing bear bile powder according to claim 3, characterized in that: The supernatant shallow feature map and the supernatant deep feature map are fused by a joint semantic propagation module to obtain the semantic joint supernatant state feature map, including: Upsampling the supernatant deep feature map to obtain a resolution reconstructed feature map; Performing point convolution, batch normalization and ReLU-based inactivation function operations on a global mean feature vector obtained by global mean pooling of the feature map reconstructed at the resolution to obtain a global semantic vector; Performing point convolution, batch normalization and ReLU-based inactivation function operations on the resolution reconstructed feature map to obtain a local semantic vector; Perform a dotted sum on the global semantic vector and the local semantic vector to obtain a semantic weight vector; Taking the semantic weight vector as a weight vector, weighted processing is performed on the shallow feature map of the surface state of the steel plate to obtain a semantic joint feature map; and The supernatant shallow feature map and the semantic joint feature map are fused to obtain the semantic joint supernatant state feature map.

5. The method for processing bear bile powder according to claim 4, characterized in that: Determining whether to stop the centrifugation operation based on the semantic combined supernatant state feature map includes: The semantically combined supernatant state feature map is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether to stop the centrifugation operation.

6. The method for processing bear bile powder according to claim 5, characterized in that: Also includes: A training step for training the pyramid network-based supernatant image feature extractor, the joint semantic propagation module and the classifier; The training step comprises: Acquiring training data, the training data comprising: a training real-time state image of a centrifuged supernatant; Passing the training real-time state image through a supernatant image feature extractor based on a pyramid network to obtain a training supernatant shallow feature map and a training supernatant deep feature map; fusing the training supernatant shallow feature map and the training supernatant deep feature map through a joint semantic propagation module to obtain a training semantic joint supernatant state feature map; Passing the training semantics-combined supernatant state feature map through a classifier to obtain a classification loss function value; The pyramid network-based supernatant image feature extractor, the joint semantic propagation module and the classifier are trained based on the classification loss function value and through back propagation of gradient descent, wherein the training semantic joint supernatant state feature map is optimized at each model iteration.

7. A bear bile powder processing system, characterized in that: include: A static clarification module, used for statically clarifying the bear bile to obtain a supernatant; A centrifugal impurity removal module, used for performing a centrifugal impurity removal treatment on the supernatant to obtain water-soluble bear bile; as well as The drying, crushing and sieving module is used to dry, crush and sieving the water-soluble bear bile to obtain bear bile powder.

8. The bear bile powder processing system according to claim 7, characterized in that: The centrifugal impurity removal module comprises: A camera acquisition unit, used to acquire a real-time state image of the centrifuged supernatant collected by a camera; An image feature extraction unit, configured to extract image features from the real-time state image to obtain a semantically-joint supernatant state feature map; and A centrifugation control unit is used to determine whether to stop the centrifugation operation based on the semantic combined supernatant state characteristic diagram.

9. The bear bile powder processing system according to claim 8, characterized in that: The image feature extraction unit comprises: a supernatant image feature extraction subunit, configured to pass the real-time state image through a supernatant image feature extractor based on a pyramid network to obtain a supernatant shallow layer feature map and a supernatant deep layer feature map; and A fusion subunit is used to fuse the supernatant shallow feature map and the supernatant deep feature map through a joint semantic propagation module to obtain the semantic joint supernatant state feature map.

10. The bear bile powder processing system according to claim 9, characterized in that: The fusion subunit is used for: Upsampling the supernatant deep feature map to obtain a resolution reconstructed feature map; Performing point convolution, batch normalization and ReLU-based inactivation function operations on a global mean feature vector obtained by global mean pooling of the feature map reconstructed at the resolution to obtain a global semantic vector; Performing point convolution, batch normalization and ReLU-based inactivation function operations on the resolution reconstructed feature map to obtain a local semantic vector; Perform a dotted sum on the global semantic vector and the local semantic vector to obtain a semantic weight vector; Taking the semantic weight vector as a weight vector, weighted processing is performed on the shallow feature map of the surface state of the steel plate to obtain a semantic joint feature map; and The supernatant shallow feature map and the semantic joint feature map are fused to obtain the semantic joint supernatant state feature map.

Images