Automatic sorting system for reproductive test samples

Abstract

The invention relates to the technical field of sample classification, in particular to an automatic classification system for reproductive test samples, which comprises the following steps: a real-time dynamic characteristic monitoring module for monitoring key characteristics of the reproductive test samples in real time by using a detection device and capturing key characteristic micro-change information of the samples; the dynamic adaptive classification control module is used for evaluating the micro-change information in real time based on the micro-change information transmitted by the real-time dynamic characteristic monitoring module and selecting an optimal classification path according to an evaluation result; and the multi-mode classification module constructs a multi-mode classification model based on a machine learning algorithm, and the multi-mode classification model classifies the specimens in combination with the optimal classification path. According to the method, the classification model can automatically adjust the model parameters when the characteristic data slightly changes, so that the classification effect is further improved, and particularly, the method has remarkable advantages when being used for processing high-dynamic reproductive samples with complex characteristics.

Description

Technical Field

[0001] The present invention relates to the technical field of sample classification, and in particular to an automated classification system for reproductive test sample specimens. Background Art

[0002] With the rapid development of modern reproductive medicine, the management and classification of reproductive test specimens has become one of the key links to ensure the accuracy and efficiency of test results. At present, many laboratories have begun to introduce automated equipment in the sample classification process to reduce manual operation errors and labor intensity. However, the existing automated classification system still faces some technical challenges when dealing with complex and dynamically changing reproductive test samples, which limits its scope of application and classification effect.

[0003] After reproductive test samples (such as sperm, eggs, embryos, etc.) are collected, their physical and chemical properties (such as metabolite concentration, redox potential and cell activity) may undergo small but important dynamic changes over time or with changes in processing conditions. Existing automated systems generally lack the ability to monitor these changes in real time, which may result in samples being misclassified during processing due to changes that are not discovered in time. In addition, due to the diversity and complexity of reproductive test samples, existing classification systems are often unable to flexibly adjust classification strategies according to the characteristics of different samples, thereby reducing the accuracy and reliability of classification.

[0004] Secondly, most existing automated classification systems use static classification algorithms, which are difficult to cope with dynamic environments where sample characteristics change over time. These systems usually rely on pre-set classification models. When the characteristics of the sample change during the classification process, the model often cannot be adjusted in time, resulting in poor classification results. Especially when dealing with reproductive samples with highly dynamic and complex characteristics, static models are prone to misclassification, which in turn affects the credibility of the test results. Summary of the invention

[0005] Based on the above objectives, the present invention provides an automated classification system for reproductive test sample specimens.

[0006] The automated classification system for reproductive test specimens includes the following steps:

[0007] Real-time dynamic characteristic monitoring module: Detection devices are used to monitor the key characteristics of reproductive test specimens in real time, including metabolite concentration, redox potential and cell activity. The monitoring data is processed through high-frequency sampling and real-time analysis algorithms to capture micro-change information of key characteristics of the specimens;

[0008] Dynamic adaptive classification control module: Based on the micro-change information transmitted by the real-time dynamic characteristic monitoring module, the dynamic adaptive classification control module evaluates the micro-change information in real time by integrating the fuzzy logic control unit and the adaptive neural fuzzy inference unit (ANFIS), and selects the optimal classification path according to the evaluation results. This module can handle the characteristic changes of different types of sample specimens under dynamic conditions, ensuring accurate classification of samples under any environmental conditions;

[0009] Multimodal classification module: A multimodal classification model is constructed based on a machine learning algorithm. The multimodal classification model classifies specimens in combination with the optimal classification path. The multimodal classification model also combines time series analysis of dynamic characteristic data. During the classification process, the model parameters are dynamically adjusted when slight changes occur in key characteristics according to the time series analysis, thereby improving the accuracy and stability of the classification.

[0010] Optionally, the detection device includes an electrochemical sensor, a microelectrode sensor and a biosensor, the metabolite concentration is detected by the electrochemical sensor, the redox potential is measured by the microelectrode sensor, the cell activity is evaluated by the biosensor, and the sensor continuously collects data.

[0011] Optionally, the real-time analysis algorithm adopts a dynamic time warping (DTW) algorithm. During the data analysis process, the dynamic time warping algorithm is used to compare and analyze the sample characteristic curves collected by the detection device, match the data collected at different time points, and identify small but significant change trends in the sample characteristic curves. By comparing the time series differences between the current data and the baseline data, the DTW algorithm can accurately capture small changes in sample characteristics and generate corresponding change signals, so that subsequent processing strategies can be adjusted in time when an abnormality is detected.

[0012] Optionally, the dynamic adaptive classification control module specifically includes:

[0013] Micro-change information processing: pre-process micro-change information and standardize key characteristic data to ensure consistency and comparability of data input;

[0014] The preprocessed micro-change information is preliminarily evaluated by the fuzzy logic control unit, the metabolite concentration, redox potential and cell activity data are input into the fuzzy logic system, the fuzzy sets (such as "low", "medium", "high") and fuzzy rules (such as "if the metabolite concentration is high and the cell activity is low, the classification path is A") of the input data are defined, and the fuzzy output is generated by the fuzzy inference engine, and the fuzzy output is provided as an input signal to the adaptive neuro-fuzzy inference unit;

[0015] Based on the fuzzy output generated by fuzzy logic control, the adaptive neural fuzzy reasoning unit makes inference selection on the classification path. Combining the advantages of neural network and fuzzy logic, the adaptive neural fuzzy reasoning unit first converts the fuzzy output into a membership function, trains and optimizes the membership function through a multi-layer feedforward neural network, and learns the nonlinear relationship between micro-change information and the optimal classification path.

[0016] After completing the reasoning process of the adaptive neuro-fuzzy inference unit, the optimal classification path is selected according to the output results.

[0017] The multi-layer feedforward neural network structure includes an input layer, a membership function layer, a rule layer, a normalization layer and an output layer, wherein;

[0018] Input layer: receives the fuzzy output (membership value) of the fuzzy logic control unit;

[0019] Membership function layer: matches the fuzzy output with the preset membership function;

[0020] Rule layer: Each node corresponds to a fuzzy rule, and the results of the membership function are weighted and summed to represent the support of a specific classification path;

[0021] Normalization layer: Normalize the output of the rule layer to ensure that the output results are within a comparable range;

[0022] Output layer: The final classification path selection is calculated based on the normalized output. The nodes in the output layer generate the final classification path decision through weighted summation:

[0023] Among them, p i is the specific classification path corresponding to each rule, is the normalized output of the normalization layer;

[0024] The adaptive neuro-fuzzy inference unit optimizes the parameters of the neural network through a training process, including the center c and standard deviation σ of the membership function and the weight w of the rule layer i The goal of training is to minimize the error between the predicted classification path and the actual classification path. The error back propagation algorithm is used to iteratively adjust the parameters in the network to improve the prediction accuracy of the model. The loss function E is expressed as: Among them, y k For the actual path selection result, is the path selection result predicted by the adaptive neural fuzzy inference unit, and M is the number of samples. The error back propagation algorithm uses the output of the loss function to calculate the error gradient, and updates the network weights through this gradient information to make the network prediction closer to the actual value, thereby minimizing the value of the loss function; through the back propagation algorithm, adjust the parameters c, σ and w in the network i By minimizing the loss function, the system can more accurately select the optimal classification path.

[0025] Optionally, the machine learning algorithm in the multimodal classification module adopts a support vector machine, and a multimodal classification model is constructed based on the support vector machine. The training is performed based on the characteristic data (metabolite concentration, redox potential and cell activity) of the reproductive test samples in the training data set. The multimodal classification model training includes normalizing the collected characteristic data, extracting features that affect the classification results through feature engineering methods, and using them as input for model training. The influential features include metabolite concentration C met , redox potential E redox , cell activity A cell ,The preprocessed data is input into the multimodal classification model for training, and the input features and the corresponding specimen category labels in the training dataset are used to optimize the model parameters to minimize the classification error.

[0026] Optionally, a multi-modal classification model is constructed by a support vector machine (SVM) algorithm to classify the multi-dimensional characteristic data of reproductive test samples. The core of SVM is to find an optimal decision boundary (hyperplane) to effectively classify sample data of different categories;

[0027] The input feature vector of the support vector machine is expressed as: x = [C met ,E redox ,A cell ], where C met is the metabolite concentration, E redox is the redox potential, A cell is cell activity; the support vector machine maps the input feature vector x to the classification result through the decision function f(x), and the support vector machine also includes a classification decision, and the classification result of the classification decision is determined by a symbolic function, which is expressed as:

[0028] Class(x)=sign(f(x))=sign(w T x+b);

[0029] When f(x)>0, the sample is classified as a positive class (such as a healthy sample); when f(x)<0, the sample is classified as a negative class (such as an abnormal sample);

[0030] The goal of the support vector machine is to maximize the distance from the decision boundary to the nearest sample point, called the margin, which is defined as: Therefore, the optimization problem is expressed as:

[0031] Constraints: y i (w T x i +b)≥1, where y i For sample x i The label is +1 or -1, indicating positive or negative class.

[0032] Optionally, the decision function f(x) is expressed as: f(x)=w T x+b, where w is the weight vector, indicating the direction of the hyperplane, and b is the bias, indicating the offset of the hyperplane. The goal is to find the optimal w and b so that sample data of different categories can be correctly classified.

[0033] Optionally, during the classification process, the multimodal classification model uses the optimal classification path as a guide, performs classification based on path guidance, and adjusts weights in the decision layer. The decision layer is a decision function. In the decision function f(x), the weight vector w is adjusted according to the optimal path.

[0034] Optionally, the multi-mode classification model is combined with time series analysis of dynamic characteristic data. During the classification process, dynamically adjusting model parameters when a key characteristic undergoes a slight change according to the time series analysis specifically includes:

[0035] Time series modeling: Build a time series model for each key feature;

[0036] Dynamically adjust model parameters: When time series analysis shows that key features have changed slightly, the multimodal classification model is adaptively updated by adjusting the weight vector w;

[0037] By regularly evaluating the prediction effect of the time series model and updating the parameter settings of the multimodal classification model in real time, it is ensured that the classification model can respond promptly and maintain high-precision classification capabilities when the sample characteristics change.

[0038] Optionally, the time series model adopts an autoregressive model, and the time series modeling includes modeling the metabolite concentration characteristic time series data to predict the change of the metabolite concentration characteristic value over time, which is expressed as:

[0039]

[0040] Among them, C met (t+1) is the predicted metabolite concentration at time point t+1;

[0041] C met (t),C met (t-1),…,C met (tp) are the metabolite concentration data at time points t, t-1, …, tp, respectively;

[0042] is the autoregressive coefficient, reflecting the degree of influence of metabolite concentration at each historical time point on future concentration;

[0043] ∈ t is white noise, representing random errors or unpredictable factors in the model.

[0044] Beneficial effects of the present invention:

[0045] The present invention, by introducing a real-time dynamic characteristic monitoring module, achieves high-precision, real-time monitoring of key characteristics of reproductive test samples, continuously tracks key indicators such as metabolite concentration, redox potential and cell activity, and captures small but important changes in samples through high-frequency sampling and real-time analysis algorithms, which not only improves the sensitivity to sample changes, but also ensures that the system can respond quickly to these changes through real-time data transmission, avoiding classification errors that may be caused by delays or data missing in traditional methods.

[0046] The present invention combines dynamic adaptive classification control with support vector machine for high-precision classification. By integrating fuzzy logic control and adaptive neural fuzzy reasoning units, it is possible to evaluate the micro-change information of key characteristics of samples in real time, and dynamically adjust the classification path accordingly. The multi-modal classification model constructed with support vector machine can accurately select the optimal classification path under different characteristic change conditions to ensure the accuracy and robustness of classification. Through the time series analysis model, the classification model can also automatically adjust the model parameters when the characteristic data undergoes micro-changes, thereby further improving the classification effect, especially showing significant advantages when processing reproductive samples with highly dynamic and complex characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the present invention or 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 in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0048] Figure 1 A schematic diagram of a classification system module according to an embodiment of the present invention;

[0049] Figure 2 Schematic diagram of a multi-modal classification model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0051] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0052] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0053] like Figure 1-Figure 2 As shown, the automated classification system for reproductive test sample specimens includes the following steps:

[0054] Real-time dynamic characteristic monitoring module: Detection devices are used to monitor the key characteristics of reproductive test specimens in real time, including metabolite concentration, redox potential and cell activity. The monitoring data is processed through high-frequency sampling and real-time analysis algorithms to capture micro-change information of key characteristics of the specimens;

[0055] Dynamic adaptive classification control module: Based on the micro-change information transmitted by the real-time dynamic characteristic monitoring module, the dynamic adaptive classification control module evaluates the micro-change information in real time by integrating the fuzzy logic control unit and the adaptive neural fuzzy inference unit (ANFIS), and selects the optimal classification path according to the evaluation results. This module can handle the characteristic changes of different types of sample specimens under dynamic conditions, ensuring accurate classification of samples under any environmental conditions;

[0056] Multimodal classification module: A multimodal classification model is constructed based on a machine learning algorithm. The multimodal classification model classifies specimens in combination with the optimal classification path. The multimodal classification model also combines time series analysis of dynamic characteristic data. During the classification process, the model parameters are dynamically adjusted when slight changes occur in key characteristics according to the time series analysis, thereby improving the accuracy and stability of the classification.

[0057] The detection device includes an electrochemical sensor, a microelectrode sensor, and a biosensor. The metabolite concentration is detected by the electrochemical sensor, the redox potential is measured by the microelectrode sensor, and the cell activity is evaluated by the biosensor. The sensor continuously collects data.

[0058] Metabolite concentration monitoring: Using electrochemical sensors, the concentration of metabolites in the sample is estimated by measuring the current generated by the electrochemical reaction on the electrode surface. The sensor calculates the concentration of the metabolite based on the intensity of the current generated.

[0059] Oxidation-reduction potential monitoring: A microelectrode sensor consisting of a reference electrode and a working electrode is used to measure the oxidation-reduction potential (ORP) of a sample to reflect the electron transfer capacity in the sample. The measurement formula for ORP is: Where E is the measured potential, E0 is the standard potential, R is the gas constant, T is the temperature, n is the number of electron transfers, F is the Faraday constant, [Ox] and [Red] are the concentrations of the oxidized state and the reduced state, respectively. The redox potential is obtained by measuring the potential difference between the electrodes.

[0060] Cell activity monitoring: Bioluminescence sensors are used to evaluate cell activity by detecting light signals generated during cell metabolism. The specific implementation method is to use fluorescence or bioluminescence technology. Cells will produce or consume specific fluorescent molecules or luminescent molecules during metabolism. The number of photons generated is detected by photodiodes to calculate the cell activity level.

[0061] The real-time analysis algorithm adopts the dynamic time warping (DTW) algorithm. During the data analysis process, the dynamic time warping algorithm is used to compare and analyze the sample characteristic curves collected by the detection device, match the data collected at different time points, and identify the small but significant change trends in the sample characteristic curves. By comparing the time series differences between the current data and the baseline data, the DTW algorithm can accurately capture the slight changes in the sample characteristics and generate corresponding change signals, so that the subsequent processing strategy can be adjusted in time when an anomaly is detected.

[0062] In the comparative analysis of the sample characteristic curves, the dynamic time warping algorithm is used to identify small but significant change trends. The calculation steps of the DTW algorithm are as follows:

[0063] Distance matrix calculation: First construct two time series X = (x1, x2, ..., x n ) and Y=(y1,y2,...,y m ), where each element D(i,j) represents x i and j The Euclidean distance between: D(i,j)=(x i -y j ) 2 ;

[0064] Cumulative distance matrix: Calculate the cumulative distance matrix C(i,j) to find the best matching path:

[0065] C(i,j)=D(i,j)+min(C(i-1,j),C(i,j-1),C(i-1,j-1));

[0066] Among them, the boundary condition is C(1,1)=D(1,1);

[0067] Path backtracking: Start from C(n,m) and backtrack to C(1,1) to determine the best matching path W = (w1,w2,...,w k ), where each w k Corresponding to a matching point in time series X and Y;

[0068] Total matching distance: Ultimately, the DTW algorithm computes the difference between two time series by minimizing the cumulative matching distance, thereby identifying small changes in the sample characteristic curve:

[0069] The DTW algorithm compares the real-time data of the sample with the historical baseline data to identify the slight changes in the sample characteristics (such as metabolite concentration, redox potential and cell activity) over time, and provides them to the dynamic adaptive classification control module.

[0070] The dynamic adaptive classification control module specifically includes:

[0071] Micro-change information processing: pre-process the micro-change information and standardize the key characteristic data to ensure the consistency and comparability of data input. Let the characteristics be represented by C met (metabolite concentration),

[0072] E redox (Redox potential), A cell (cell activity), the standardized processing formula is as follows:

[0073]

[0074] Among them, μ C , μE , μ A is the average value of historical data of each characteristic, σ C , σ E , σ A is the standard deviation of each characteristic, and the standardized data C met ′、E redox ′、A cell ′ is used as the input of the subsequent fuzzy logic control and adaptive neuro-fuzzy inference unit;

[0075] The preprocessed micro-change information is preliminarily evaluated by the fuzzy logic control unit, the metabolite concentration, redox potential and cell activity data are input into the fuzzy logic system, the fuzzy sets of the input data (such as "low", "medium", "high") and fuzzy rules (such as "if the metabolite concentration is high and the cell activity is low, the classification path is A") are defined, and the fuzzy output is generated by the fuzzy inference engine. The fuzzy output will be provided as an input signal to the adaptive neuro-fuzzy inference unit; define the fuzzy set of the input data, for example, the metabolite concentration C met ′ is divided into three fuzzy sets of “low”, “medium” and “high”, and its membership function μ 低 (C met ′),μ 中 (C met ′),μ 高 (C met ′), are defined as:

[0076]

[0077]

[0078] Define fuzzy rules, for example:

[0079] Rule 1: If C met 'High and A cell ′ is low, then the classification path selects A;

[0080] Rule 2: If E redox ' and A cell ′, the classification path selects B.

[0081] Through the fuzzy inference system, the fuzzy logic control unit calculates the output of each rule and synthesizes the final fuzzy output through the weighted average method. Let the membership degree of the output classification path be μ pathA and μ pathB , the synthesis is expressed as:

[0082] Among them, W A and W B is the rule weight.

[0083] Based on the fuzzy output generated by fuzzy logic control, the adaptive neural fuzzy reasoning unit makes inference selection on the classification path. Combining the advantages of neural network and fuzzy logic, the adaptive neural fuzzy reasoning unit first converts the fuzzy output into a membership function, trains and optimizes the membership function through a multi-layer feedforward neural network, and learns the nonlinear relationship between micro-change information and the optimal classification path.

[0084] After completing the reasoning process of the adaptive neuro-fuzzy inference unit, the optimal classification path is selected according to the output results. The optimal classification path is not only based on the slight changes in the key characteristics of the current sample, but also takes into account historical data and dynamic adjustment strategies, thereby ensuring that the classification path can achieve the highest classification accuracy and stability under various environmental conditions.

[0085] The fuzzy output is converted into a membership function:

[0086] The output of the fuzzy logic control unit, such as the fuzzy membership value μ for each possible classification path (such as path A and path B) pathA and μ pathB , is passed as input to the adaptive neuro-fuzzy inference unit. For each fuzzy output, the adaptive neuro-fuzzy inference unit represents it as a membership function, which describes how the degree of the input variable (such as key feature micro-change information) affects the selection of different classification paths.

[0087] For example, suppose the fuzzy logic output is μ pathA =0.7 and μ pathB =0.4, ANFIS regards these values ​​as the activation values ​​of the input layer nodes, indicating the support degree of the input data for path A and path B.

[0088] The multi-layer feedforward neural network structure includes an input layer, a membership function layer, a rule layer, a normalization layer and an output layer, wherein;

[0089] Input layer: receives the fuzzy output (membership value) of the fuzzy logic control unit;

[0090] Membership function layer: matches the fuzzy output with the preset membership function. The membership function is a Gaussian function, and its formula is:

[0091] Among them, x is the input variable (such as micro-change information), c is the center of the Gaussian function, and σ is the standard deviation. By adjusting the c and σ parameters, the adaptive neural fuzzy inference unit can perform fuzzy processing on the input data;

[0092] Rule layer: Each node corresponds to a fuzzy rule, and the results of the membership function are weighted and summed to represent the support of a specific classification path, for example:

[0093] f i =w i ×μ pathA +(1-w i )×μ pathB , where f i is the output of the i-th rule, w i is the weight, indicating the support of path A;

[0094] Normalization layer: Normalize the output of the regularization layer to ensure that the output results are within a comparable range. Normalization is expressed as: in, is the normalized output;

[0095] Output layer: The final classification path selection is calculated based on the normalized output. The nodes in the output layer generate the final classification path decision through weighted summation:

[0096] Among them, p i is the specific classification path corresponding to each rule, is the normalized output of the normalization layer;

[0097] The adaptive neuro-fuzzy inference unit optimizes the parameters of the neural network through a training process, including the center c and standard deviation σ of the membership function and the weight w of the rule layer. i The goal of training is to minimize the error between the predicted classification path and the actual classification path. The error back propagation algorithm is used to iteratively adjust the parameters in the network to improve the prediction accuracy of the model. The loss function E is expressed as: Among them, y k For the actual path selection result, is the path selection result predicted by the adaptive neural fuzzy inference unit, and M is the number of samples. The error back propagation algorithm uses the output of the loss function to calculate the error gradient, and updates the network weights through this gradient information to make the network prediction closer to the actual value, thereby minimizing the value of the loss function; through the back propagation algorithm, adjust the parameters c, σ and w in the network i By minimizing the loss function, the system can more accurately select the optimal classification path.

[0098] The machine learning algorithm in the multimodal classification module uses a support vector machine. A multimodal classification model is constructed based on the support vector machine and trained based on the characteristic data (metabolite concentration, redox potential, and cell activity) of reproductive test samples in the training data set. The multimodal classification model training includes normalizing the collected characteristic data, extracting features that affect the classification results through feature engineering methods, and using them as input for model training. The influential features include metabolite concentration C met , redox potential Eredox , cell activity A cell ,The preprocessed data is input into the multimodal classification model for training, and the input features and the corresponding specimen category labels in the training dataset are used to optimize the model parameters to minimize the classification error.

[0099] The multi-modal classification model was constructed by the support vector machine (SVM) algorithm to classify the multi-dimensional characteristic data of reproductive test samples. The core of SVM is to find an optimal decision boundary (hyperplane) to effectively classify sample data of different categories;

[0100] The input feature vector of the support vector machine is expressed as: x = [C met ,E redox ,A cell ], where C met is the metabolite concentration, E redox is the redox potential, A cell is the cell activity; the support vector machine maps the input feature vector x to the classification result through the decision function f(x). The support vector machine also includes the classification decision. The classification result of the classification decision is determined by the symbolic function, which is expressed as:

[0101] Class(x)=sign(f(x))=sign(w T x+b);

[0102] When f(x)>0, the sample is classified as a positive class (such as a healthy sample); when f(x)<0, the sample is classified as a negative class (such as an abnormal sample);

[0103] The goal of the support vector machine is to maximize the distance from the decision boundary to the nearest sample point, called the margin, which is defined as: Therefore, the optimization problem is expressed as:

[0104] Constraints: y i (w T x i +b)≥1, where y i For sample x i The label is +1 or -1, indicating positive or negative class.

[0105] The decision function f(x) is expressed as: f(x) = w T x+b, where w is the weight vector, indicating the direction of the hyperplane, and b is the bias, indicating the offset of the hyperplane. The goal is to find the optimal w and b so that sample data of different categories can be correctly classified.

[0106] The dynamic adaptive classification control module determines the features (metabolite concentration, redox potential, cell activity) that have the greatest impact on the classification results and the best classification path by evaluating the micro-change information of the key characteristics of the samples in real time. The relationship between these characteristics and the classification results has been analyzed and optimized in this module. Therefore, the optimal classification path actually represents the features that the model should focus on. When a certain feature has a significant impact on the classification results, adjusting the corresponding feature weights can enable the model to better reflect the impact of this feature on the final classification, thereby improving the accuracy and stability of the classification. Therefore, in the classification process, the multimodal classification model uses the optimal classification path as a guide, classifies based on the path guidance, and adjusts the weights in the decision layer. The decision layer is the decision function. In the decision function f(x), the weight vector w is adjusted according to the optimal path. For example, if the optimal path indicates that the metabolite concentration has a greater impact on the classification results, the model will increase the corresponding C met The weight of is increased to make it more influential in classification decisions, so as to enhance the sensitivity of the model to changes in key features.

[0107] The multi-mode classification model is combined with the time series analysis of dynamic characteristic data. During the classification process, the model parameters are dynamically adjusted when the key characteristics obtained by the time series analysis undergo slight changes. Specifically, the following are included:

[0108] Time series modeling: Build a time series model for each key feature;

[0109] Dynamically adjust model parameters: When time series analysis shows that key features have changed slightly, the multimodal classification model is adaptively updated by adjusting the weight vector w. If the prediction model shows that C met It will increase significantly at a future time point, and the model will increase the metabolite concentration weight w met The weight of t is used to reflect this change: w(t+1)=w(t)+Δw, where Δw is determined based on the forecast results of the time series model;

[0110] By regularly evaluating the prediction effect of the time series model and updating the parameter settings of the multimodal classification model in real time, it is ensured that the classification model can respond promptly and maintain high-precision classification capabilities when the sample characteristics change.

[0111] The time series model adopts an autoregressive model. Time series modeling includes modeling the metabolite concentration characteristic time series data to predict the change of the metabolite concentration characteristic value over time, which is expressed as:

[0112]

[0113] Among them, C met (t+1) is the predicted metabolite concentration at time point t+1;

[0114] C met (t),C met (t-1),…,C met (tp) are the metabolite concentration data at time points t, t-1, …, tp, respectively;

[0115] is the autoregressive coefficient, reflecting the degree of influence of metabolite concentration at each historical time point on future concentration;

[0116] ∈ t is white noise, representing random errors or unpredictable factors in the model.

[0117] The autoregressive model predicts future concentration changes by using historical data on metabolite concentrations. This prediction result can help the SVM model understand the possible changing trends of key characteristics in advance, so as to better adjust the classification decision. Specifically, if the autoregressive model predicts that the metabolite concentration will increase significantly at a certain point in the future, the SVM model can dynamically adjust its decision boundary based on this prediction result to improve the accuracy of classification.

[0118] Autoregressive coefficient is determined by minimizing the prediction error. In practice, these coefficients are obtained by training with historical data and reflect the correlation between feature values ​​at different time points. For example, if is larger, indicating that the current metabolite concentration has a greater impact on the prediction of future concentrations.

[0119] In the support vector machine classification model, slight changes in feature values ​​can affect the classification results. When the autoregressive model predicts the trend of feature value changes, the SVM can respond to these changes by adjusting the weight vector w. For example, if the metabolite concentration is predicted to increase, the SVM will increase the weight of the metabolite concentration feature in the classification decision to make the model more sensitive to this feature change. This process helps improve the classification performance of the model under changing conditions.

[0120] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0121] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. The automated classification system for reproductive test specimens is characterized by: The following steps are involved: Real-time dynamic characteristic monitoring module: Detection devices are used to monitor the key characteristics of reproductive test specimens in real time, including metabolite concentration, redox potential and cell activity. The monitoring data is processed through high-frequency sampling and real-time analysis algorithms to capture micro-change information of key characteristics of the specimens; Dynamic adaptive classification control module: Based on the micro-change information transmitted by the real-time dynamic characteristic monitoring module, the dynamic adaptive classification control module evaluates the micro-change information in real time by integrating the fuzzy logic control unit and the adaptive neural fuzzy reasoning unit, and selects the optimal classification path according to the evaluation results; Multimodal classification module: A multimodal classification model is constructed based on a machine learning algorithm. The multimodal classification model classifies specimens in combination with an optimal classification path. The multimodal classification model also combines time series analysis of dynamic characteristic data. During the classification process, the model parameters are dynamically adjusted when slight changes occur in key characteristics according to the time series analysis.

2. The automated classification system for reproductive test specimens according to claim 1, characterized in that: The detection device includes an electrochemical sensor, a microelectrode sensor and a biosensor. The metabolite concentration is detected by the electrochemical sensor, the redox potential is measured by the microelectrode sensor, the cell activity is evaluated by the biosensor, and the sensor continuously collects data.

3. The automated classification system for reproductive test specimens according to claim 2, characterized in that: The real-time analysis algorithm adopts a dynamic time warping algorithm. During the data analysis process, the dynamic time warping algorithm is used to compare and analyze the sample characteristic curves collected by the detection device, match the data collected at different time points, identify the slight change trends in the sample characteristic curves, and capture slight changes in sample characteristics by comparing the time series differences between the current data and the baseline data, and generate corresponding change signals.

4. The automated classification system for reproductive test specimens according to claim 1, characterized in that: The dynamic adaptive classification control module specifically includes: Micro-change information processing: pre-process micro-change information and standardize key characteristic data to ensure consistency and comparability of data input; The pre-processed micro-change information is preliminarily evaluated by the fuzzy logic control unit, the metabolite concentration, redox potential and cell activity data are input into the fuzzy logic system, the fuzzy set and fuzzy rules of the input data are defined, and the fuzzy output is generated by the fuzzy inference engine, which is provided as an input signal to the adaptive neuro-fuzzy inference unit; Based on the fuzzy output generated by fuzzy logic control, the adaptive neuro-fuzzy reasoning unit selects the classification path and converts the fuzzy output into a membership function. The membership function is trained and optimized through a multi-layer feedforward neural network to learn the nonlinear relationship between micro-change information and the optimal classification path. After completing the reasoning process of the adaptive neuro-fuzzy reasoning unit, the optimal classification path is selected according to the output results; The multi-layer feedforward neural network structure includes an input layer, a membership function layer, a rule layer, a normalization layer and an output layer, wherein; Input layer: receiving the fuzzy output of the fuzzy logic control unit; Membership function layer: matches the fuzzy output with the preset membership function; Rule layer: Each node corresponds to a fuzzy rule, and the results of the membership function are weighted and summed to represent the support of a specific classification path; Normalization layer: Normalize the output of the rule layer to ensure that the output results are within a comparable range; Output layer: The final classification path selection is calculated based on the normalized output. The nodes in the output layer generate the final classification path decision through weighted summation: Among them, p i is the specific classification path corresponding to each rule, is the normalized output of the normalization layer; The adaptive neural fuzzy inference unit optimizes the parameters of the neural network through a training process. The goal of the training is to minimize the error between the predicted classification path and the actual classification path. The error back propagation algorithm is used to iteratively adjust various parameters in the network.

5. The automated classification system for reproductive test specimens according to claim 4, characterized in that: The machine learning algorithm in the multi-modal classification module adopts a support vector machine, and a multi-modal classification model is constructed based on the support vector machine. The training is performed based on the characteristic data of the reproductive test samples in the training data set. The multi-modal classification model training includes normalizing the collected characteristic data, extracting features that affect the classification results through feature engineering methods, and using them as input for model training. The influential features include metabolite concentration C met , redox potential E redox , cell activity A cell ,The preprocessed data is input into the multimodal classification model for training, and the input features and the corresponding specimen category labels in the training dataset are used to optimize the model parameters to minimize the classification error.

6. The automated classification system for reproductive test specimens according to claim 5, characterized in that: The input feature vector of the support vector machine is expressed as: x = [C met ,E redox ,A cell ], where C met is the metabolite concentration, E redox is the redox potential, A cell is cell activity; the support vector machine maps the input feature vector x to the classification result through the decision function f(x), and the support vector machine also includes a classification decision, and the classification result of the classification decision is determined by a symbolic function, which is expressed as: Class(x)=sign(f(x))=sign(w T x+b); When f(x)>0, the sample is classified as positive; when f(x)<0, the sample is classified as negative; The goal of the support vector machine is to maximize the distance from the decision boundary to the nearest sample point, called the margin, which is defined as: Therefore, the optimization problem is expressed as: Subject to constraints: where y i For sample x i The label is +1 or -1, indicating positive or negative class.

7. The automated classification system for reproductive test specimens according to claim 6, characterized in that: The decision function f(x) is expressed as: f(x)=w T x+b, where w is the weight vector, indicating the direction of the hyperplane, and b is the bias, indicating the offset of the hyperplane. The goal is to find the optimal w and b so that sample data of different categories can be correctly classified.

8. The automated classification system for reproductive test specimens according to claim 7, characterized in that: During the classification process, the multi-modal classification model uses the optimal classification path as a guide, performs classification based on the path guidance, and performs weight adjustment in the decision layer. The decision layer is a decision function. In the decision function f(x), the weight vector w is adjusted according to the optimal path.

9. The automated classification system for reproductive test specimens according to claim 1, characterized in that: The multi-mode classification model combines the time series analysis of dynamic characteristic data. During the classification process, the model parameters are dynamically adjusted when the key characteristics obtained by the time series analysis undergo slight changes. Specifically, the model parameters are: Time series modeling: Build a time series model for each key feature; Dynamically adjust model parameters: When time series analysis shows that key features have changed slightly, the multimodal classification model is adaptively updated by adjusting the weight vector w; By regularly evaluating the prediction effect of the time series model and updating the parameter settings of the multimodal classification model in real time, it is ensured that the classification model can respond promptly and maintain high-precision classification capabilities when the sample characteristics change.

10. The automated classification system for reproductive test specimens according to claim 9, characterized in that: The time series model adopts an autoregressive model. The time series modeling includes modeling the metabolite concentration characteristic time series data to predict the change of the metabolite concentration characteristic value over time, which is expressed as: Among them, C met (t+1) is the predicted metabolite concentration at time point t+1; C met (t),C met (t-1),…,C met (tp) are the metabolite concentration data at time points t, t-1, …, tp, respectively; is the autoregressive coefficient, which reflects the degree of influence of metabolite concentration at each historical time point on future concentration; ∈ t is white noise, representing random errors or unpredictable factors in the model.

Images