Intelligent analysis system and method for microbial detection based on federated learning

Abstract

The present invention relates to the field of microbial detection technology, and discloses an intelligent analysis system and method for microbial detection based on federated learning. The system includes: a user instruction parsing module, which is used to parse detection requests through natural language processing and image recognition technology and output structured strain labels; a federated resource library module, whose input end is connected to the output end of the user instruction parsing module through an encrypted communication interface, and is used to store and update strain feature vectors and magnetic bead parameters through the FedAvg algorithm; an intelligent collaborative decision engine module, which is used to call data from the federated resource library module; a hardware execution control module; a federated feedback optimization module, which is used to collect hardware execution data; and a visual report generation module. A distributed strain feature library and magnetic bead parameter pool are constructed through federated learning, and a magnetic bead weight mapping of the strain is dynamically established through a cross-library association unit. The resource library is updated in a closed loop through a lightweight optimization package, which significantly improves the response speed and detection accuracy of new strains.

Description

Technical Field

[0001] The present invention relates to the field of microbial detection technology, and specifically to a microbial detection intelligent analysis system and method based on federated learning. Background Art

[0002] Microbial testing in areas such as food safety and environmental monitoring faces the dual challenges of collaborative processing of multi-source heterogeneous data and privacy protection. On the one hand, testing scenarios are scattered, such as supermarket rapid testing stations, laboratories, and field sites, making it difficult to centrally manage strain characteristic data. On the other hand, magnetic bead performance parameters and sample matrix information need to be updated in real time to respond to new pathogens. The federated learning framework provides a technical basis for building a dynamic strain characteristic library through distributed node local training and parameter encryption aggregation mechanisms. Under the federated learning framework, full-process intelligent testing of instruction parsing, resource optimization, decision execution, and feedback traceability is realized. Related technologies often use centralized databases, such as cloud-based centralized strain libraries, requiring each scenario to upload original test data.

[0003] The use of a centralized database makes it difficult to achieve efficient collaboration and dynamic optimization of data across institutions, and there is a disconnect between hardware execution and decision-making analysis, resulting in delayed responses to new pathogens and insufficient accuracy in joint detection of multiple bacterial species. Summary of the Invention

[0004] In response to the shortcomings of existing technologies, the present invention provides an intelligent analysis system and method for microbial detection based on federated learning, which solves the problems of difficulty in achieving efficient collaboration and dynamic optimization of cross-institutional data, resulting in delayed response to new pathogens and low accuracy in joint detection of multiple strains.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a microbial detection intelligent analysis system based on federated learning, comprising:

[0006] User instruction parsing module, used to parse detection requests through natural language processing and image recognition technology, and output structured bacterial species labels;

[0007] The federated resource library module has an input end connected to the output end of the user instruction parsing module through an encrypted communication interface, and is used to store and update bacterial strain feature vectors and magnetic bead parameters using the FedAvg algorithm;

[0008] The intelligent collaborative decision-making engine module is used to call the data of the federated resource library module and generate the magnetic bead configuration strategy;

[0009] A hardware execution control module is used to convert the magnetic bead configuration strategy into a magnetic bead position selection signal, a volume control signal and a temperature parameter signal;

[0010] The federated feedback optimization module is used to collect hardware execution data, identify abnormal events in the bead recovery rate, and generate lightweight optimization parameters to update the bead priority index of the federated resource library module;

[0011] The visual report generation module is used to integrate the test analysis results with the literature basis to generate a visual report with error tracing.

[0012] Preferably, the user instruction parsing module includes:

[0013] A natural language processing unit is used to parse the text instructions entered by the user and extract keywords of the bacterial species name;

[0014] An image recognition unit, configured to recognize a type label in a sample image by using image recognition technology;

[0015] The structured coding unit has an input end connected to the output end of the natural language processing unit and the image recognition unit, and is used to generate standardized instruction data including bacterial species identifiers, sample categories, and detection parameters.

[0016] Preferably, the federal resource library module includes:

[0017] Feature encoding unit, used to aggregate the strain feature parameters of distributed nodes through horizontal federated learning to generate strain feature vectors;

[0018] A parameter management unit, which is used to store magnetic bead performance data, including antibody type, magnetic response threshold, and optimal pH value, using inverted index technology;

[0019] Cross-library association unit, used to establish the weight mapping relationship between bacterial species feature vectors and magnetic bead parameters ;in, :node The bacterial strain feature vector or magnetic bead parameters, For nodes The amount of data, is the total data volume of the entire network, This is the federal aggregation round.

[0020] Preferably, the intelligent collaborative decision engine module includes:

[0021] Local matching calculation unit, used to calculate the matching score using the following cosine similarity algorithm ;

[0022] in, is the bacterial species feature vector, is the magnetic bead parameter vector;

[0023] Federated aggregate decision-making unit for executing through secure multi-party computing technology , output the global optimal magnetic bead combination solution;

[0024] in, is the number of nodes participating in the calculation; For nodes The weight of Allocation based on historical detection accuracy;

[0025] Multi-strain optimization unit for implementing greedy algorithms ;

[0026] in, is a candidate magnetic bead combination set, is the number of target bacteria, Magnetic beads cross-reaction rate.

[0027] Preferably, the hardware execution control module includes:

[0028] A position selection unit, used for generating storage coordinate codes according to the type of magnetic beads;

[0029] a volume control unit for generating a volume control signal by pulse width modulation;

[0030] The temperature synchronization unit is used to transmit the temperature instruction code through the serial peripheral interface protocol.

[0031] Preferably, the federated feedback optimization module includes:

[0032] An abnormality identification unit, used to mark abnormal characteristics when the magnetic bead recovery rate is less than 90%;

[0033] Distillation compression unit, used to compress abnormal data into a lightweight optimization package with a parameter volume ≤ 10% of the original model;

[0034] The index updating unit is used to update the magnetic bead priority order of the federated resource library module according to the optimization package.

[0035] Preferably, the visual report generating module includes:

[0036] The knowledge graph mapping unit is used to match the abnormal data of the federated feedback optimization module with the literature library and output error solutions;

[0037] The multimodal generation unit is used to integrate raw data, conclusion analysis and literature basis, and output graphic and text reports.

[0038] Preferably, when the parameter management unit detects that the performance of the magnetic beads has declined by more than 15%, it sends a warning mark through the intelligent collaborative decision engine module;

[0039] The early warning flag triggers the federated aggregation decision unit to exclude the magnetic bead solution and select an alternative solution with a suboptimal matching score.

[0040] Preferably, the intelligent collaborative decision engine module further includes a multidimensional feature extraction unit for extracting spatial features of the detection data through a CNN convolutional neural network and extracting time series features in combination with an LSTM long short-term memory network;

[0041] The weight mapping relationship of the federated resource library modules is updated in real time through a dynamic adjustment mechanism.

[0042] The intelligent analysis method for microbial detection based on federated learning includes the following steps:

[0043] S1. Parsing user instructions, parsing detection requests and outputting structured bacterial species labels;

[0044] S2, federated resource management, storing and updating strain feature vectors and magnetic bead parameters through the FedAvg algorithm;

[0045] S3, intelligent collaborative decision-making, generates magnetic bead configuration strategies based on S2 data;

[0046] S4, hardware execution control, converting the magnetic bead configuration strategy into magnetic bead position selection signals, volume control signals and temperature parameter signals;

[0047] S5, federated feedback optimization, collects hardware execution data from S4, identifies abnormal events in magnetic bead recovery rate, and updates the magnetic bead priority index;

[0048] S6. Visual report generation: integrating analysis results with literature references, and outputting a visual report with error traceability.

[0049] The present invention provides a microbial detection intelligent analysis system based on federated learning. It has the following beneficial effects:

[0050] 1. The present invention constructs a distributed strain feature library and magnetic bead parameter pool based on federated learning, dynamically establishes a magnetic bead weight mapping of strains through cross-library association units, and updates the resource library through a lightweight optimization package closed loop, significantly improving the response speed and detection accuracy of new strains.

[0051] 2. The present invention uses CNN to extract the spatial distribution characteristics of magnetic beads, combines it with LSTM to analyze the detection time series characteristics, and integrates spatiotemporal data to generate magnetic bead strategies. It uses a greedy algorithm to compress the cross-reaction rate of multiple bacteria species, and realizes the precise conversion of strategy to equipment through coordinate encoding, PWM signal and SPI protocol, eliminating human errors and ensuring detection reliability.

[0052] 3. The present invention uses a knowledge graph mapping unit to intelligently match the abnormal data fed back by the federation with the cloud document library and output targeted solutions. The multimodal generation unit automatically generates structured graphic reports by integrating the original data, analysis conclusions and document basis, and updates the resource library parameters in combination with the dynamic weight adjustment mechanism to achieve full-process self-driving of anomaly detection, knowledge tracing and closed-loop optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is an architecture diagram of the microbial detection intelligent analysis system based on federated learning of the present invention;

[0054] Figure 2 This is a flow chart of the intelligent analysis method for microbial detection based on federated learning of the present invention. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] Please see the attached Figure 1 , an embodiment of the present invention provides a microbial detection intelligent analysis system based on federated learning, comprising:

[0057] User instruction parsing module, used to parse detection requests through natural language processing and image recognition technology, and output structured bacterial species labels;

[0058] The federated resource library module has an input end connected to the output end of the user instruction parsing module through an encrypted communication interface, and is used to store and update bacterial strain feature vectors and magnetic bead parameters using the FedAvg algorithm;

[0059] The intelligent collaborative decision-making engine module is used to call the data of the federated resource library module and generate the magnetic bead configuration strategy;

[0060] A hardware execution control module is used to convert the magnetic bead configuration strategy into a magnetic bead position selection signal, a volume control signal and a temperature parameter signal;

[0061] The federated feedback optimization module is used to collect hardware execution data, identify abnormal events in the bead recovery rate, and generate lightweight optimization parameters to update the bead priority index of the federated resource library module;

[0062] The visual report generation module is used to integrate the test analysis results with the literature basis to generate a visual report with error tracing.

[0063] Specifically, the user instruction parsing module uses natural language processing technology to parse text instructions (such as "Detect Salmonella in fresh chicken") and extract strain keywords. It also uses image recognition technology to identify type labels (such as "fresh chicken") in sample images. The module ultimately outputs a structured strain label (such as strain ID: Salmonella_0157) containing standardized strain identifiers, sample categories, and test parameters. This shortens the traditional manual data entry process and eliminates transcription errors, providing high-precision input for the subsequent federated resource library module.

[0064] The federated resource library module receives structured strain labels from the user command parsing module through an encrypted communication interface. It then uses the FedAvg algorithm to aggregate local strain feature data (such as surface antigen proteins and metabolic markers) and magnetic bead performance parameters (such as antibody type and magnetic response threshold) from distributed nodes (such as supermarket rapid testing stations and CDCs). While protecting data privacy, it dynamically updates the global strain feature vector library and magnetic bead parameter pool, shortening the storage cycle for new strains (such as drug-resistant E. coli), reducing false positives, and providing real-time and accurate data support for the intelligent decision-making engine.

[0065] The intelligent collaborative decision-making engine module calls upon the strain feature vectors and magnetic bead parameters from the federated resource library module, generates a magnetic bead configuration strategy using a federated feature matching algorithm (cosine similarity calculation) and secure multi-party computing technology, extracts spatiotemporal features from the test data using a convolutional neural network and a long-short-term memory network, and optimizes the magnetic bead combination scheme for multi-strain joint testing using a greedy algorithm. This provides operational instructions to the hardware execution control module, enabling fully automated decision-making throughout the entire process.

[0066] The hardware execution control module converts the magnetic bead configuration strategy generated by the intelligent collaborative decision-making engine into executable instructions for the device. It uses magnetic bead position selection signals to locate the storage bin, volume control signals to accurately quantify the bead volume (e.g., 100 μL of anti-Salmonella magnetic beads), and temperature parameter signals to synchronize with the qPCR temperature control system. This enables automated conversion from strategy to physical operation, eliminating human error and ensuring standardized execution of the testing process.

[0067] The federated feedback optimization module continuously collects hardware execution data (such as the deviation between the actual bead recovery rate and the target recovery rate), identifies recovery rate anomalies (such as insufficient recovery rate due to bead performance degradation), generates lightweight optimization parameters (compresses model size), and dynamically updates the bead priority index of the federated resource library module. This implements closed-loop resource library optimization based on execution feedback, ensuring the accuracy and reliability of subsequent bead configuration strategies.

[0068] The visual report generation module integrates the detection and analysis results of the federated feedback optimization module and calls on authoritative literature in the cloud knowledge base (such as bacterial species identification standards and magnetic bead performance specifications) to generate a structured graphic report that includes error tracing analysis (such as non-specific adsorption correction suggestions), realizing visual traceability of the entire detection process and automatic knowledge empowerment.

[0069] The user instruction parsing module includes:

[0070] A natural language processing unit is used to parse the text instructions entered by the user and extract keywords of the bacterial species name;

[0071] An image recognition unit, configured to recognize a type label in a sample image by using image recognition technology;

[0072] The structured coding unit has an input end connected to the output end of the natural language processing unit and the image recognition unit, and is used to generate standardized instruction data including bacterial species identifiers, sample categories, and detection parameters.

[0073] Specifically, the natural language processing unit parses user-entered text instructions (such as "detect salmonella in chicken"), accurately extracts bacterial species name keywords (such as "Salmonella"), eliminates ambiguities (such as abbreviations and aliases), and matches standardized bacterial species identifiers. This provides structured input without semantic bias for subsequent modules, ensuring the reliability of decision-making across the entire system.

[0074] The image recognition unit processes sample images (such as fresh food labels or microscopic images of microorganisms) to capture the visual features of biological morphology and label text, automatically identifying and classifying sample types (such as "poultry - Salmonella risk level A"), eliminating manual input errors and providing standardized sample category labels for the federal resource library, ensuring the accuracy of subsequent magnetic bead configuration strategies;

[0075] The structured coding unit eliminates multi-source input conflicts (such as the difference between the text instruction "chicken" and the image label "poultry") by fusing the strain keywords extracted by the natural language processing unit and the sample type labels identified by the image recognition unit, and generates standardized instruction data containing a unique strain identifier (such as Salmonella_0157), sample category (fresh poultry) and preset detection parameters (temperature control range 55-60°C), providing the federated resource library with an unambiguous, distributable callable input source, ensuring the accuracy of the collaborative execution of subsequent modules.

[0076] The federated repository modules include:

[0077] Feature encoding unit, used to aggregate the strain feature parameters of distributed nodes through horizontal federated learning to generate strain feature vectors;

[0078] A parameter management unit, which is used to store magnetic bead performance data, including antibody type, magnetic response threshold, and optimal pH value, using inverted index technology;

[0079] The cross-library association unit has an input end connected to the output end of the feature encoding subunit and the parameter management subunit, and is used to establish a weight mapping relationship between the bacterial strain feature vector and the magnetic bead parameters.

[0080] Specifically, the feature encoding unit securely aggregates local strain feature data (such as antigen protein expression profiles) from distributed nodes (such as supermarkets and laboratories) through a horizontal federated learning algorithm, generating a unified strain feature vector while protecting data privacy. This eliminates cross-institutional data barriers in traditional methods and enables the dynamic expansion of new strain features into the global resource library.

[0081] The parameter management unit uses inverted index technology to structuredly store magnetic bead parameters (such as antibody type and magnetic response threshold). It quickly associates multi-dimensional performance data with magnetic bead IDs, optimizing traditional linear searches to millisecond-level queries, ensuring that the decision engine can call accurate parameters in real time.

[0082] Cross-library association unit, used to establish the weight mapping relationship between bacterial species feature vectors and magnetic bead parameters ;in, :node The bacterial strain feature vector or magnetic bead parameters, For nodes The amount of data, is the total data volume of the entire network, This is the federal aggregation round.

[0083] The intelligent collaborative decision-making engine module includes:

[0084] Local matching calculation unit, used to calculate the matching score using the following cosine similarity algorithm ;

[0085] in, is the bacterial species feature vector, is the magnetic bead parameter vector;

[0086] Federated aggregate decision-making unit for executing through secure multi-party computing technology , output the global optimal magnetic bead combination solution;

[0087] in, is the number of nodes participating in the calculation; For nodes The weight of Allocation based on historical detection accuracy;

[0088] Multi-strain optimization unit for implementing greedy algorithms ;

[0089] in, is a candidate magnetic bead combination set, is the number of target bacteria, Magnetic beads cross-reaction rate.

[0090] Specifically, the local matching calculation unit uses the cosine similarity algorithm to calculate the matching score between the bacterial strain feature vector and the magnetic bead parameter vector (such as the vector angle between the surface antigen of E. coli and the anti-E. coli magnetic beads), quantifying the degree of adaptation of a single bacterial strain and providing a preliminary screening basis for federated aggregation decision-making;

[0091] The federated aggregate decision-making unit uses secure multi-party computing technology to fuse the local matching results of each node, dynamically assigns weights based on the node's historical accuracy, and outputs the globally optimal magnetic bead combination solution (such as anti-Salmonella magnetic beads + anti-Listeria magnetic beads), eliminating single-node data bias;

[0092] The multi-strain optimization unit performs a greedy algorithm iteration on the candidate magnetic bead combination set, with the goal of minimizing the cross-reaction rate, and screens out a multi-strain magnetic bead configuration strategy with a total cross-reaction below a safety threshold.

[0093] The hardware execution control module includes:

[0094] A position selection unit, used for generating storage coordinate codes according to the type of magnetic beads;

[0095] a volume control unit for generating a volume control signal by pulse width modulation;

[0096] The temperature synchronization unit is used to transmit the temperature instruction code through the serial peripheral interface protocol.

[0097] Specifically, the position selection unit generates a binary coordinate code based on a preset mapping table between magnetic bead types and physical storage bins, such as anti-Salmonella magnetic beads - bin number N03, to drive the automated equipment to accurately locate the target magnetic bead storage location, eliminating the position deviation of manual selection;

[0098] The volume control unit converts the target volume value into a pulse width modulation signal and accurately controls the amount of magnetic bead reagent extracted by adjusting the stepper motor drive current, replacing traditional manual pipetting operations;

[0099] Based on the Serial Peripheral Interface (SPI) communication protocol, the temperature synchronization unit compiles temperature parameters into instruction code frames that can be parsed by the device, achieving real-time data synchronization with the qPCR temperature control and ensuring accurate execution of the amplification reaction temperature.

[0100] The federated feedback optimization module includes:

[0101] An abnormality identification unit, used to mark abnormal characteristics when the magnetic bead recovery rate is less than 90%;

[0102] Distillation compression unit, used to compress abnormal data into a lightweight optimization package with a parameter volume ≤ 10% of the original model;

[0103] The index update unit is used to update the magnetic bead priority sorting of the federated resource library module according to the optimization package.

[0104] Specifically, the anomaly recognition unit continuously monitors the actual recovery rate data of the magnetic beads in the hardware execution module. When the recovery rate falls below the preset threshold of 90%, it automatically marks abnormal features (such as a surge in nonspecific adsorption) and triggers the optimization process.

[0105] The distillation compression unit compresses the marked abnormal data into lightweight optimized packages through model distillation technology, significantly reducing federated communication overhead;

[0106] The index update unit dynamically adjusts the priority ranking of magnetic beads in the federated resource library based on the lightweight optimization package (high cross-reaction magnetic beads are downgraded), realizing closed-loop iterative optimization of resource library parameters.

[0107] The visual report generation module includes:

[0108] The knowledge graph mapping unit is used to match the abnormal data of the federated feedback optimization module with the literature library and output error solutions;

[0109] The multimodal generation unit is used to integrate raw data, conclusion analysis and literature basis, and output graphic and text reports.

[0110] Specifically, the knowledge graph mapping unit intelligently matches abnormal data (magnetic bead nonspecific adsorption rate exceeds the standard) from the federated feedback optimization module with the cloud-based literature library (including ISO standards and authoritative journal solutions), outputting targeted error correction solutions (such as increasing Tween-20 concentration) to improve traceability efficiency.

[0111] The multimodal generation unit integrates the original detection data (such as positive results of bacterial strains), the error analysis conclusions of federated feedback (such as attribution of recovery rate deviation), and the literature basis provided by the knowledge graph mapping unit to automatically generate a structured graphic report containing data charts, conclusion summaries and solutions.

[0112] When the parameter management unit detects that the performance of the magnetic beads has declined by more than 15%, it sends a warning mark through the intelligent collaborative decision-making engine module;

[0113] The early warning flag triggers the federated aggregation decision unit to exclude the magnetic bead solution and select an alternative solution with a suboptimal matching score.

[0114] Specifically, when the parameter management unit detects that the performance of the magnetic beads has declined by more than 15%, it sends a warning mark to the intelligent collaborative decision-making engine module to identify the failure risk magnetic beads in real time, preventing false negative results caused by performance degradation in traditional testing;

[0115] At the same time, the early warning mark triggers the federal aggregation decision-making unit to actively exclude high-risk magnetic bead solutions, and based on secure multi-party computing technology, screen out alternative magnetic bead combinations with suboptimal matching scores from the candidate pool, ensuring that the detection process seamlessly switches to backup solutions when magnetic bead performance is abnormal, maintaining system robustness.

[0116] The intelligent collaborative decision-making engine module also includes a multi-dimensional feature extraction unit, which is used to extract the spatial features of the detection data through the CNN convolutional neural network and extract the time series features in combination with the LSTM long short-term memory network;

[0117] The weight mapping relationship of the federated resource library modules is updated in real time through a dynamic adjustment mechanism.

[0118] Specifically, the multidimensional feature extraction unit uses the CNN convolutional neural network to analyze the spatial distribution characteristics of magnetic beads (such as the aggregation morphology of magnetic bead-microorganism complexes in microfluidic chips), and combines it with the LSTM long short-term memory network to capture the temporal dynamic characteristics of the qPCR amplification curve (such as the impact of temperature jumps on fluorescence signals), extracting spatiotemporal fusion features from the detection data, and solving the technical defect of traditional single-dimensional analysis that ignores spatiotemporal correlation.

[0119] The dynamic adjustment mechanism updates the weight mapping relationship of the federated resource library module in real time according to the feature extraction results, establishes dynamic adaptation rules for strain-magnetic bead parameters, and eliminates matching deviations caused by changes in the sample matrix (such as meat fat interference).

[0120] Please see the attached Figure 2 ,The intelligent analysis method for microbial detection based on federated learning includes the following steps:

[0121] S1. Parsing user instructions, parsing detection requests and outputting structured bacterial species labels;

[0122] S2, federated resource management, storing and updating strain feature vectors and magnetic bead parameters through the FedAvg algorithm;

[0123] S3, intelligent collaborative decision-making, generates magnetic bead configuration strategies based on S2 data;

[0124] S4, hardware execution control, converting the magnetic bead configuration strategy into magnetic bead position selection signals, volume control signals and temperature parameter signals;

[0125] S5, federated feedback optimization, collects hardware execution data from S4, identifies abnormal events in magnetic bead recovery rate, and updates the magnetic bead priority index;

[0126] S6. Visual report generation: integrating analysis results with literature references, and outputting a visual report with error traceability.

[0127] Specifically, structured strain labels are generated by parsing user instructions, and distributed node data are dynamically integrated using a federated learning algorithm to construct a strain feature and magnetic bead parameter resource library. Deep learning feature extraction technology is called to generate magnetic bead configuration strategies, and the strategies are converted into positioning, volume, and temperature control signals that can be executed by the device to achieve precise operation. Based on hardware execution feedback, recovery rate anomalies are identified and the resource library index is optimized in a closed loop. Finally, authoritative literature is integrated to generate a visual report with error traceability. Under the premise of data privacy protection, a full-process automated detection chain is constructed to solve the data island problem and eliminate manual operation errors.

[0128] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The microbial detection intelligent analysis system based on federated learning is characterized by: include: User instruction parsing module, used to parse detection requests through natural language processing and image recognition technology, and output structured bacterial species labels; The user instruction parsing module includes: A natural language processing unit is used to parse the text instructions entered by the user and extract keywords of the bacterial species name; An image recognition unit, configured to recognize a type label in a sample image by using image recognition technology; a structured coding unit, whose input end is connected to the output end of the natural language processing unit and the image recognition unit, and is used to generate standardized instruction data including bacterial species identifiers, sample categories, and detection parameters; The federated resource library module has an input end connected to the output end of the user instruction parsing module through an encrypted communication interface, and is used to store and update bacterial strain feature vectors and magnetic bead parameters using the FedAvg algorithm; The federal resource library module includes: Feature encoding unit, used to aggregate the strain feature parameters of distributed nodes through horizontal federated learning to generate strain feature vectors; A parameter management unit, which is used to store magnetic bead performance data, including antibody type, magnetic response threshold, and optimal pH value, using inverted index technology; Cross-library association unit, used to establish the weight mapping relationship between bacterial species feature vectors and magnetic bead parameters ;in, :node The bacterial strain feature vector or magnetic bead parameters, For nodes The amount of data, is the total data volume of the entire network, for the federal aggregation round; The intelligent collaborative decision-making engine module is used to call the data of the federated resource library module and generate the magnetic bead configuration strategy; The intelligent collaborative decision engine module includes: Local matching calculation unit, used to calculate the matching score using the following cosine similarity algorithm ; in, is the bacterial species feature vector, is the magnetic bead parameter vector; Federated aggregate decision-making unit for executing through secure multi-party computing technology , output the global optimal magnetic bead combination solution; in, is the number of nodes participating in the calculation; For nodes The weight of Allocation based on historical detection accuracy; Multi-strain optimization unit for implementing greedy algorithms ; in, is a candidate magnetic bead combination set, is the number of target bacterial species, Magnetic beads Cross-reactivity rate; The intelligent collaborative decision engine module also includes a multidimensional feature extraction unit for extracting spatial features of detection data through a CNN convolutional neural network and extracting time series features in combination with an LSTM long short-term memory network; The weight mapping relationship of the federated resource library modules is updated in real time through a dynamic adjustment mechanism; A hardware execution control module is used to convert the magnetic bead configuration strategy into a magnetic bead position selection signal, a volume control signal and a temperature parameter signal; The federated feedback optimization module is used to collect hardware execution data, identify abnormal events in the bead recovery rate, and generate lightweight optimization parameters to update the bead priority index of the federated resource library module; The visual report generation module is used to integrate the test analysis results with the literature basis to generate a visual report with error tracing.

2. The microbial detection intelligent analysis system based on federated learning according to claim 1 is characterized in that: The hardware execution control module includes: A position selection unit, used for generating storage coordinate codes according to the type of magnetic beads; a volume control unit for generating a volume control signal by pulse width modulation; The temperature synchronization unit is used to transmit the temperature instruction code through the serial peripheral interface protocol.

3. The microbial detection intelligent analysis system based on federated learning according to claim 1 is characterized in that: The federated feedback optimization module includes: An abnormality identification unit, used to mark abnormal characteristics when the magnetic bead recovery rate is less than 90%; Distillation compression unit, used to compress abnormal data into a lightweight optimization package with a parameter volume ≤ 10% of the original model; The index updating unit is used to update the magnetic bead priority order of the federated resource library module according to the optimization package.

4. The microbial detection intelligent analysis system based on federated learning according to claim 1 is characterized in that: The visual report generation module includes: The knowledge graph mapping unit is used to match the abnormal data of the federated feedback optimization module with the literature library and output error solutions; The multimodal generation unit is used to integrate raw data, conclusion analysis and literature basis, and output graphic and text reports.

5. The microbial detection intelligent analysis system based on federated learning according to claim 1 is characterized in that: When the parameter management unit detects that the performance of the magnetic beads has declined by more than 15%, it sends a warning mark through the intelligent collaborative decision engine module; The early warning mark triggers the federated aggregation decision unit to exclude the output global optimal magnetic bead combination solution and select an alternative solution with a suboptimal matching score.

6. An intelligent analysis method for microbial detection based on federated learning, characterized in that: The microbial detection intelligent analysis system based on federated learning according to any one of claims 1 to 5 comprises the following steps: S1. Parsing user instructions, parsing detection requests and outputting structured bacterial species labels; S2, federated resource management, storing and updating strain feature vectors and magnetic bead parameters through the FedAvg algorithm; S3, intelligent collaborative decision-making, generates magnetic bead configuration strategies based on S2 data; S4, hardware execution control, converting the magnetic bead configuration strategy into magnetic bead position selection signals, volume control signals and temperature parameter signals; S5, federated feedback optimization, collects hardware execution data from S4, identifies abnormal events in magnetic bead recovery rate, and updates the magnetic bead priority index; S6. Visual report generation: integrating analysis results with literature references, and outputting a visual report with error traceability.

Images