Doxycycline hydrochloride rapid analysis system and method based on automatic detection

Abstract

The invention provides a doxycycline hydrochloride rapid analysis system and method based on automatic detection, and the system comprises an automatic sample injection module which is used for automatically mixing a to-be-detected sample and a fluorescent probe according to a preset proportion, and injecting the mixture into a microfluidic reaction path; the multi-channel signal acquisition module is used for synchronously acquiring multi-mode detection signals such as fluorescence, scattered light and temperature; the fusion model module is used for performing feature extraction and fusion analysis on the acquired multi-channel signals and generating an optimal adjustment instruction; the closed-loop execution module is used for automatically adjusting detection parameters based on model output and realizing dynamic closed-loop control in the detection process; and the result output module is used for outputting a doxycycline hydrochloride concentration judgment result and can be used for collecting user feedback to assist the iterative optimization of the model. The method has the advantages of high automation degree, high detection speed, high result accuracy and strong anti-interference capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chemical detection, and more particularly to a doxycycline hydrochloride rapid analysis system and method based on automatic detection. BACKGROUND

[0002] Doxycycline hydrochloride is a broad-spectrum tetracycline antibiotic widely used in human medical treatment and livestock breeding. Its residue problem poses potential risks to food safety and ecological environment, so it is of great practical significance to develop rapid, accurate and efficient doxycycline hydrochloride detection technology.

[0003] At present, the detection methods for doxycycline hydrochloride mainly include high performance liquid chromatography, liquid chromatography-mass spectrometry and other methods. Although these methods have high sensitivity and accuracy, they generally have the following limitations:

[0004] First, the pretreatment process is complex and time-consuming, requiring professional operators, which is difficult to meet the needs of on-site rapid detection;

[0005] Second, the instrument equipment is large and expensive, with high running costs, which is not conducive to popularization and promotion.

[0006] In order to meet the needs of rapid detection, fluorescence sensing method is introduced. Among them, the detection method based on fluorescence probe shows great potential due to its high sensitivity. However, the existing fluorescence detection method still faces many challenges:

[0007] (1) Poor anti-interference ability: the actual sample matrix is complex, and factors such as background fluorescence and turbidity will seriously interfere with the detection signal, leading to deviation in the measurement results. Traditional methods usually rely on a single fluorescence signal, and lack effective correction means for non-specific adsorption and environmental noise.

[0008] (2) Detection process is rigid: existing automatic equipment usually uses preset fixed parameters for detection, and cannot dynamically adjust according to changes in reaction progress, sample characteristics or environmental temperature. When encountering low reaction efficiency, signal saturation or poor signal-to-noise ratio, the system cannot be optimized automatically, affecting the reliability and adaptability of the detection.

[0009] In summary, there is an urgent need in the art for a technical solution that can overcome the above-mentioned shortcomings, i.e. to realize the automatic and intelligent rapid analysis of doxycycline hydrochloride with strong anti-interference ability and self-adaptive optimization function under the premise of ensuring high sensitivity and high accuracy. SUMMARY

[0010] The purpose of the present application is to provide a doxycycline hydrochloride rapid analysis system and method based on automatic detection, which realizes the full-process automation and intelligentization from sample introduction to result output through the synergistic effect of automatic sampling module, multi-channel signal acquisition module, fusion model module, closed-loop execution module and result output module, and has the advantages of realizing multi-modal optical signal fusion perception, guaranteeing dynamic closed-loop optimization of the detection process, improving system detection precision, anti-interference ability and self-adaptive learning ability.

[0011] In a first aspect, the present application provides a doxycycline hydrochloride rapid analysis system based on automatic detection, comprising:

[0012] An automatic sampling module is used to automatically mix the sample to be tested and the fluorescent probe in a predetermined ratio, and inject the mixed liquid into the microfluidic reaction path of the multi-channel signal acquisition module;

[0013] A multi-channel signal acquisition module is an integrated multifunctional microfluidic optical detector, which integrates a micro-reaction cavity and multiple optical detection interfaces, and is used to acquire fluorescence signals, scattered light signals and temperature information;

[0014] A fusion model module is signal connected with the multi-channel signal acquisition module, used to receive multi-channel detection data and construct a feature vector, input into the fusion model for analysis and decision-making, to generate optimal adjustment instructions;

[0015] A closed-loop execution module is signal connected with the fusion model module, used to automatically adjust the detection parameters based on the optimal adjustment instructions, to realize dynamic closed-loop control of the detection process.

[0016] A result output module is signal connected with the fusion model module, used to output the concentration determination result of the fusion model, and can receive user feedback to assist model correction or training.

[0017] Further, the present application provides that the multi-channel signal acquisition module is an integrated microfluidic optical detector, and the detection area of the micro-reaction cavity is respectively provided with a fluorescence detection port, a scattered light detection port and a temperature detection port, and is respectively connected to a corresponding fluorescence detection submodule, a scattered light detection submodule and a temperature monitoring submodule, for simultaneously acquiring different types of optical and thermal detection signals.

[0018] Further, the present application provides that the closed-loop execution module comprises:

[0019] A fluid control submodule is signal connected with the fusion model module, used to adjust the instantaneous flow ratio of the sample and the fluorescent probe, to control the reaction mixing efficiency;

[0020] The constant temperature control submodule is in signal connection with the fusion model module and is used for adjusting a constant temperature setting value of the micro-reaction cavity to ensure temperature control consistency of a reaction process.

[0021] The integration time adjustment submodule is arranged in the fluorescence detection submodule in the multi-channel signal acquisition module and is used for adjusting fluorescence acquisition integration time according to fluorescence signal intensity feedback to optimize a signal-to-noise ratio.

[0022] Specifically, the integration time adjustment submodule is configured to:

[0023] The fluid control submodule is in signal connection with the fusion model module and is used for:

[0024] receiving a reaction optimization instruction output by the fusion model module;

[0025] dynamically adjusting a transient flow ratio of the sample and the fluorescent probe;

[0026] and feeding back an actual execution value to the fusion model module in real time to form a control closed loop to correct errors or drifts.

[0027] The constant temperature control submodule is configured to:

[0028] receiving a temperature setting value output by the fusion model module;

[0029] controlling a constant temperature interval in the micro-reaction cavity;

[0030] and collecting temperature stability data and thermal drift data, which are sent back to the fusion model module for dynamic parameter adjustment after being compared with a setting label.

[0031] The integration time adjustment submodule is integrated in the fluorescence detection submodule and is configured to:

[0032] according to a signal quality score result provided by the fusion model module;

[0033] adjusting integration time in a fluorescence acquisition process in real time;

[0034] and binding a collection number, integration time setting and detection result as a data baseline for subsequent model training and abnormal tracing.

[0035] Further, the fusion model module includes:

[0036] a feature extraction submodule, which is used for constructing a multi-dimensional feature vector and a label thereof based on fluorescence signals, scattering signals and temperature data acquired by the multi-channel signal acquisition module;

[0037] a model decision submodule, which is used for calling a trained fusion decision model to analyze and calculate the feature vector to generate a decision instruction for detecting parameter regulation;

[0038] The model updating sub-module is configured to dynamically update or optimize the weights of the fusion model according to user calibration information or historical detection results.

[0039] Further, the application provides that the fusion decision model is an ensemble learning model or a deep neural network model, and a nonlinear response relationship between heterogeneous signals is constructed based on multi-source feature data, for predicting a detection sensitivity change trend and automatically generating a control strategy.

[0040] Further, the application provides that the feature extraction sub-module is configured to:

[0041] constructing multi-modal fusion features based on fluorescence signals, scattering light signals and temperature signals;

[0042] The fusion features include a fluorescence-to-scattering light intensity ratio, a fluorescence time gradient change rate, a temperature dynamic correction factor and a signal-to-noise variance ratio.

[0043] A unique feature label is constructed for each fusion feature to realize full-process traceability of the cross-channel fusion features.

[0044] Further, the application provides that the fusion model module further includes a decision feedback sub-module configured to:

[0045] receive the fusion feature vector and its label from the feature extraction sub-module;

[0046] dynamically infer the current reaction state based on historical modeling rules and current environmental variables;

[0047] and generate parameter feedback instructions according to the model decision results, which are sent to the fluid control sub-module, the constant temperature control sub-module and the integral time adjustment sub-module, respectively, for performing closed-loop adjustment operations;

[0048] The feedback instructions and feature labels correspond one-to-one, ensuring that the execution process has reverse checking capability.

[0049] Specifically, the fusion decision model is trained to:

[0050] when the ratio of fluorescence and scattering light intensity exceeds a preset first threshold value, and the time gradient change rate of the fluorescence signal is lower than a second threshold value,

[0051] The output control strategy is to preferentially increase the reaction temperature and reduce the sample flow rate to cope with the interference caused by reaction lag and turbidity. The first threshold value and the second threshold value can be dynamically adjusted according to the sample type.

[0052] Further, the application provides that the data management sub-module is configured to:

[0053] The original fluorescence signal, the fusion feature vector, the model decision parameter, and the actual execution feedback value are collected from the fluorescence detection sub-module, the feature extraction sub-module, the fusion model module, and the closed-loop execution module, respectively.

[0054] The above data are associated and bound one by one to establish a data link indexed by the detection number.

[0055] A data visualization query interface and a traceability interface are provided to support the traceability of the sampling path, the response decision, and the final detection conclusion through any detection number.

[0056] The bound data can be used for model optimization training, abnormality traceability analysis, and closed-loop correction to ensure the interpretability and traceability of the system operation.

[0057] Further, the application further comprises a model training submodule configured to:

[0058] The bound historical detection number, the original input signal, the fusion feature vector, and the model output result are called from the data management submodule to construct a training sample set.

[0059] Based on a supervised learning or semi-supervised learning strategy, the model parameters in the fusion model module are iteratively updated.

[0060] After the model training and verification are completed, a new model version number is generated and pushed to the fusion model module to ensure that the online inference process has the latest modeling capability and verifiability.

[0061] Specifically, the model training submodule is further included to support the continuous learning and capability enhancement of the fusion model module. The model training submodule calls the historical detection data stored in the data management submodule to iteratively update the fusion model parameters based on a supervised learning or semi-supervised learning strategy, and generates a model version number. The updated model weight is pushed to the fusion model module to replace the current model, thereby improving the system's adaptability to sample complexity and reaction state changes, and realizing model dynamic updating and verifiable closed loop.

[0062] Further, the application further comprises a model training submodule configured to:

[0063] The fusion feature vector,

[0064] The composite control instruction combination,

[0065] And the signal-to-noise ratio of the detection terminal reached after the execution of the instruction;

[0066] The model training submodule takes the signal-to-noise ratio as a supervised learning optimization target to iteratively update the parameters of the fusion decision model.

[0067] In a second aspect, the application provides a rapid analysis method of doxycycline hydrochloride based on automatic detection, comprising the following steps:

[0068] A sample injection step: injecting a sample containing doxycycline hydrochloride into a microfluidic channel through an automatic injection module;

[0069] A signal acquisition step: synchronously acquiring fluorescence signals, scattering light signals and environmental temperature signals of the target reaction chamber;

[0070] A feature extraction step: constructing a fusion feature vector based on the fluorescence signals, scattering light signals and temperature signals, the features including a fluorescence-to-scattering light intensity ratio, a time gradient change rate of the fluorescence signals, a correction factor based on a temperature change trend, and a signal-to-noise variance ratio;

[0071] A state discrimination step: inputting the fusion feature vector into a trained fusion model to infer the current reaction state;

[0072] A feedback control step: generating multi-parameter feedback control instructions such as temperature adjustment, excitation light intensity adjustment and sample flow rate adjustment according to the inference result;

[0073] A detection output step: extracting a final doxycycline hydrochloride detection value based on the state adjusted by the closed-loop control, and generating a detection number.

[0074] Further, the application provides that the fusion model used in the state discrimination step is an integrated neural network model, which is trained based on historical detection numbers and labels and has cross-channel feature fusion capability.

[0075] Further, the application provides that in the feature extraction step, when the fluorescence-to-scattering light intensity ratio in the fusion feature vector exceeds a first threshold and the fluorescence time gradient change rate is lower than a second threshold, the model preferentially generates a combination of temperature rising and flow rate decreasing instructions to deal with detection lag and background interference.

[0076] Further, the application provides that the feedback control step comprises:

[0077] Generating parameter control instructions and binding the current feature label;

[0078] Returning the instructions to the control module and synchronously recording them in the data traceability interface.

[0079] Further, the application provides that in the detection output step, each detection number is bound with the following fields: original signal sampling data, fusion features, control response parameters, final detection value and model version number, which can be retrieved through the unique detection number.

[0080] Further, the application further includes a model training step, the training step includes:

[0081] extracting a historical detection sample set from the data management module, the sample set including a fusion feature vector, a corresponding composite control instruction and a final signal-to-noise ratio after execution of the composite control instruction;

[0082] using a supervised or semi-supervised learning algorithm to optimize the fusion model with the final signal-to-noise ratio as the objective function;

[0083] after completing the model training, a new model version number is pushed for calling by the inference module.

[0084] Compared with the prior art, the application has the following beneficial effects:

[0085] Firstly, the application realizes synchronous acquisition and fusion analysis of multi-modal optical signals, and improves detection accuracy and anti-interference capability. The application synchronously acquires fluorescence signals, scattered light signals and temperature information through an integrated microfluidic optical detector, constructs a multi-dimensional feature vector including a fluorescence-to-scattered light intensity ratio, a fluorescence time gradient change rate, a temperature dynamic correction factor and a signal-to-noise variance ratio, and inputs the fusion model for decision-making. Compared with the traditional single signal detection method, multi-source signal fusion can effectively distinguish target reactions from background noise, compensate for environmental disturbances such as turbidity and temperature drift through a nonlinear response relationship, and significantly improve the accuracy and repeatability of doxycycline hydrochloride detection.

[0086] Secondly, the application realizes dynamic closed-loop adjustment of detection parameters, and optimizes reaction efficiency and signal-to-noise ratio. Based on the decision instruction output by the fusion model, the application adjusts the instantaneous flow ratio of the sample and the fluorescent probe through the fluid control submodule, maintains the temperature stability of the micro-reaction cavity through the constant temperature control submodule, and adaptively optimizes the fluorescence acquisition integration time through the integration time adjustment submodule, forming a real-time feedback control closed loop. The traditional method mainly uses fixed parameters, which is difficult to deal with sample variation or reaction lag; the system adjusts multiple parameters to ensure uniform reaction mixing, consistent temperature control and optimal signal-to-noise ratio, and improves detection sensitivity and robustness.

[0087] Thirdly, the application realizes continuous learning and adaptive updating of the intelligent fusion model, and enhances the adaptability of the system to complex samples. The application introduces a model training submodule, uses the historical detection data, fusion feature vectors and execution feedback bound in the data management submodule to iteratively optimize the fusion decision model parameters with a supervised learning or semi-supervised learning strategy. The model can dynamically adjust the weight according to the user calibration information or new samples, avoid model degradation caused by changes in reaction conditions, and realize long-term stable operation and prediction accuracy improvement.

[0088] Fourthly, the full-process data traceability and explainable decision are realized to guarantee the verifiability of the detection process. The original signal, the fusion feature vector, the model decision parameter and the actual execution value are bound by the data management submodule with the detection number as the index to establish an end-to-end data link. The user can trace back the sampling path, the response decision and the final conclusion of any detection through the visual query interface, support abnormal traceability and model correction, and solve the pain point of the opaque decision process of the traditional black box system.

[0089] Fifthly, the system has good engineering adaptability and expansion capability due to the modular design and standardized integration. The boundaries of the modules of the application are clear, the interfaces are uniform, the automatic sampling, multi-channel signal acquisition and closed-loop execution modules can be flexibly connected to the existing microfluidic platform or detection equipment, the on-demand deployment and function expansion are supported, the hardware does not need to be greatly modified, and the application threshold and popularization cost are reduced. BRIEF DESCRIPTION OF DRAWINGS

[0090] Figure 1 FIG. 1 is a schematic diagram of the automatic detection-based doxycycline hydrochloride rapid analysis system of the application.

[0091] Figure 2 FIG. 2 is a schematic diagram of the automatic detection-based doxycycline hydrochloride rapid analysis method of the application.

[0092] The following specific embodiments will further illustrate the application in combination with the above drawings. DETAILED DESCRIPTION

[0093] Embodiment:

[0094] In the prior art, the detection of doxycycline hydrochloride mainly relies on traditional laboratory detection technologies such as high performance liquid chromatography and ultraviolet spectrophotometry. Although these methods have high accuracy and stability, the detection process usually involves complex sample pretreatment, manual operation and long reaction time, which seriously restricts their practicality and popularization in on-site detection, rapid screening and other application scenarios.

[0095] To improve the detection efficiency, most schemes still use a single signal channel (such as fluorescence intensity) for analysis and judgment, and fail to realize the cooperative processing of multi-dimensional detection data. In addition, the existing microfluidic detection system generally adopts an open-loop control strategy, i.e., the detection parameters are preset and the running process is not adjustable, which leads to significant fluctuations in system sensitivity and stability when facing different sample backgrounds, environmental temperature changes or excitation light source drift.

[0096] Especially in the aspect of automation processing, the traditional scheme usually adopts a fixed proportion of sample and reagent mixing mode, lacks intelligent adjustment ability, and the sample injection process is mostly manual or semi-automatic, which is easy to introduce errors. For data processing, the mainstream scheme usually extracts a single feature for comparison and discrimination, lacks the extraction and fusion of complex features such as signal-to-noise ratio and gradient change trend, and is difficult to cope with the weak signal recognition problem under low concentration background.

[0097] In order to solve the above problems, the inventors realize that a rapid analysis architecture should be constructed, which fuses multi-modal detection signals and intelligent closed-loop control. On the one hand, an automatic sampling module should be introduced to realize the automatic mixing and accurate injection of reagents and samples in proportion; on the other hand, an optical module with multi-channel acquisition capability should be designed to support the synchronous acquisition of fluorescence, scattered light and temperature signals, and a fusion feature vector should be constructed, including fluorescence scattering ratio, time gradient, temperature correction factor and other parameters. Further, combined with the trained fusion discrimination model, intelligent judgment of the reaction state can be realized, and control instructions can be generated in real time for feedback adjustment of key parameters such as flow rate, light intensity and temperature, thereby significantly improving the detection accuracy, stability and system adaptability.

[0098] As shown in Figure 1 , the present application proposes a rapid analysis system for doxycycline hydrochloride based on automatic detection, which comprises:

[0099] An automatic sampling module is used to automatically mix the sample to be tested and the fluorescence probe in a predetermined proportion, and inject the mixed liquid into the microfluidic reaction path of the multi-channel signal acquisition module;

[0100] A multi-channel signal acquisition module is an integrated multifunctional microfluidic optical detector, which integrates a micro-reaction cavity and multiple optical detection interfaces for acquiring fluorescence signals, scattered light signals and temperature information;

[0101] A fusion model module is connected in signal with the multi-channel signal acquisition module, used to receive multi-channel detection data and construct a feature vector, input into the fusion model for analysis and decision-making to generate optimal adjustment instructions;

[0102] A closed-loop execution module is connected in signal with the fusion model module, used to automatically adjust the detection parameters based on the optimal adjustment instructions to realize dynamic closed-loop control of the detection process.

[0103] A result output module is connected in signal with the fusion model module, used to output the concentration judgment result of the fusion model, and can receive user feedback to assist model correction or training.

[0104] The application realizes automatic mixing of the to-be-tested sample and the fluorescent probe and accurate injection into a reaction path by setting an automatic sampling module, effectively avoids problems such as unstable mixing ratio and large injection error of traditional manual mixing, improves the repeatability and automation level of sampling, and meets the scene requirements of high-throughput and online detection.

[0105] Further, the application provides that the multi-channel signal acquisition module is an integrated microfluidic optical detector, detection areas of micro-reaction cavities of the integrated microfluidic optical detector are respectively provided with fluorescence detection ports, scattered light detection ports and temperature detection ports, and are respectively connected to corresponding fluorescence detection sub-modules, scattered light detection sub-modules and temperature monitoring sub-modules, so as to simultaneously acquire different types of optical and thermal detection signals.

[0106] The application integrates the three types of detection ports of fluorescence, scattered light and temperature in the integrated microfluidic optical detector, and establishes accurate corresponding connection with the sub-modules, solves the problems of poor synchronism and response mismatch caused by physical separation of signal channels in the prior art, and enhances the real-time performance of signal acquisition and the fusion analysis capability between channels.

[0107] Further, the application provides that the closed-loop execution module comprises:

[0108] A fluid control sub-module is in signal connection with the fusion model module and is used for adjusting a transient flow ratio of the sample and the fluorescent probe, so as to control a reaction mixing efficiency.

[0109] A constant temperature control sub-module is in signal connection with the fusion model module and is used for adjusting a constant temperature setting value of the micro-reaction cavity, so as to ensure temperature control consistency of a reaction process.

[0110] An integration time adjustment sub-module is arranged in the fluorescence detection sub-module in the multi-channel signal acquisition module and is used for adjusting a fluorescence acquisition integration time according to fluorescence signal intensity feedback, so as to optimize a signal-to-noise ratio.

[0111] Specifically, the fluid control sub-module is in signal connection with the fusion model module and is used for:

[0112] receiving a reaction optimization instruction output by the fusion model module;

[0113] receiving a reaction optimization instruction output by the fusion model module;

[0114] dynamically adjusting a transient flow ratio of the sample and the fluorescent probe;

[0115] and feeding back an actual execution value to the fusion model module in real time to form a control closed loop to correct errors or drifts.

[0116] The constant temperature control sub-module is used for:

[0117] receiving a temperature setting value output by the fusion model module;

[0118] controlling the constant temperature interval in the micro-reaction cavity;

[0119] and collecting temperature stability data and thermal drift data, and after comparing with the set label, sending back to the fusion model module for dynamic parameter adjustment.

[0120] The integral time adjustment submodule is integrated in the fluorescence detection submodule, and is configured to:

[0121] according to the signal quality score result provided by the fusion model module;

[0122] adjusting the integral time in real time during the fluorescence acquisition process;

[0123] and binding the acquisition number, integral time setting and detection result as the data baseline for subsequent model training and abnormal tracing.

[0124] The present application solves the problems of low efficiency of manual parameter adjustment and unstable reaction conditions in the traditional detection process by setting the execution submodule including fluid control, constant temperature control and integral time adjustment, and realizing linkage adjustment by the composite instruction output by the fusion model.

[0125] Further, the present application proposes that the fusion model module comprises:

[0126] a feature extraction submodule for constructing a multi-dimensional feature vector and its label based on the fluorescence signal, scattering signal and temperature data obtained by the multi-channel signal acquisition module;

[0127] a model decision submodule for calling the trained fusion decision model to analyze and calculate the feature vector, and generating a decision instruction for detecting parameter regulation;

[0128] a model updating submodule for dynamically updating or weight optimizing the fusion model according to user calibration information or historical detection results.

[0129] The execution submodule not only receives the adjustment instruction of the fusion model, but also can real-time feedback the execution result to the fusion model module, constructs a feedback closed loop, realizes error real-time correction and parameter adjustment adaptive function.

[0130] Further, the present application proposes that the fusion decision model is an ensemble learning model or a deep neural network model, and a nonlinear response relationship between heterogeneous signals is constructed based on multi-source feature data, which is used to predict the detection sensitivity change trend and automatically generate a control strategy.

[0131] The integral time adjustment submodule is integrated in the fluorescence detection submodule, which can adjust the integral time in real time according to the score result of the signal quality of the fusion model, solve the problem of overexposure or underexposure caused by fixed parameters in the traditional system, and realize dynamic optimization of signal-to-noise ratio.

[0132] Further, the application proposes that the feature extraction submodule is configured to:

[0133] constructing multi-modal fusion features based on fluorescence signals, scattering light signals and temperature signals;

[0134] The fusion features include fluorescence and scattering light intensity ratio, fluorescence time gradient change rate, temperature dynamic correction factor and signal-to-noise variance ratio.

[0135] A unique feature label is constructed for each fusion feature to realize full-process traceability of cross-channel fusion features.

[0136] Further, the application proposes that the fusion model module further includes a decision feedback submodule configured to:

[0137] receive fusion feature vectors and their labels from the feature extraction submodule;

[0138] Infer the current reaction state based on historical modeling rules and current environmental variables;

[0139] and generate parameter feedback instructions according to the model decision results, which are sent to the fluid control submodule, the constant temperature control submodule and the integral time adjustment submodule respectively to perform closed-loop regulation operations;

[0140] The feedback instructions and feature labels correspond one-to-one to ensure that the execution process has reverse checking capability.

[0141] Specifically, the fusion decision model is trained to:

[0142] When the ratio of fluorescence and scattering light intensity exceeds a preset first threshold value, and the time gradient change rate of the fluorescence signal is lower than a second threshold value,

[0143] The output control strategy is to preferentially increase the reaction temperature and reduce the sample flow rate to cope with the interference caused by reaction lag and turbidity. The first threshold value and the second threshold value can be dynamically adjusted according to the sample type.

[0144] The application constructs multi-modal fusion features including fluorescence intensity, scattering light ratio, gradient change and temperature correction through the feature extraction submodule, and attaches a unique label for identification, effectively improving the model's ability to distinguish complex reaction states, and enhancing the overall recognition accuracy and data traceability of the system.

[0145] Further, the application proposes that the data management submodule is configured to:

[0146] The original fluorescence signal, the fusion feature vector, the model decision parameter and the actual execution feedback value are collected from the fluorescence detection sub-module, the feature extraction sub-module, the fusion model module and the closed-loop execution module respectively;

[0147] The above data are associated and bound one by one to establish a data link indexed by detection number;

[0148] A data visualization query interface and a traceability interface are provided to support the reverse query of the sampling path, response decision and final detection conclusion through any detection number;

[0149] The bound data can be used for model optimization training, abnormality traceability analysis and closed-loop correction to ensure the interpretability and traceability of system operation.

[0150] The fusion model module further introduces a decision feedback sub-module, so that the model not only generates instructions, but also makes judgments based on historical modeling experience and current state, and accurately distributes the results to the control end, realizing the complete closed loop of intelligent perception, judgment, control and verification. Compared with the traditional model which only outputs concentration value, it has significant intelligent adjustment ability.

[0151] Further, the application further comprises a model training sub-module configured to:

[0152] The bound historical detection number, original input signal, fusion feature vector and model output result are called from the data management sub-module to construct a training sample set;

[0153] Based on the supervised learning or semi-supervised learning strategy, the model parameters in the fusion model module are iteratively updated;

[0154] After completing the model training and verification, a new model version number is generated and pushed to the fusion model module to ensure that the online inference process has the latest modeling capability and verifiability.

[0155] Specifically, the model training sub-module is further included for supporting continuous learning and capability enhancement of the fusion model module. The model training sub-module iteratively updates the fusion model parameters through the historical detection data stored in the data management sub-module by a supervised learning or semi-supervised learning strategy, and generates a model version number. The updated model weight is pushed to the fusion model module for replacing the current model, thereby improving the adaptability of the system to sample complexity and reaction state changes, and realizing model dynamic updating and verifiable closed loop.

[0156] The model training sub-module is introduced to realize the supervised or semi-supervised learning mechanism based on historical detection data, support continuous updating and version replacement of the fusion model parameters, break through the technical bottleneck of "static modeling and inability to optimize" of traditional detection systems, and build an online learning framework with self-evolution ability.

[0157] Further, the application provides the training sample set provided by the data management and traceability module, and the sample set includes:

[0158] the fusion feature vector,

[0159] the composite control instruction combination,

[0160] and the detection terminal signal-to-noise ratio reached after the execution of the instruction;

[0161] The model training submodule takes the signal-to-noise ratio as a supervised learning optimization target, and is used for iteratively updating the parameters of the fusion decision model. The training sample takes the final detection signal-to-noise ratio as an optimization target, highly binds the control strategy and the detection performance, realizes that the model not only pursues the concentration fitting accuracy, but also takes into account the detection quality effectiveness, and has stronger engineering usability and target orientation.

[0162] As shown in Figure 2 The application provides a doxycycline hydrochloride rapid analysis method based on automatic detection, which includes the following steps:

[0163] A sample injection step: injecting a sample containing doxycycline hydrochloride into a microfluidic channel through an automatic sampling module;

[0164] A signal acquisition step: synchronously acquiring a fluorescence signal, a scattering light signal and an environmental temperature signal of the target reaction chamber;

[0165] A feature extraction step: constructing a fusion feature vector based on the fluorescence signal, the scattering light signal and the temperature signal, and the features include a fluorescence-to-scattering light intensity ratio, a time gradient change rate of the fluorescence signal, a correction factor based on a temperature change trend, and a signal-to-noise variance ratio;

[0166] A state discrimination step: inputting the fusion feature vector into a trained fusion model to infer the current reaction state;

[0167] A feedback control step: generating a multi-parameter feedback control instruction such as temperature adjustment, excitation light intensity adjustment and sample flow rate adjustment according to the inference result;

[0168] A detection output step: extracting a final doxycycline hydrochloride detection value according to the state adjusted by the closed-loop control, and generating a detection number.

[0169] Further, the application provides that the fusion model used in the state discrimination step is an integrated neural network model, which is obtained based on historical detection numbers and labels, and has a cross-channel feature fusion capability.

[0170] Further, the application proposes that in the feature extraction step, when the fluorescence and scattering light intensity ratio in the fusion feature vector exceeds the first threshold value and the fluorescence time gradient change rate is lower than the second threshold value, the model generates the instruction combination of temperature rise and flow reduction in priority to cope with detection delay and background interference.

[0171] Further, the application proposes that the feedback control step includes:

[0172] Generating parameter control instructions and binding the current feature label;

[0173] Returning the instructions to the control module and synchronously recording in the data traceability interface.

[0174] Further, the application proposes that in the detection output step, each detection number is bound with the following fields: original signal sampling data, fusion features, control response parameters, final detection value and model version number, and the fields can be retrieved through the unique detection number.

[0175] Further, the application proposes that it further includes a model training step, and the training step includes:

[0176] Extracting a historical detection sample set from the data management module, the sample set including fusion feature vectors, corresponding composite control instructions and final signal-to-noise ratios after execution of the instructions;

[0177] Using a supervised or semi-supervised learning algorithm to optimize the fusion model with the final signal-to-noise ratio as the objective function;

[0178] After completing the model training, a new model version number is pushed for calling by the inference module.

[0179] The above-described embodiments only express several embodiments of the application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent of the application. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the application, a number of modifications and improvements can be made, which are all within the protection scope of the application. Therefore, the protection scope of the patent of the application should be subject to the appended claims.

Claims

1. A rapid analysis system for doxycycline hydrochloride based on automated detection, characterized by, The system comprises: an automatic sampling module for automatically mixing the sample to be tested with the fluorescent probe at a preset ratio and injecting the mixed solution into the microfluidic reaction path of the multi-channel signal acquisition module; a multi-channel signal acquisition module, which is an integrated multifunctional microfluidic optical detector, integrates a micro-reaction chamber and multiple optical detection interfaces, and is used for acquiring fluorescence signals, scattered light signals and temperature information; a fusion model module connected to the multi-channel signal acquisition module for receiving multi-channel detection data and constructing a feature vector, inputting the feature vector into a fusion model for analysis and decision-making to generate optimal adjustment instructions; a closed-loop execution module connected to the fusion model module for automatically adjusting detection parameters based on the optimal adjustment instructions to realize dynamic closed-loop control of the detection process. a result output module connected to the fusion model module for outputting the concentration determination result of the fusion model and receiving user feedback to assist model correction or training.

2. The automated detection-based rapid analysis system for doxycycline hydrochloride according to claim 1, characterized in that, The multi-channel signal acquisition module is an integrated microfluidic optical detector, and the detection area of the micro-reaction chamber is respectively provided with a fluorescence detection port, a scattered light detection port and a temperature detection port, and is respectively connected to a fluorescence detection submodule, a scattered light detection submodule and a temperature monitoring submodule for simultaneously acquiring different types of optical and thermal detection signals.

3. The automatic detection-based doxycycline hydrochloride rapid analysis system according to claim 1 or 2, characterized in that the closed-loop execution module comprises: a fluid control submodule connected to the fusion model module for adjusting the instantaneous flow ratio of the sample and the fluorescent probe to control the reaction mixing efficiency; a constant temperature control submodule connected to the fusion model module for adjusting the constant temperature setting value of the micro-reaction chamber to ensure the temperature control consistency of the reaction process; an integration time adjustment submodule arranged in the fluorescence detection submodule of the multi-channel signal acquisition module for adjusting the fluorescence acquisition integration time according to the fluorescence signal intensity feedback to optimize the signal-to-noise ratio.

4. The automated detection-based rapid analysis system for doxycycline hydrochloride according to any one of claims 1 to 3, characterized in that, The fusion model module comprises: a feature extraction submodule for constructing a multi-dimensional feature vector and its label based on the fluorescence signal, scattered signal and temperature data acquired by the multi-channel signal acquisition module; a model decision submodule for calling a trained fusion decision model to analyze and calculate the feature vector to generate a decision instruction for detecting parameter regulation; a model updating submodule for dynamically updating or optimizing the weight of the fusion model according to user calibration information or historical detection results.

5. The automated detection-based rapid analysis system for doxycycline hydrochloride according to any one of claims 1 to 4, characterized in that, The fusion decision model is an ensemble learning model or a deep neural network model, and is used for predicting the detection sensitivity trend and automatically generating a control strategy based on the nonlinear response relationship between the heterogeneous signals constructed based on the multi-source feature data.

6. The automated detection-based rapid analysis system for doxycycline hydrochloride according to claim 4, wherein The feature extraction submodule is configured to: construct multi-modal fusion features based on fluorescence signals, scattered light signals and temperature signals; the fusion features include fluorescence and scattered light intensity ratio, fluorescence time gradient change rate, temperature dynamic correction factor and signal-to-noise variance ratio; a unique feature label is constructed for each fusion feature to realize full-process traceability of the cross-channel fusion features.

7. The automated detection-based rapid analysis system for doxycycline hydrochloride according to claim 4, characterized in that, The fusion model module further comprises a decision feedback submodule configured to: receive the fusion feature vector from the feature extraction submodule and its label; inference the current reaction state based on historical modeling rules and current environmental variables; and generate parameter feedback instructions according to the model decision results, which are sent to the fluid control submodule, thermostat control submodule and integral time adjustment submodule respectively to perform closed-loop regulation operation; The feedback instructions correspond one-to-one with the feature labels, ensuring that the execution process has reverse checking capability; The fusion decision model is trained to: When the ratio of fluorescence intensity to scattering light intensity exceeds a preset first threshold value, and the time gradient change rate of the fluorescence signal is lower than a second threshold value, The output control strategy is to preferentially increase the reaction temperature and reduce the sample flow rate to cope with the interference caused by reaction lag and turbidity. Wherein, the first threshold value and the second threshold value can be dynamically adjusted according to the sample type.

8. The automated detection-based rapid analysis system for doxycycline hydrochloride according to claim 4, characterized in that, The data management submodule is configured to: Collect raw fluorescence signals, fusion feature vectors, model decision parameters and actual execution feedback values from the fluorescence detection submodule, feature extraction submodule, fusion model module and closed-loop execution module respectively; One-to-one association and binding of the above data is performed to establish a data link indexed by detection number; Provide data visualization query interface and traceability interface to support the reverse query of sample path, response decision and final detection conclusion through any detection number; The bound data can be used for model optimization training, abnormal traceability analysis and closed-loop correction to ensure the explainability and traceability of system operation.

9. The automated detection-based rapid analysis system for doxycycline hydrochloride according to claim 4, wherein Further comprising a model training submodule configured to: Call the bound historical detection number, original input signal, fusion feature vector and model output result from the data management submodule to construct a training sample set; Iteratively update the model parameters in the fusion model module based on supervised learning or semi-supervised learning strategy; After completing model training and verification, generate a new model version number and push it to the fusion model module to ensure that the online inference process has the latest modeling capability and verifiability.

10. A rapid analysis method of doxycycline hydrochloride based on automated detection, characterized by, The method comprises the following steps: Sample injection step: under the control of the automatic sampling module, mix the sample to be tested with the fluorescence probe at a preset ratio and inject it into the microfluidic channel; Signal acquisition step: simultaneously acquire the fluorescence signal, scattering light signal and environmental temperature signal of the target reaction chamber; Feature extraction step: based on the fluorescence signal, scattering light signal and temperature signal, construct a fusion feature vector, which includes the ratio of fluorescence intensity to scattering light intensity, the time gradient change rate of the fluorescence signal, the correction factor based on the temperature change trend, and the signal-to-noise variance ratio; State discrimination step: input the fusion feature vector into the trained fusion model to infer the current reaction state; Feedback control step: generate temperature adjustment, excitation light intensity adjustment and sample flow rate adjustment instructions according to the inference results; Detection output step: extract the final doxycycline hydrochloride detection value according to the adjusted state of the closed-loop control, and generate a detection number.

Images