Intelligent substance in-situ extracting and sampling method and system

Abstract

The invention relates to an intelligent substance in-situ extraction sampling method and system. The system comprises a multi-dimensional feature fusion depth profile prediction module, a physical characteristic boundary identification module, a waste liquid-effective sample dynamic identification module and an integrated intelligent decision module. The multi-dimensional feature fusion depth profile prediction module establishes an internal structure virtual model of a target substance through multi-modal scanning; the physical characteristic boundary recognition module monitors physical characteristic changes in the microneedle puncture process in real time and dynamically recognizes different material level boundaries; the waste liquid-effective sample dynamic identification module dynamically judges the effectiveness of the sample based on real-time component analysis; and the integrated intelligent decision module integrates data of the three modules to realize closed-loop feedback and a self-optimization sampling strategy. According to the method, multi-level accurate directional sampling in the substance can be realized, a sampling strategy is dynamically adjusted through substance characteristics, the sample utilization rate is maximized based on intelligent sample identification of components, and adaptive optimization is realized by adopting a deep learning network structure.

Description

Technical Field

[0001] The present invention belongs to the technical fields of precise sampling, intelligent detection, and microfluidics, and relates to an intelligent in-situ extraction and sampling method and system for substances. Background Art

[0002] In-situ extraction and sampling of substances, as an essential basic link in many fields such as medical diagnosis, environmental monitoring, and materials science research, is of self-evident importance. This technology is directly related to the accuracy and reliability of subsequent analysis and is a crucial step in scientific research exploration and practical applications. Although traditional substance sampling methods have met the basic needs to a certain extent, with the progress of science and technology and the in-depth of research, their limitations have become increasingly prominent.

[0003] Traditional sampling methods mainly focus on fixed-depth puncture and preset waste liquid volume, and can be roughly divided into three categories: fixed-depth sampling method, vision-guided sampling method, and experience-based sampling method. The fixed-depth sampling method has been widely used to a certain extent due to its simplicity of operation. However, this method ignores the individual differences between substances and the complexity of internal structures. As a result, the sampling position often deviates from the ideal point, the effective sample volume is insufficient, and even unnecessary damage may be caused to the sample due to too deep puncture. Obviously, this method cannot meet the requirements of high-precision and high-efficiency sampling. The vision-guided sampling method attempts to guide the sampling process through the human eye or auxiliary vision equipment and locates based on the surface characteristics of the substance. However, this method also has its limitations and knows little about the internal structural changes of the substance. When facing substances with complex structures or non-uniformity, such as multi-layer tissues, composite materials, etc., it is often difficult for vision guidance to accurately capture the best sampling area, resulting in the sampling result deviating from the expectation. The experience-based sampling method completely relies on the subjective experience and intuition of the operator. Although this method reflects human wisdom and flexibility to a certain extent, its repeatability and standardization are greatly reduced. Different operators may obtain completely different sampling results due to differences in experience and skill levels, which undoubtedly brings great uncertainty to subsequent analysis and research.

[0004] In recent years, with the rapid development of sensing technology, microfluidics technology, and artificial intelligence technology, intelligent sampling methods have gradually emerged and become a research hotspot. These methods can be roughly divided into two categories: image-guided and sensor-feedback-based. The image-guided method establishes a three-dimensional model of the target substance through pre-scanning, providing more intuitive guidance for sampling. However, this method has insufficient perception of the real-time changes in substance characteristics and is difficult to accurately capture the dynamic changes of the substance.

[0005] On the other hand, the method based on sensing feedback can real-time sense parameters such as resistance and temperature during the puncture process, providing more refined control for sampling. However, this method often only focuses on a single physical property and lacks comprehensive judgment of multi-dimensional physical properties. In addition, there are also obvious deficiencies in the definition of waste liquid and effective samples in the existing technology. Currently, it mainly relies on fixed volume or time thresholds for determination, rather than dynamically adjusting according to the actual component characteristics of the sample.

[0006] In summary, in the related fields, the existing technology has the following four major problems: First, there is a lack of accurate identification and hierarchical sampling ability for the multi-level structure inside the substance; second, the perception dimension of substance characteristics is single, making it difficult to cope with the challenges of complex heterogeneous substances; third, the determination of waste liquid and effective samples relies too much on preset parameters, while ignoring the differences in actual component characteristics; fourth, there is a lack of an intelligent decision-making system for collaborative optimization of each link, resulting in the lack of integrity and systematicness in the sampling process. These problems not only affect the accuracy of sampling and the representativeness of samples, but also cause waste of effective components and excessive sampling damage, seriously restricting the research progress and application effects in related fields. Therefore, exploring and developing new intelligent sampling methods has become an urgent task. Summary of the Invention

[0007] In view of this, the purpose of the present invention is to provide an intelligent in-situ extraction and sampling method and system for substances. This solution realizes the visualization of the internal structure of the target substance based on multi-dimensional feature fusion depth profile prediction, uses physical property gradient monitoring to accurately identify the hierarchical boundary, abandons the concept of fixed waste liquid volume through a waste liquid-effective sample dynamic recognition algorithm, integrates an intelligent decision-making model to achieve closed-loop feedback and self-optimizing sampling strategies, and adopts a deep learning network structure to achieve adaptive optimization, greatly improving the sampling efficiency and sample representativeness while ensuring the sampling accuracy.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] An intelligent in-situ extraction and sampling method for substances, which specifically includes the following steps:

[0010] S1: Establish a multi-dimensional feature fusion depth profile prediction module, obtain the internal structure information of the target substance through multi-modal scanning, and construct a virtual model of the internal structure of the target substance based on this information to predict the optimal sampling depth, path, and multi-level sampling positions;

[0011] S2: Construct a physical property boundary recognition module, monitor the physical property parameters in real-time during the micro-needle puncture process of sampling, calculate the physical property gradient, dynamically identify the boundaries of different substance levels, and determine the optimal sampling positions;

[0012] S3: Construct a waste liquid - effective sample dynamic recognition module, import the collected liquid into the microfluidic analysis unit of this dynamic recognition module to detect the liquid composition in real - time, calculate the sample effectiveness index, automatically determine the conversion point between waste liquid and effective sample, and determine the optimal sampling time and duration;

[0013] S4: Construct an integrated intelligent decision - making module, integrate the output data of the above - mentioned three modules, realize the adaptive adjustment of system parameters, and cooperate to control the whole process of microneedle puncture, layer recognition, sample collection and system withdrawal, and complete the intelligent in - situ extraction and sampling of substances.

[0014] Furthermore, in step S1, using the multi - dimensional feature fusion depth profile prediction module, obtain the internal structure information of the target substance in different dimensions through multiple modalities such as ultrasonic scanning, optical imaging and impedance measurement; apply the depth prediction model to integrate the multi - modal information, and construct a virtual internal model of the target substance according to the information; the depth prediction model uses the following equation:

[0015]

[0016] where x represents the horizontal coordinate position on the plane of the target substance, y represents the vertical coordinate position on the plane of the target substance, D(x, y) is the predicted optimal sampling depth, F i (x, y) is the feature function of different modalities, w i is the weight coefficient of each feature, β is the gradient adjustment factor, is the physical property gradient function. By determining the (x, y) coordinates, the system locates the precise puncture point position on the surface of the target substance, and the D(x, y) function calculates the optimal depth of puncture in the depth direction at the position (x, y). Fi(x, y) represents the feature information obtained by different modality scans (such as ultrasonic, optical imaging, impedance measurement, etc.) at the position (x, y), represents the physical property gradient at the position (x, y).

[0017] Determine the multi - level sampling positions based on the predicted depth:

[0018] L i = D(x, y)×α i +Δi

[0019] where L i is the sampling position of the i - th layer, α i is the relative depth coefficient, and Δi is the adaptive adjustment amount.

[0020] Furthermore, in step S2, the microneedle system in the physical property boundary recognition module integrates multiple sensors such as mechanics, electricity, acoustics, optics and thermotics to form a set of physical property parameters P j(d); wherein the set of physical property parameters Pj specifically includes: mechanical property parameters (puncture resistance, tissue hardness, shear stress, and viscoelastic coefficient); electrical property parameters (electrical impedance, dielectric constant, capacitance value, and bioelectric potential); acoustic property parameters (acoustic impedance, echo intensity, acoustic wave attenuation coefficient, and sound velocity change); optical property parameters (light reflectivity, light scattering coefficient, light absorption coefficient, and fluorescence intensity); and thermal property parameters (thermal conductivity, heat capacity, and temperature gradient).

[0021] The calculation physical property gradient monitoring equation is:

[0022] G(d) = ∑[λ j × (dP j / dd)]

[0023] wherein, G(d) is the comprehensive physical property gradient at depth d, P j is the jth physical property parameter, and λ j is the characteristic weight coefficient;

[0024] Apply the boundary recognition function B(d) = {1, if |G(d)| > τ(d); 0, otherwise} to identify the hierarchical boundary, where τ(d) is the depth-dependent dynamic threshold function and B(d) is the boundary indication function;

[0025] Determine the sampling optimization position: Sopt = argmax{Q(d)|d ∈ [d1, d2]}, where Q(d) is the sampling quality evaluation function and [d1, d2] is the target layer depth interval.

[0026] Furthermore, in step S3, in the waste liquid-effective sample dynamic recognition module: Introduce the collected liquid into the microfluidic analysis unit, and detect the liquid components in real time through the micro-spectroscopic and electrochemical sensor arrays to obtain the target component concentration Ck(t) and the interferent concentration Im(t);

[0027] Calculate the sample validity determination equation as follows:

[0028] E(t) = ∑[μ k × Ck(t)] - ∑[ν m × Im(t)], where E(t) is the sample validity index at time t, and μ k and ν m are the weight coefficients of the target component and the interferent respectively;

[0029] Apply the switching decision function: S(t) = {1, if E(t) > θ(t) and dE(t) / dt > 0; 0, otherwise}, where θ(t) is the dynamic threshold function and S(t) is the sampling switching function;

[0030] Determine the optimal sampling duration: Topt = min{t|dE(t) / dt<εand t>tstart}, where ε is the gradient threshold and tstart is the effective sampling start time.

[0031] Further, in step S4, the integrated intelligent decision module integrates the output data of the three modules and applies the system integrated decision function:

[0032] D(t)=w1·F1(Pd(t),Bd(t))+w2·F2(Sd(t),E(t))+w3·F3(H(t))

[0033] Among them, D(t) is the comprehensive decision output, F1, F2, and F3 are the decision functions of the three modules, Pd(t) is the prediction depth index, Bd(t) is the boundary recognition index, Sd(t) is the sample validity index, E(t) is the sampling efficiency index, H(t) is the historical experience index, and w1, w2, and w3 are dynamic weight coefficients;

[0034] Implement adaptive update of sampling strategy: Among them, θ(t) is the system parameter set, J(θ) is the sampling quality evaluation function, and η is the learning rate.

[0035] Furthermore, the multi-dimensional feature fusion depth profile prediction module, physical property boundary recognition module, waste liquid-effective sample dynamic recognition module and integrated intelligent decision-making module are all implemented using a deep learning network structure. The system continuously optimizes parameters through a self-learning algorithm to improve adaptability to different types of substances and sampling accuracy.

[0036] The present invention also provides an intelligent material in-situ extraction sampling system, which includes a multi-dimensional feature fusion depth profile prediction module, a physical property boundary recognition module, a waste liquid-effective sample dynamic recognition module and an integrated intelligent decision-making module; the multi-dimensional feature fusion depth profile prediction module is used to obtain the internal structure information of the target material through multi-modal scanning, and build a virtual model of the target material based on the information, and predict the optimal sampling depth, path and multi-level sampling position; the physical property boundary recognition module is used to monitor the physical property parameters in real time during the microneedle puncture process, calculate the physical property gradient, dynamically identify the boundaries of different material levels, and determine the optimal sampling position; the waste liquid-effective sample dynamic recognition module is used to introduce the collected liquid into the microfluidic analysis unit of the dynamic recognition module to detect the liquid composition in real time, calculate the sample validity index, automatically determine the conversion point between the waste liquid and the effective sample, and determine the optimal sampling time and duration; the integrated intelligent decision-making module is used to integrate the output data of the above three modules, realize the adaptive adjustment of system parameters, and coordinately control the entire process of microneedle puncture, layer recognition, sample collection and system withdrawal to complete the intelligent material in-situ extraction sampling.

[0037] Furthermore, the system further includes:

[0038] Segmented microneedle system: The segmented microneedle system includes microneedle structures independently controlled in multiple segments, each segment equipped with an independent sampling channel, a sensor array, and an actuator, to achieve synchronous or sequential sampling at multiple depth levels;

[0039] Sample collection and identification system: The sample collection and identification system is used for automatically marking, separating, storing, and processing samples at different depths.

[0040] Furthermore, the segmented microneedle system is the core hardware for realizing multi-level sampling, consisting of microneedle structures independently controlled in multiple segments, each segment equipped with an independent sampling channel, a sensor array, and an actuator, and specifically includes:

[0041] The segmented microneedle system adopts a modular design, and different segments can be freely combined to adapt to the requirements of different application scenarios; each segment of the microneedle contains 4 core components: a precision advance and retreat driver, a sensor array, a sampling channel, and an independent control unit; the precision advance and retreat driver adopts piezoelectric ceramic drive technology to achieve displacement control with micron-level precision; the sensor array adopts various micro sensors such as mechanical, electrical, acoustic, and optical sensors to achieve multi-dimensional monitoring of physical properties; the sampling channel adopts a multi-channel design with independent flow path control to ensure that samples at different layers do not mix; the independent control unit is responsible for local data processing and drive control to reduce the burden on the central processing system;

[0042] The surface of the microneedle is treated with a special coating, having characteristics of drag reduction, anti-adhesion, and anti-pollution, reducing puncture damage and improving sampling efficiency. The tip design adopts bionics principle, simulating the structure of mosquito mouthparts, to achieve low-damage puncture and efficient liquid collection.

[0043] Furthermore, the sample collection and identification system is responsible for automatically marking, separating, storing, and processing samples collected at different depths, and specifically includes:

[0044] Sample collection unit, including a micro peristaltic pump, a multi-channel switching valve, and a sample storage bin, capable of precisely controlling sample flow and separation;

[0045] Automatic marking system, adopting RFID technology, attaching an electronic tag to each sample container to record information such as sampling time, depth, physical property data, and validity index, to ensure sample traceability;

[0046] The sample preservation module automatically adjusts environmental parameters such as temperature, humidity, and oxygen content according to different sample characteristics to ensure sample stability; for special samples, the system can also perform preprocessing operations such as dilution, concentration, filtration, or chemical fixation to improve the efficiency of subsequent analysis.

[0047] The system integration analysis function conducts preliminary detection and evaluation on the collected samples, generates a sample quality report, and assists users in judging the sampling effect and subsequent analysis strategies.

[0048] The beneficial effects of the present invention are as follows:

[0049] 1) Break through the limitations of traditional fixed-depth sampling, achieve multi-level precise directional sampling inside substances, and greatly improve sampling accuracy and layer resolution. Through multi-dimensional feature fusion depth profile prediction, it is possible to accurately plan the sampling path and depth before puncture, reducing blind operations.

[0050] 2) Dynamically adjust the sampling strategy based on the physical properties of substances rather than preset parameters, improving the system's adaptability. The physical property boundary recognition algorithm can real-time monitor changes in multiple physical parameters, accurately identify the boundaries of different substance layers, and ensure that the sampling is located in the target area.

[0051] 3) Abandon the concept of fixed waste liquid volume, achieve intelligent sample recognition based on composition, and maximize sample utilization. The waste liquid-effective sample dynamic recognition algorithm accurately determines the optimal sampling timing by real-time analyzing the liquid composition, improving sample purity and representativeness.

[0052] 4) The integrated decision-making model enables the system to be self-adaptive and self-learning, continuously optimizing the sampling effect. The system can automatically adjust the parameters of each algorithm based on historical experience and real-time feedback, continuously improving the sampling performance.

[0053] 5) Adopt a deep learning network structure to implement the module functions, continuously optimize the parameters through self-learning algorithms, improve the adaptability to different substance types and sampling accuracy, and achieve a significant improvement in the degree of intelligence.

[0054] 6) The integration of the segmented microneedle system and the sample collection and identification system realizes multi-level synchronous or sequential sampling, automatic sample marking and separation storage, and greatly improves work efficiency and sample management level.

[0055] 7) The multi-mode configuration design makes the system highly applicable, significantly improving the accuracy and representativeness of in-situ extraction of various substances. The system architecture is flexible, the parameters are adjustable, and it can adapt to the sampling requirements in different fields such as medicine, environment, and materials.

[0056] Other advantages, objectives, and features of the present invention will be elaborated to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following specification. Brief Description of the Drawings

[0057] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail and preferably below in conjunction with the accompanying drawings, where:

[0058] Figure 1 is the overall architecture diagram of the intelligent in-situ extraction and sampling system of the present invention;

[0059] Figure 2 is the structural diagram of the multi-dimensional feature fusion depth profile prediction module of the present invention;

[0060] Figure 3 is the working principle diagram of the physical property boundary recognition module of the present invention;

[0061] Figure 4 is the working principle diagram of the waste liquid-effective sample dynamic recognition module of the present invention;

[0062] Figure 5 is the architecture diagram of the integrated intelligent decision-making module of the present invention;

[0063] Figure 6 is the mode switching flowchart of the system of the present invention under different application scenarios. Detailed implementation manners

[0064] The technical solutions of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0065] Figure 1 is the overall architecture diagram of the intelligent in-situ extraction and sampling system of the present invention. As Figure 1 shown, this embodiment provides an intelligent in-situ extraction and sampling system, which includes six main parts: a multi-dimensional feature fusion depth profile prediction module, a physical property boundary recognition module, a waste liquid-effective sample dynamic recognition module, an integrated intelligent decision-making module, a segmented microneedle system, and a sample collection and identification system. Each part works together to achieve intelligent in-situ extraction and sampling of substances. In this embodiment, it specifically includes:

[0066] I. Multi-dimensional feature fusion depth profile prediction module

[0067] As Figure 2 shown, the multi-dimensional feature fusion depth profile prediction module constructs a virtual model of the internal structure of the target substance by integrating multi-modal scanning data, and realizes the prediction of the optimal sampling depth, path, and multi-level sampling positions. Specifically, it includes:

[0068] First, the target substance is non-invasively scanned using multiple modalities such as ultrasonic scanning, optical imaging, and impedance measurement to obtain the substance structure information F i (x,y) in different dimensions. Among them, ultrasonic scanning provides acoustic impedance information, optical imaging provides surface and near-surface structure information, and impedance measurement provides electrical property distribution information.

[0069] Then, apply the depth prediction model to integrate multi-modal information. The core equation of this model is:

[0070]

[0071] where D(x, y) is the predicted optimal sampling depth, w i are the weight coefficients of each feature (automatically adjusted by the deep learning network), β is the gradient adjustment factor, is the physical property gradient function. This model can fuse multi-dimensional information according to the weight distribution of different modal features, and at the same time consider the change trend of physical properties to predict the optimal sampling depth.

[0072] Furthermore, determine the multi-level sampling positions according to the predicted depth:

[0073] L i = D(x, y) × α i + Δ i

[0074] where L i is the sampling position of the i-th layer, α i is the relative depth coefficient (0 < α i < 1, and ∑α i = 1), Δ i is the adaptive adjustment amount based on real-time feedback. Through this formula, the system can automatically plan multiple sampling layers to achieve multi-level sampling in one puncture.

[0075] This module uses a hybrid architecture of convolutional neural network (CNN) and recurrent neural network (RNN) to achieve multi-modal data fusion and depth prediction. An attention mechanism is introduced into the network to enhance the recognition ability of key structural features. The model gradually improves the prediction accuracy for different material types through pre-training and online fine-tuning.

[0076] II. Physical Property Boundary Recognition Module

[0077] As Figure 3 shown, the physical property boundary recognition module dynamically recognizes the boundaries of different material layers and accurately locates the optimal sampling area by real-time monitoring the changes in physical properties during the micro-needle puncture process. Specifically, it includes:

[0078] First, the micro-needle system integrates multiple sensors of mechanics, electricity, acoustics, and optics to form a set of physical property parameters P j (d), and real-time monitors the physical properties at each depth d during the puncture process. After the sensor data is preprocessed and feature extracted, it is input into the physical property analysis unit.

[0079] Then, calculate the physical property gradient monitoring equation:

[0080] G(d) = ∑[λ j ×(dP j / dd)]

[0081] where G(d) is the gradient of the comprehensive physical properties at depth d, P j is the physical property parameter of the j-th type, and λ j is the characteristic weight coefficient (dynamically adjusted through a self-learning algorithm according to different material types). This equation can sensitively capture the change rate of physical properties and serve as an indicator of the boundary between different material levels.

[0082] Furthermore, a boundary recognition function is applied to identify the hierarchical boundary:

[0083] B(d) = {1, if |G(d)| > τ(d) {0, otherwise}

[0084] where τ(d) is a depth-dependent dynamic threshold function that is automatically adjusted according to the penetration depth and historical data, and B(d) is the boundary indicator function. A value of 1 indicates that a boundary is detected.

[0085] Finally, based on the boundary recognition result, the optimal sampling position is determined:

[0086] Sopt = argmax{Q(d)|d ∈ [d1, d2]}

[0087] where Q(d) is the sampling quality evaluation function, which comprehensively considers the predicted value of the target substance concentration, uniformity, and representativeness, and [d1, d2] is the depth interval of the target layer determined based on boundary recognition. Through this formula, the system can find the optimal sampling point in the identified layer.

[0088] This module adopts a deep learning architecture based on the long short-term memory network (LSTM) to achieve real-time analysis of sensor time-series data and boundary recognition. The system continuously accumulates experience during use through a self-supervised learning mechanism to optimize the boundary recognition threshold and the sampling position evaluation function.

[0089] III. Waste Liquid - Effective Sample Dynamic Recognition Module

[0090] As Figure 4 shown, the waste liquid - effective sample dynamic recognition module abandons the traditional concept of fixed waste liquid volume and dynamically determines the effectiveness of the sample based on real-time composition analysis. Specifically, during implementation:

[0091] First, the microfluidic system immediately introduces the collected liquid into the analysis unit, and the composition of the liquid is detected in real time through a micro-spectroscopic, electrochemical sensor array, etc., to obtain the target component concentration Ck(t) and the interferent concentration Im(t). The analysis unit adopts a miniaturized and integrated design to ensure the minimum analysis delay.

[0092] Then, calculate the sample validity judgment equation:

[0093] E(t) = ∑[μ k ×C k (t)] - ∑[ν m ×I m (t)]

[0094] where E(t) is the sample validity index at time t, C k (t) is the concentration of the target component k, I m (t) is the concentration of the interferent m, μ k and ν m are the weight coefficients of the target component and the interferent respectively. This equation comprehensively evaluates the sample validity by weighing the concentrations of the target component and the interferent.

[0095] Furthermore, apply the switching decision function to determine the conversion point between the waste liquid and the valid sample:

[0096] S(t) = {1, if E(t) > θ(t) and dE(t) / dt > 0 {0, otherwise}

[0097] where θ(t) is the dynamic threshold function (adaptively adjusted according to the substance type and sampling history), and S(t) is the sampling switching function. When the value is 1, it means that the current liquid should be collected as a valid sample. The switching function not only considers the current validity index but also its change trend to ensure that sampling starts when the validity continues to improve.

[0098] Finally, determine the optimal sampling duration:

[0099] Topt = min{t | dE(t) / dt < ε and t > tstart}

[0100] where ε is the gradient threshold (a small positive number close to 0), and tstart is the start time of valid sampling. This formula ensures that sampling ends before the sample validity index starts to decline, maximizing the sample validity.

[0101] This module adopts a component analysis model based on a deep neural network (DNN), combines wavelet transform technology to extract features from sensor signals, and realizes high-sensitivity recognition of trace target components through a residual network structure. The system adopts an online learning method and dynamically adjusts the sample validity evaluation criteria according to different application scenarios.

[0102] IV. Integrated Intelligent Decision Module

[0103] As Figure 5 shown, the integrated intelligent decision module integrates the data of the three major modules to realize closed-loop feedback and self-optimizing sampling strategies.

[0104] In specific implementation:

[0105] First, integrate the output data of the three major modules and apply the system integration decision function:

[0106] D(t) = w1·F1(Pd(t), Bd(t)) + w2·F2(Sd(t), E(t)) + w3·F3(H(t))

[0107] Where D(t) is the comprehensive decision output at time t, F1, F2, and F3 are the decision functions of the depth prediction, physical property identification, and waste liquid identification modules respectively, Pd(t) is the predicted depth index, Bd(t) is the boundary identification index, Sd(t) is the sample validity index, E(t) is the sampling efficiency index, H(t) is the historical experience index, and w1, w2, and w3 are dynamic weight coefficients. This function realizes the fusion decision of multi-dimensional information.

[0108] Then, based on the decision output, the system sampling strategy is adaptively updated:

[0109]

[0110] Where θ(t) is the system parameter set (including weight coefficients, thresholds, etc. in each algorithm), J(θ) is the sampling quality evaluation function (comprehensively considering sample purity, representativeness, sampling efficiency, etc.), and η is the learning rate. Through this formula, the system can automatically optimize each parameter and continuously improve the sampling performance.

[0111] Next, the system controls the segmented microneedle system according to the comprehensive decision to perform precise puncture, hierarchical identification, waste liquid discrimination, and effective sample collection. In this process, the three modules work simultaneously and feedback to each other to form a closed-loop control: the depth prediction module provides the initial path planning, the physical property identification module adjusts the sampling position in real time, and the waste liquid identification module dynamically controls the sampling timing.

[0112] Finally, after completing the sampling of all layers, the system intelligently plans the withdrawal path and executes the system self-cleaning process to prevent cross-contamination. At the same time, the system incorporates the sampling data of this time into the historical database for the optimization of future sampling strategies.

[0113] This module adopts a reinforcement learning framework, uses the sampling quality as the reward function, and realizes the continuous optimization of the decision-making model. The system uses a graph neural network (GNN) structure to achieve cross-module information fusion and introduces a Bayesian optimization mechanism to accelerate the parameter search and optimization process.

[0114] V. Segmented Microneedle System

[0115] The segmented microneedle system is the core hardware for multi-level sampling. It consists of microneedle structures independently controlled in multiple segments, with each segment equipped with an independent sampling channel, a sensor array, and an actuator. Specifically, in implementation:

[0116] The segmented microneedle system adopts a modular design, and different segments can be freely combined to meet the requirements of different application scenarios. Each segment of the microneedle contains 4 core components: a precision advancing and retracting driver, a sensor array, a sampling channel, and an independent control unit.

[0117] The precision advancing and retracting driver uses piezoelectric ceramic drive technology to achieve displacement control with micron-level precision. The sensor array integrates various micro sensors such as mechanics, electricity, acoustics, and optics to achieve multi-dimensional monitoring of physical characteristics. The sampling channel adopts a multi-channel design with independent flow path control to ensure that samples at different layers do not mix. The independent control unit is responsible for local data processing and drive control, reducing the burden on the central processing system.

[0118] The surface of the microneedle is treated with a special coating, which has characteristics such as drag reduction, anti-adhesion, and anti-pollution, reducing puncture damage and improving sampling efficiency. The tip design refers to the bionics principle, simulating the structure of the mosquito mouthpart to achieve low-damage puncture and efficient liquid collection.

[0119] The system supports three sampling modes: synchronous sampling mode (all segments collect simultaneously), sequential sampling mode (collect different layers sequentially), and combined sampling mode (flexibly configured according to requirements). Under different modes, the system automatically adjusts the microneedle control strategy and flow path configuration to achieve the optimal sampling effect.

[0120] VI. Sample Collection and Identification System

[0121] The sample collection and identification system is responsible for automatically marking, separating, storing, and processing samples collected at different depths. Specifically, in implementation:

[0122] The system uses microfluidic chip technology to build a multi-channel sample processing platform. The sample collection unit includes a micro peristaltic pump, a multi-channel switching valve, and a sample storage bin, which can precisely control the sample flow and separation.

[0123] The automatic marking system uses RFID technology to attach an electronic tag to each sample container, recording information such as sampling time, depth, physical characteristic data, and validity index to ensure the traceability of the samples.

[0124] The sample preservation module automatically adjusts environmental parameters such as temperature, humidity, and oxygen content according to different sample characteristics to ensure sample stability. For special samples, the system can also perform preprocessing operations such as dilution, concentration, filtration, or chemical fixation to improve the efficiency of subsequent analysis.

[0125] The system also integrates basic analysis functions, which can conduct preliminary detection and evaluation on the collected samples, generate sample quality reports, and assist users in judging the sampling effect and subsequent analysis strategies.

[0126] VI. System Mode Configuration

[0127] As Figure 6 shown, the system of the present invention can be configured into four main working modes according to different application scenarios: medical biological tissue sampling mode, environmental and geological sampling mode, material detection mode, and food quality control mode. Under each mode, the system automatically adjusts relevant algorithm parameters and hardware configurations.

[0128] In the medical biological tissue sampling mode, the system optimizes the biocompatibility parameters, matches the medical imaging navigation system, and strengthens the minimally invasive performance, which is applicable to applications such as tumor heterogeneity stratification sampling, liquid biopsy, and drug permeability research.

[0129] In the environmental and geological sampling mode, the system enhances the mechanical strength, adjusts the physical property recognition parameters, and optimizes the anti-pollution ability, which is applicable to soil stratification pollutant monitoring, sediment chronology research, water body stratification precise sampling, etc.

[0130] In the material detection mode, the system expands the physical property monitoring range, improves the boundary recognition sensitivity, and strengthens the high temperature and high pressure resistance ability, which is applicable to composite material structure analysis, welding quality detection, material aging research, etc.

[0131] In the food quality control mode, the system strengthens the prevention and control of microbial contamination, enhances the sensitivity of component analysis, and optimizes the sample preservation conditions, which is applicable to fruit ripeness analysis, fermentation process monitoring, dairy product quality control, etc.

[0132] The system adopts a modular software and hardware design. Users can select the working mode through the touch screen interface or the remote control software. The system automatically loads the corresponding parameter set and configuration file to complete the mode switching. Under specific application requirements, the system also supports custom modes, and users can set key parameters to construct a dedicated sampling plan.

[0133] VIII. Implementation of Deep Learning Network

[0134] The multi-dimensional feature fusion depth profile prediction module, physical property boundary recognition module, waste liquid-effective sample dynamic recognition module, and integrated intelligent decision module of the system of the present invention are all implemented by using a deep learning network structure, specifically as follows:

[0135] The multi-dimensional feature fusion depth profile prediction module adopts a U-Net-like segmentation network architecture to achieve multi-modal data fusion and three-dimensional structure prediction through an encoder-decoder structure. The network introduces an attention mechanism and residual connections to improve the feature extraction efficiency and prediction accuracy.

[0136] The physical property boundary recognition module adopts a BiLSTM (Bidirectional Long Short-Term Memory Network) structure to process the sensor time-series data and accurately identify the mutation points of physical properties. The network introduces a gradient-enhanced tree ensemble method to improve the robustness of boundary recognition.

[0137] The waste liquid-valid sample dynamic recognition module adopts a hybrid architecture of one-dimensional convolutional neural network (1D-CNN) and Transformer to achieve the extraction of component spectrogram features and sequence analysis, and accurately identify the changing trend of sample validity.

[0138] The integrated intelligent decision-making module adopts a deep reinforcement learning framework, models the sampling process as a Markov decision-making process, and realizes the search for the optimal decision-making strategy through the policy network and the value network. The system adopts a meta-learning method to improve the cross-scenario generalization ability.

[0139] The system continuously optimizes the parameters through a self-learning algorithm, mainly including the following mechanisms: (1) Online fine-tuning: Dynamically adjust the network parameters according to real-time feedback; (2) Experience replay: Store historical sampling experiences for network training; (3) Multi-task learning: Optimize multiple objective functions simultaneously to improve the generality of the model; (4) Transfer learning: Utilize the knowledge accumulated in a certain field to quickly adapt to new application scenarios.

[0140] The deep learning architecture of the system adopts a distributed computing design. The key algorithms are deployed on the main control unit, and lightweight inference models are deployed on each functional module to achieve efficient collaborative computing. The system uses edge computing technology to reduce energy consumption while ensuring real-time performance and improve portability.

[0141] The implementation scenarios of the present invention are extensive and applicable to the in-situ extraction sampling requirements of various substances, mainly including the following categories:

[0142] First, medical biological tissue sampling. Configure micro needles made of biocompatible materials and integrate a medical imaging navigation system to achieve tumor heterogeneity stratified sampling, minimally invasive tissue acquisition, enrichment of circulating tumor cells in liquid biopsy, detection of trace biomarkers in cerebrospinal fluid, etc.

[0143] Second, environmental and geological sampling. Equip enhanced micro needles to adapt to different hardness strata to achieve soil stratified pollutant monitoring, sediment chronology research sampling, water body stratified precise sampling, analysis of groundwater layer quality differences, etc.

[0144] Third, advanced material detection. It is applicable to the analysis of the interlayer structure of composite materials, multi-level evaluation of coating quality, detection of the depth distribution of welding quality, research on the vertical distribution of material aging degree, etc.

[0145] Fourth, food and agricultural product quality control. Achieve the gradient analysis of fruit maturity, detection of the penetration depth of food additives, dynamic monitoring of the fermentation process, quality control of dairy products, etc.

[0146] In addition, the present invention is also applicable to fields such as cultural relics and art analysis, drug research and development and evaluation, environmental microorganism monitoring, etc. In these application scenarios, the system can adjust relevant parameters according to specific requirements and flexibly adapt to different material characteristics and sampling requirements.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified without departing from the purpose and scope of the present technical solution, and they should all be covered within the scope of the claims of the present invention.

Claims

1. An intelligent in-situ extraction and sampling method for substances, characterized in that: The method specifically includes the following steps: S1: Establish a multi-dimensional feature fusion depth profile prediction module, obtain the internal structure information of the target substance through multi-modal scanning, and construct a virtual model of the interior of the target substance based on this information to predict the optimal sampling depth, path, and multi-level sampling positions; S2: Construct a physical property boundary recognition module, monitor the physical property parameters in real time during the micro-needle puncture process for sampling, calculate the physical property gradient, dynamically identify the boundaries of different substance levels, and determine the optimal sampling positions; S3: Construct a waste liquid - effective sample dynamic recognition module, import the collected liquid into the microfluidic analysis unit of this dynamic recognition module to detect the liquid composition in real time, calculate the sample effectiveness index, automatically determine the conversion point between waste liquid and effective sample, and determine the optimal sampling timing and duration; S4: Construct an integrated intelligent decision-making module, integrate the output data of the above three modules, realize the adaptive adjustment of system parameters, and cooperate to control the whole process of micro-needle puncture, layer identification, sample collection, and system withdrawal to complete the intelligent in-situ extraction and sampling of substances.

2. The intelligent in-situ extraction and sampling method of substances according to claim 1, characterized in that: In step S1, using the multi-dimensional feature fusion depth profile prediction module, obtain the internal structure information of the target substance in different dimensions through multiple modalities such as ultrasonic scanning, optical imaging, and impedance measurement; apply a depth prediction model to integrate the multi-modal information and construct a virtual model of the interior of the target substance based on this information; the depth prediction model uses the following equation: Among them, D(x, y) is the predicted optimal sampling depth, and F i (x, y) is the feature function of different modalities, and w i is the weight coefficient of each feature, β is the gradient adjustment factor, is the physical property gradient function; Determine the multi-level sampling positions based on the predicted depth: L i = D(x,y) × α i + Δi where L i is the sampling position of the i-th layer, and α i is the relative depth coefficient, and Δi is the adaptive adjustment amount.

3. The intelligent in-situ extraction and sampling method of substances according to claim 2, characterized in that: In step S2, the microneedle system in the physical property boundary recognition module integrates various sensors of mechanics, electricity, acoustics, optics, and thermotics to form a physical property parameter set P j (d); The physical property gradient monitoring equation is calculated as follows: G(d) = ∑[λ j ×(dP j / dd)] Among them, G(d) is the comprehensive physical property gradient at depth d, and P j is the physical property parameter of the j-th type, and λ j is the characteristic weight coefficient; Apply the boundary recognition function B(d) = {1, if |G(d)| > τ(d); 0, otherwise} to identify the layer boundaries, where τ(d) is a depth-related dynamic threshold function and B(d) is a boundary indication function; Determine the optimized sampling position: Sopt = argmax{Q(d)|d ∈ [d1, d2]}, where Q(d) is a sampling quality evaluation function and [d1, d2] is the depth interval of the target layer.

4. The intelligent in-situ extraction and sampling method of substances according to claim 3, wherein: In step S3, in the waste liquid - effective sample dynamic recognition module: import the collected liquid into the microfluidic analysis unit, and detect the liquid composition in real time through a micro-spectroscopy and electrochemical sensor array to obtain the target component concentration Ck(t) and the interferent concentration Im(t); Calculate the sample effectiveness judgment equation as follows: E(t) = ∑[μ k × Ck(t)] - ∑[ν m × Im(t)], where E(t) is the sample validity index at time t, and μ k and ν m are the weight coefficients of the target component and the interferent, respectively; Apply the switching decision function: S(t) = {1, if E(t) > θ(t) and dE(t) / dt > 0; 0, otherwise}, where θ(t) is a dynamic threshold function and S(t) is a sampling switching function; Determine the optimal sampling duration: Topt = min{t|dE(t) / dt < ε and t > tstart}, where ε is a gradient threshold and tstart is the start time of effective sampling.

5. The intelligent in-situ extraction and sampling method of substances according to claim 4, characterized in that: In step S4, in the integrated intelligent decision-making module, integrate the output data of the three modules and apply the system integration decision function: D(t) = w1·F1(Pd(t), Bd(t)) + w2·F2(Sd(t), E(t)) + w3·F3(H(t)) Among them, D(t) is the comprehensive decision-making output, F1, F2, and F3 are the decision-making functions of three modules, Pd(t) is the predicted depth index, Bd(t) is the boundary recognition index, Sd(t) is the sample validity index, E(t) is the sampling efficiency index, H(t) is the historical experience index, and w1, w2, and w3 are dynamic weight coefficients; Realize the adaptive update of the sampling strategy: Among them, θ(t) is the system parameter set, J(θ) is the sampling quality evaluation function, and η is the learning rate.

6. The intelligent in-situ extraction and sampling method of substances according to claim 5, characterized in that: The multi-dimensional feature fusion depth profile prediction module, physical property boundary recognition module, waste liquid-effective sample dynamic recognition module, and integrated intelligent decision-making module are all implemented using a deep learning network structure. The system continuously optimizes parameters through a self-learning algorithm to improve the adaptability to different substance types and sampling accuracy.

7. An intelligent in-situ extraction and sampling system for substances, characterized in that: The system includes a multi-dimensional feature fusion depth profile prediction module, a physical property boundary recognition module, a waste liquid-effective sample dynamic recognition module, and an integrated intelligent decision-making module; the multi-dimensional feature fusion depth profile prediction module is used to obtain the internal structure information of the target substance through multi-modal scanning, and based on this information, construct a virtual model inside the target substance to predict the optimal sampling depth, path, and multi-level sampling positions; the physical property boundary recognition module is used to monitor the physical property parameters in real time during the micro-needle puncture process, calculate the physical property gradient, dynamically identify the boundaries of different substance layers, and determine the optimal sampling positions; the waste liquid-effective sample dynamic recognition module is used to import the collected liquid into the microfluidic analysis unit of the dynamic recognition module to detect the liquid composition in real time, calculate the sample validity index, automatically determine the conversion point between waste liquid and effective samples, and determine the optimal sampling timing and duration; The integrated intelligent decision-making module is used to integrate the output data of the above three modules, realize the adaptive adjustment of system parameters, and cooperate to control the whole process of micro-needle puncture, layer identification, sample collection, and system withdrawal, and complete the intelligent in-situ extraction and sampling of substances.

8. An intelligent in-situ extraction and sampling system for substances according to claim 7, characterized in that: The system also includes: Segmented micro-needle system: The segmented micro-needle system includes a micro-needle structure with multiple independently controlled segments, each segment is equipped with an independent sampling channel, a sensor array, and an actuator, to achieve synchronous or sequential sampling at multiple depth levels; Sample collection and identification system: The sample collection and identification system is used to automatically label, separate, store, and process samples at different depths.

9. An intelligent in-situ extraction and sampling system for substances according to claim 8, characterized in that: The segmented micro-needle system is composed of a micro-needle structure with multiple independently controlled segments, each segment is equipped with an independent sampling channel, a sensor array, and an actuator, and specifically includes: The segmented micro-needle system adopts a modular design, and different segments can be freely combined to meet the requirements of different application scenarios; each segment of the micro-needle contains 4 core components: a precision advance and retreat driver, a sensor array, a sampling channel, and an independent control unit; the precision advance and retreat driver uses piezoelectric ceramic drive technology to achieve displacement control with micron-level accuracy; the sensor array uses a variety of micro sensors such as mechanics, electricity, acoustics, and optics to achieve multi-dimensional monitoring of physical properties; the sampling channel adopts a multi-channel design with independent flow path control to ensure that samples at different layers do not mix; the independent control unit is responsible for local data processing and drive control; The surface of the micro-needle is treated with a special coating, and the tip design adopts the principle of bionics, simulating the structure of a mosquito's mouthpart, to achieve low-damage puncture and efficient liquid collection.

10. An intelligent in-situ extraction and sampling system for substances according to claim 9, characterized in that: The described sample collection and identification system is responsible for automatically labeling, separating, storing, and processing samples collected at different depths, specifically including: The sample collection unit, which includes a micro peristaltic pump, a multi-channel switching valve, and a sample storage bin, and can precisely control sample flow and separation; The automatic labeling system, which uses RFID technology to attach an electronic tag to each sample container, records information such as sampling time, depth, physical property data, and validity index, and ensures sample traceability; The sample preservation module automatically adjusts environmental parameters such as temperature, humidity, and oxygen content according to different sample characteristics to ensure sample stability; The system integrates analysis functions to conduct preliminary detection and evaluation of the collected samples, generate a sample quality report, and assist users in judging the sampling effect and subsequent analysis strategies.

Images