A rapid detection method and system for soil heavy metal ecological risk microorganisms

Abstract

This invention relates to the field of environmental monitoring and analysis technology, and particularly to a rapid detection method and system for soil heavy metal ecological risk microorganisms. The method includes: collecting microbial sensor signals from soil samples and extracting biotoxicity response features to construct a biotoxicity response spectrum dataset; establishing a dose-response relationship matrix from the dataset, training a support vector machine model to obtain stability indices and determine information fusion weights, and establishing a mapping model between the biological signal baseline and heavy metal concentration after feature fusion; acquiring signals from the sample to be tested and extracting features, inputting the biotoxicity response features into the support vector machine model to output prediction confidence to correct the fusion weights, obtaining the fused features to be tested, calculating the biological signal bias, and outputting the predicted heavy metal concentration value, thereby determining the pollution level and ecological risk level. This invention improves the stability of results and the usability for on-site screening.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and analysis technology, and in particular to a rapid detection method and system for soil heavy metal ecological risk microorganisms. Background Technology

[0002] In scenarios such as tailings ponds and mine remediation, the spatial distribution of heavy metal pollution exhibits significant locality and temporal variability. Rapid on-site screening capabilities directly impact the efficiency of zoned remediation, investment decisions, and secondary pollution control. Existing detection methods largely rely on laboratory chemical analysis or on-site elemental determination. While these methods provide content information, they often fail to simultaneously reflect bioavailability and ecological effects, and suffer from long sampling and testing cycles, high costs, and difficulties in high-frequency sampling. On the other hand, rapid assessments based on bioresponse and microbial sensor readouts, while more closely aligned with ecological risk, are susceptible to changes in soil matrix, conductivity, microbial sensor drift, and microbial state fluctuations. This leads to insufficient comparability of results across different batches and unstable threshold determinations, limiting large-scale application. Summary of the Invention

[0003] This invention provides a rapid detection method and system for soil heavy metal ecological risk microorganisms, which at least solves the problem of how to stably obtain comparable detection results that can be used for heavy metal pollution classification and ecological risk assessment under the interference of soil matrix and sensor drift.

[0004] In a first aspect, the present invention provides a rapid detection method for soil heavy metal ecological risk microorganisms, comprising the following steps:

[0005] Acquire microbial sensor signals from soil samples, extract biotoxicity response features, and construct a biotoxicity response spectrum dataset;

[0006] A dose-response matrix was established using the biotoxicity response spectrum dataset. A support vector machine model was trained using the dose-response matrix to obtain a stability index. The information fusion weights were determined based on the stability index. Biotoxicity response features were fused to obtain fusion features. The fusion features were statistically analyzed to obtain the biosignal baseline. A heavy metal concentration mapping model was established according to the correspondence between heavy metal standard concentrations and fusion features.

[0007] Microbial sensor signals from soil samples to be tested are acquired, biotoxicity response features are extracted, the biotoxicity response features are input into a support vector machine model to output prediction confidence, the information fusion weights are corrected, the biotoxicity response features are fused to obtain the fused features to be tested, the fused features to be tested are input into a heavy metal concentration mapping model to output the predicted heavy metal concentration, the predicted heavy metal concentration is compared with the pollution level threshold sequence to determine the pollution level, and the ecological risk level is determined based on the pollution level.

[0008] In one possible implementation, the microbial sensor signal includes an electrochemical current signal and an electrochemical impedance signal.

[0009] In one possible implementation, extracting biotoxicity response characteristics includes: determining a growth rate inhibition curve using electrochemical current signals and electrochemical impedance signals, and determining an enzyme activity inhibition gradient using electrochemical current signals and electrochemical impedance signals.

[0010] In one possible implementation, constructing a biotoxicity response spectrum dataset includes: acquiring microbial sensor signals from soil samples with known heavy metal standard concentrations, extracting biotoxicity response features, and establishing a correspondence between the biotoxicity response features and the heavy metal standard concentrations to construct the biotoxicity response spectrum dataset.

[0011] In one possible implementation, establishing the dose-response matrix involves arranging the biotoxicity response characteristics according to different levels of heavy metal standard concentrations, with the rows of the dose-response matrix corresponding to the heavy metal standard concentrations and the columns corresponding to the biotoxicity response characteristics.

[0012] In one possible implementation, training the support vector machine model involves training with a linear support vector machine model, where the stability metric is determined by the minimum classification margin between the classification hyperplane of the linear support vector machine model and the training samples.

[0013] In one possible implementation, determining the information fusion weights based on the stability index includes obtaining the feature weights of the linear support vector machine model and performing normalization to obtain the information fusion weights. The fusion of biotoxicity response features to obtain the fusion features includes weighting and summing the biotoxicity response features according to the information fusion weights to obtain the fusion features.

[0014] In one possible implementation, statistical analysis of the fusion features to obtain the biological signal baseline includes statistical analysis of the fusion features of the biotoxicity response spectrum dataset to obtain the biological signal baseline, and establishing a heavy metal concentration mapping model includes training a support vector regression model according to the correspondence between the heavy metal standard concentration and the fusion features to obtain the heavy metal concentration mapping model.

[0015] In one possible implementation, the prediction confidence is determined by the classification confidence of the output of the soil sample to be tested by the support vector machine model. The information fusion weights are corrected by weighting and updating the information fusion weights according to the prediction confidence. The biological signal bias is determined by the difference between the fused feature to be tested and the biological signal baseline. The pollution level is determined by comparing the predicted heavy metal concentration with the pollution level threshold sequence. A correspondence is established between the ecological risk level interval threshold sequence and the pollution level. The potential ecological risk index is calculated by the predicted heavy metal concentration, the background heavy metal concentration and the heavy metal toxicity coefficient, and the potential ecological risk index is compared with the ecological risk level interval threshold sequence to determine the level.

[0016] Secondly, the present invention provides a rapid detection system for soil heavy metal ecological risk microorganisms, used to implement the rapid detection method for soil heavy metal ecological risk microorganisms, the system comprising:

[0017] The biosignal acquisition module is used to acquire microbial sensor signals from soil samples, extract biotoxicity response characteristics, and construct a biotoxicity response spectrum dataset.

[0018] The model building module is used to establish a dose-response relationship matrix from the biotoxicity response spectrum dataset, train a support vector machine model using the dose-response relationship matrix to obtain a stability index, determine the information fusion weights based on the stability index, fuse biotoxicity response features to obtain fusion features, statistically analyze the fusion features to obtain the biological signal baseline, and establish a heavy metal concentration mapping model according to the correspondence between heavy metal standard concentration and fusion features.

[0019] The risk assessment module is used to acquire microbial sensor signals from the soil sample to be tested, extract biotoxicity response features, input the biotoxicity response features into the support vector machine model to output the prediction confidence, correct the information fusion weights, fuse the biotoxicity response features to obtain the fused features to be tested, input the fused features to be tested into the heavy metal concentration mapping model to output the predicted heavy metal concentration, compare the predicted heavy metal concentration with the pollution level threshold sequence to determine the pollution level, and determine the ecological risk level based on the pollution level.

[0020] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0021] By jointly acquiring and synchronously extracting electrochemical current and electrochemical impedance signals, a multidimensional characterization of the physiological inhibitory effect of microorganisms was achieved, reducing the interference of soil matrix differences on single signals. Through the structured construction of biotoxicity response spectrum datasets and dose-effect relationship matrices, consistency of training data caliber and solidification of concentration gradient information were achieved. By generating stability indices through support vector machine models and determining information fusion weights accordingly, data-driven updates of feature fusion weights were achieved. By correcting information fusion weights through prediction confidence and combining them with biological signal baselines, drift and anomalous features were suppressed. By outputting concentration predictions through a heavy metal concentration mapping model and linking pollution levels with ecological risk levels, an integrated output of concentration-level-risk was achieved. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the execution flow of the method of the present invention;

[0023] Figure 2 This is a structural block diagram of the system of the present invention. Detailed Implementation

[0024] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] Rapid microbial detection refers to using microbial cells or metabolic systems as sensitive units, combined with microbial sensors, to convert external chemical stresses into measurable signal changes within a short time, and obtaining comparable discrimination results through standardized data processing. Unlike detection approaches centered on elemental content, rapid microbial detection emphasizes the direct characterization of "biological response," reflecting the inhibitory characteristics of pollutants on microbial growth activity, metabolic intensity, and key enzyme activity within a short time window, thus providing rapid evidence for risk screening, zoning decisions, and process monitoring. In soil scenarios, rapid microbial detection typically relies on microbial immobilization carriers and electrochemical microbial sensor readout methods, using electrochemical current signals and electrochemical impedance signals as response carriers. Then, through feature extraction and model mapping, complex response changes are converted into an indicator system that can be used for judgment. Based on the above technical approach, this invention focuses on the acquisition of microbial sensor signals from soil samples, the construction of toxicity response characteristics, and data-driven fusion modeling to form a rapid detection scheme for ecological risk assessment.

[0028] like Figure 1 As shown, a rapid detection method for soil heavy metal ecological risk microorganisms includes the following steps:

[0029] Acquire microbial sensor signals from soil samples, extract biotoxicity response features, and construct a biotoxicity response spectrum dataset;

[0030] In this embodiment, after obtaining soil samples from the target area, the soil samples are mixed with a buffer solution and agitated. A soil extract is obtained through solid-liquid separation. The soil extract is injected into the reaction chamber of a microbial sensor, bringing the immobilized microbial electrode into contact with the extract. Electrochemical current and electrochemical impedance signals are collected to form a microbial sensor signal. The microbial sensor signal is segmented and baseline-corrected, and biotoxicity response features are extracted. These features include response amplitude, rate of change, recovery time, and inhibition rate, and are recorded according to sample number. The above collection and feature extraction process is repeated for soil samples with known heavy metal standard concentrations. A one-to-one correspondence is established between the biotoxicity response features and the heavy metal standard concentrations, forming a biotoxicity response spectrum dataset, thereby improving the consistency of subsequent modeling.

[0031] Microbial sensor signals include electrochemical current signals and electrochemical impedance signals.

[0032] In this embodiment, the microbial sensor comprises a reaction chamber, an immobilized microbial electrode, a reference electrode, a counter electrode, a signal acquisition circuit, and a data processing unit. The immobilized microbial electrode supports the microbial membrane layer, and the microbial membrane layer and the conductive substrate of the electrode form a stable electron transfer interface. The reaction chamber contains the soil extract and provides constant fluid contact conditions. The signal acquisition circuit employs an electrochemical workstation or an equivalent multi-channel electrochemical measurement circuit to perform constant potential current measurement and AC impedance measurement, respectively, to obtain electrochemical current signals and electrochemical impedance signals, which are then packaged together into a microbial sensor signal.

[0033] During the acquisition of electrochemical current signals, soil extract was injected into the reaction chamber and allowed to stabilize for a preset time. Subsequently, a constant potential was applied to the immobilized microbial electrode using a reference electrode as the potential benchmark, and the current change sequence between the working electrode and the counter electrode was continuously acquired. The electrochemical current signal was used to characterize the changes in metabolic electron transport intensity of microorganisms under the action of soil extract. When heavy metal ions inhibited the microbial metabolic process, the electrochemical current signal exhibited characteristics such as a decrease in amplitude, a slowdown in the rate of rise, or a delayed recovery, thus providing direct observational data for subsequent extraction of biotoxicity response characteristics.

[0034] When acquiring electrochemical impedance spectroscopy (EIS) signals, the state of the medium and electrode interface within the reaction chamber is kept consistent. A small-amplitude AC excitation is superimposed under a constant bias potential, and multi-frequency scanning measurements are performed on the immobilized microbial electrode to obtain impedance spectrum data that varies with frequency. The EIS signal is used to characterize the comprehensive changes in charge transfer and diffusion processes at the electrode interface. When heavy metal ions alter the activity of the microbial membrane or the interfacial charge transfer behavior, the morphology of the impedance spectrum changes identifiablely in the low-frequency and mid-to-high-frequency ranges, thus providing auxiliary criteria for distinguishing disturbances from different sources such as signal noise, changes in medium conductivity, and changes in microbial activity.

[0035] To ensure the alignment and fusion of electrochemical current and electrochemical impedance signals under the same sample conditions, the data processing unit implements a unified acquisition process control and timestamp marking for both types of signals: within the same soil extract injection cycle, the electrochemical current signal is acquired first and reaches a stable phase before the electrochemical impedance signal is acquired; or, within the same time window, the two types of signals are acquired alternately and the sampling order is recorded. Both types of signals undergo consistent quality control processing before feature extraction, including removing abnormal peaks, correcting slow drift, and normalizing dimensional differences, so that the electrochemical current and electrochemical impedance signals can jointly reflect the toxic effects of the soil extract on the microbial system. By simultaneously acquiring the electrochemical current and electrochemical impedance signals, the sensitivity to heavy metal inhibition effects is improved. Furthermore, even with interference such as sensor aging, changes in solution conductivity, or bubble adhesion, the detection stability can still be maintained by relying on the other type of signal, thus providing a more consistent input data foundation for subsequent establishment of a biotoxicity response spectrum dataset, construction of a dose-effect relationship matrix, and model training.

[0036] Extracting biotoxicity response characteristics includes: determining the growth rate inhibition curve using electrochemical current signals and electrochemical impedance signals, and determining the enzyme activity inhibition gradient using electrochemical current signals and electrochemical impedance signals.

[0037] In this embodiment, biotoxicity response characteristics are used to characterize the degree of inhibition of microbial activity on the immobilized microbial electrode by the soil extract. To ensure the comparability of characteristics, the acquisition of microbial sensor signals follows a consistent timing sequence: after the soil extract enters the reaction chamber, a steady-state establishment is first performed, followed by continuous acquisition of electrochemical current signals; electrochemical impedance signals are acquired under the same reaction chamber conditions, and the charge transfer resistance reflecting the ease of electron transfer at the interface is obtained from the impedance data. By simultaneously utilizing electrochemical current and electrochemical impedance signals, the influence of factors such as soil extract conductivity fluctuations and bubble adhesion on a single signal can be reduced, making the subsequently constructed characteristics more stable.

[0038] The process for determining the growth rate inhibition curve is as follows: A steady-state time window is selected from the electrochemical current signal, and the steady-state current amplitude is calculated. The charge transfer resistance is obtained by equivalent circuit fitting in the electrochemical impedance signal. The steady-state current amplitude and charge transfer resistance are combined to obtain the growth rate characterization value. A heavy metal-free control sample is used as a benchmark to calculate the corresponding inhibition rate. The above process is repeated for samples with different heavy metal standard concentrations. The relationship between the inhibition rate and the heavy metal standard concentration is recorded as the growth rate inhibition curve and included as part of the biotoxicity response characteristics in the biotoxicity response spectrum dataset.

[0039] The process for determining the enzyme activity inhibition gradient is as follows: Under the same soil leachate conditions, a preset substrate solution is injected into the reaction chamber to trigger the current response of the microbial enzymatic reaction. The increment of the current response is extracted and combined with the charge transfer resistance obtained by the electrochemical impedance signal to form the enzyme activity characterization value. The enzyme activity inhibition rate is calculated using a heavy metal-free control sample. The difference in enzyme activity inhibition rate corresponding to adjacent heavy metal standard concentration samples is used to obtain the enzyme activity inhibition gradient, which reflects the sensitivity of enzyme activity to concentration changes and facilitates the subsequent model to distinguish between different pollution levels.

[0040] The core calculation expression involved in this embodiment is as follows:

[0041]

[0042] This is a value representing the growth rate; This represents the amplitude of the steady-state electrochemical current signal. The charge transfer resistance is obtained by fitting the electrochemical impedance signal.

[0043]

[0044] This represents the enzyme activity characterization value; This represents the increment in the electrochemical current signal response after substrate injection; The charge transfer resistance is obtained by fitting the electrochemical impedance signal.

[0045]

[0046] Inhibition rate; These are bioactivity characterization values; These are the bioactivity characterization values ​​for the heavy metal-free control sample. Pick To obtain the growth rate inhibition rate, Pick The enzyme activity inhibition rate was obtained.

[0047]

[0048] The enzyme activity inhibition gradient; For the first Enzyme activity inhibition rate of samples with standard concentrations of heavy metals; For the first Each heavy metal standard concentration.

[0049] Through the above treatment, the growth rate inhibition curve reflects the trend of overall growth activity with concentration, and the enzyme activity inhibition gradient reflects the sensitivity of key metabolic processes to concentration changes. Together, they improve the stability and distinguishability of subsequent modeling and judgment.

[0050] The construction of the biotoxicity response spectrum dataset includes: acquiring microbial sensor signals from soil samples with known heavy metal standard concentrations, extracting biotoxicity response features, and establishing a correspondence between the biotoxicity response features and heavy metal standard concentrations to construct the biotoxicity response spectrum dataset.

[0051] In this embodiment, to construct a biotoxicity response spectrum dataset, the target heavy metal species are first identified, and a heavy metal standard concentration sequence covering different pollution levels is established. Using a substrate soil with known background heavy metal content and uniform texture as a carrier, the target heavy metal standard solution is added to the substrate soil at a set concentration level. Mechanical mixing is used to uniformly disperse the heavy metals in the soil, and the sample is then sealed and allowed to stand to complete the adsorption and speciation stabilization process, thereby obtaining a soil sample sequence with known heavy metal standard concentrations. For each concentration level of soil sample, a soil extract is prepared at a consistent solid-liquid ratio, and a supernatant for sensing detection is obtained through solid-liquid separation. The soil extract is injected into the reaction chamber of the microbial sensor, ensuring full contact between the immobilized microbial electrode and the soil extract. Electrochemical current and electrochemical impedance signals are simultaneously acquired according to a preset acquisition sequence, forming the microbial sensor signal for that soil sample. Subsequently, the microbial sensor signals underwent quality control and standardization processing, including removing abnormal peaks, correcting slow drift, and aligning sampling time windows. Based on the aforementioned biotoxicity response feature extraction rules, the correlation features of growth rate inhibition curves and enzyme activity inhibition gradients were determined using electrochemical current and electrochemical impedance signals, respectively, resulting in biotoxicity response feature records for the soil sample. Soil samples at the same concentration level were repeatedly measured, and the consistency of the repeated results was verified. Abnormal records with significant deviations were removed, and biotoxicity response feature records that met the consistency requirements were retained. Finally, each biotoxicity response feature record was bound to its corresponding heavy metal standard concentration, forming a set of sample entries with concentration labels. These entries were then numbered and archived according to heavy metal type, concentration level, and collection batch, thus constructing a biotoxicity response spectrum dataset. This dataset provides stable input for subsequent dose-response matrix construction and model training, improving the distinguishability and robustness of the heavy metal concentration mapping model across different pollution levels.

[0052] A dose-response matrix was established using the biotoxicity response spectrum dataset. A support vector machine model was trained using the dose-response matrix to obtain a stability index. The information fusion weights were determined based on the stability index. Biotoxicity response features were fused to obtain fusion features. The fusion features were statistically analyzed to obtain the biosignal baseline. A heavy metal concentration mapping model was established according to the correspondence between heavy metal standard concentrations and fusion features.

[0053] In this embodiment, the biotoxicity response spectrum dataset is sorted according to the heavy metal standard concentration sequence to establish a dose-response matrix. The rows of the dose-response matrix correspond to the heavy metal standard concentration levels, and the columns correspond to features related to growth rate inhibition curves and enzyme activity inhibition gradients. After normalizing the features in each column, a linear support vector machine model is trained using concentration interval labels as supervisory information. A stability index is obtained by repeatedly randomly partitioning the training and validation sets and statistically analyzing the consistency of classification results. Based on the stability index, feature weights for fusion are determined and normalized to obtain information fusion weights. The biotoxicity response features are then weighted and summed according to these information fusion weights to obtain the fused features. The fused features of zero-concentration and low-concentration samples are statistically analyzed to obtain a biosignal baseline. Using the fused features as input and the heavy metal standard concentration as output, a support vector regression model is trained to obtain a heavy metal concentration mapping model, thereby improving the stability and discriminability of concentration prediction.

[0054] Establishing the dose-response matrix involves arranging the biotoxicity response characteristics according to different levels of heavy metal standard concentrations. The rows of the dose-response matrix correspond to the heavy metal standard concentrations, and the columns of the dose-response matrix correspond to the biotoxicity response characteristics.

[0055] In this embodiment, the dose-response matrix is ​​directly obtained from the biotoxicity response spectrum dataset. This matrix establishes a structured correspondence between heavy metal standard concentrations and biotoxicity response features, providing consistent input for subsequent support vector machine model training and information fusion weight determination. Each sample entry in the biotoxicity response spectrum dataset contains at least a heavy metal standard concentration label, a biotoxicity response feature record, a collection batch identifier, and a sample number. The biotoxicity response feature records are stored in a unified order according to a pre-agreed feature list. This feature list includes features related to growth rate inhibition curves and features related to enzyme activity inhibition gradients, and the same feature uses the same calculation method and dimension across different entries.

[0056] When establishing the dose-response matrix, the biotoxicity response spectrum dataset is first grouped according to the heavy metal standard concentration, and the heavy metal standard concentration levels are then sorted from low to high. Subsequently, a consistency check is performed on the sample entries within each concentration level. This consistency check includes: checking the completeness of the feature list, checking the completeness of the acquisition time sequence markers, and checking for any entries with significant distortion due to sensor anomalies. For sample entries that pass the check, biotoxicity response feature records are extracted according to sample number to form a feature set for that concentration level. When there are repeated measurements at the same concentration level, outlier removal is performed on the feature sets of repeated measurements. Outlier removal uses a threshold determination method based on interquartile range or standard deviation to remove feature records that significantly deviate from the distribution of the same group. After outlier removal, the remaining feature records are summarized within the group using either mean summarization or median summarization to obtain the representative biotoxicity response feature vector corresponding to that concentration level.

[0057] After generating representative biotoxicity response feature vectors for each concentration level, these vectors are stacked sequentially in concentration order to obtain a dose-response matrix. The rows of the dose-response matrix correspond to the standard concentration levels of heavy metals, and the columns correspond to the biotoxicity response features, with the column order consistent with the feature list. To ensure comparability of different features in subsequent training stages, each column is normalized after matrix generation. Normalization employs either minimum-maximum scaling based on training data or mean-variance standardization based on training data. The normalization parameters are saved along with the dose-response matrix and reused in subsequent processing of soil samples.

[0058] Through the matrix construction method described above, the correspondence between heavy metal standard concentration levels and biotoxicity response characteristics is structurally solidified. This not only reflects the overall inhibitory trend brought about by increasing concentration, but also retains the sensitivity differences of different characteristics to concentration changes, thereby improving the repeatability of subsequent stability index calculations and providing a stable data foundation for determining information fusion weights.

[0059] Training a support vector machine model includes training with a linear support vector machine model. The stability metric is determined by the minimum classification margin between the classification hyperplane of the linear support vector machine model and the training samples.

[0060] In this embodiment, a linear support vector machine model is used to learn the discrimination boundary between biotoxicity response features and heavy metal standard concentration levels from the dose-response matrix, and to generate a stability index based on the classification interval of the training samples. Specifically, each row of the dose-response matrix is ​​used as an input feature vector for a training sample, and the heavy metal standard concentration level corresponding to that row is mapped to a category label according to a preset concentration range, thereby forming supervised training data. To avoid the difference in the scale of different features affecting the classification boundary, the biotoxicity response features in each column are normalized in the same way as described above before training, and the normalization parameters are saved for reuse in subsequent test samples.

[0061] The linear support vector machine (SVM) model is trained using a maximum margin classification strategy that prioritizes linearly separable or approximately linearly separable models, outputting classification parameters and bias terms. After training, the classification margin is calculated for each training sample, and the statistical measure of the classification margin is used as a stability index. The classification margin is calculated using the following expression:

[0062]

[0063] For the first The classification margin of each training sample; For the first The class labels of the training samples; For the first The input feature vector of each training sample; This represents the classification parameter vector of the linear support vector machine model. This represents the bias term of the linear support vector machine model.

[0064] Stability metrics are used to characterize the separability and boundary robustness of the dose-response relationship matrix under the current feature set. In this embodiment, the quantile statistics of the training sample classification margin are used as stability metrics to reduce the impact of individual outliers on the results.

[0065]

[0066] For stability indicators; The training sample classification margin set of Quantiles.

[0067] In implementation, A fixed value is taken and saved along with the model parameters to ensure that the stability index has a consistent judgment caliber across different batches of data. Through the above training and calculation method, the stability index can reflect the contribution of different biotoxicity response characteristics to the differentiation of concentration levels, providing a reliable basis for subsequent feature fusion based on information fusion weights, and maintaining the stability of the model training and weight determination process when there are batch differences in sensor input.

[0068] Determining the information fusion weights based on stability indices includes obtaining the feature weights of the linear support vector machine model and performing normalization to obtain the information fusion weights. The fusion of biotoxicity response features to obtain fusion features includes weighting and summing the biotoxicity response features according to the information fusion weights.

[0069] In this embodiment, the information fusion weights are used to transform biotoxicity response features into fused features that can be used for subsequent modeling, thereby reducing the impact of fluctuations caused by differences in soil matrix and sensor batches on individual features. After the linear support vector machine model is trained, the data processing unit extracts the feature weights corresponding to each biotoxicity response feature from the classification parameter vector output by the linear support vector machine model. To make the weights interpretable and easy to use directly for weighted fusion, the feature weights are first denegated to avoid the positive and negative signs causing mutual cancellation of the fusion results; then, the denegated feature weights are normalized so that the sum of all information fusion weights is 1, thus obtaining the information fusion weight set.

[0070] In terms of implementation, the data processing unit performs multiple training partitions and iterations on the dose-response matrix to obtain multiple sets of linear support vector machine (SVM) models. Each SVM model has a stability index obtained based on the aforementioned classification interval statistics. The data processing unit uses the stability index as a criterion for model reliability, selecting the SVM model with the highest stability index and extracting feature weights from this model as the source of information fusion weights, thus ensuring that the information fusion weights are consistent with the separability of the training data. After determining the information fusion weights, the data processing unit performs a weighted summation of the biotoxicity response features corresponding to the same sample according to the information fusion weights to obtain fusion features. A correspondence is established between the fusion features and heavy metal standard concentrations, used for subsequent statistical analysis of the fusion features to obtain a biological signal baseline. A heavy metal concentration mapping model is then established based on the correspondence between heavy metal standard concentrations and fusion features.

[0071] The core calculation expression involved in this embodiment is as follows:

[0072]

[0073] For the first Information fusion weights for individual biological toxicity response characteristics; For the classification parameter vector of the linear support vector machine model, the one that is related to the first... Feature weights corresponding to each biological toxicity response characteristic; This represents the number of biotoxicity response characteristics.

[0074]

[0075] Features of fusion; For the first Information fusion weights for individual biological toxicity response characteristics; For the first sample Each biological toxicity response characteristic takes a value; This represents the number of biotoxicity response characteristics.

[0076] In this way, the information fusion weights are obtained by training data. The fusion features are uniformly summarized under the same weight system for the biological toxicity response features from different sources, making the fusion features more consistent with the changes in heavy metal standard concentrations. This improves the stability and distinguishability of the subsequent biological signal baseline establishment and heavy metal concentration mapping model training process.

[0077] The biological signal baseline is obtained by statistically analyzing the fusion features of the biotoxicity response spectrum dataset. The heavy metal concentration mapping model is established by training a support vector regression model according to the correspondence between the heavy metal standard concentration and the fusion features.

[0078] In this embodiment, the fusion feature is obtained by weighting and summing the biotoxicity response features according to the information fusion weight. The fusion feature is used to uniformly characterize the comprehensive inhibitory effect of soil extract on immobilized microbial electrodes. When establishing the biosignal baseline, sample entries for characterizing low-contamination or background states are selected from the biotoxicity response spectrum dataset. These sample entries include zero-concentration sample entries and low-concentration sample entries. The fusion features of these sample entries are subjected to consistency verification and outlier removal to eliminate abnormal fusion features caused by poor electrode contact, bubble adhesion, or abnormal collection. Subsequently, the retained fusion features are statistically analyzed to obtain the biosignal baseline. In this embodiment, median statistics are used to improve robustness to occasional perturbations. The biosignal baseline is determined by the following expression:

[0079]

[0080] As a baseline for biological signals; For the first Fusion features of individual sample entries; This represents the number of sample entries included in the statistics. The biosignal baseline, information fusion weights, and normalization parameters are saved together for determining the biosignal bias of subsequent soil samples, ensuring comparability between different batches of measurements under the same benchmark.

[0081] When establishing the heavy metal concentration mapping model, the fusion feature of each sample item in the biotoxicity response spectrum dataset is used as the model input, and the corresponding heavy metal standard concentration is used as the model output, forming training sample pairs. To ensure that the training data covers different pollution levels, the training sample pairs are divided into training and validation sets by sampling according to the heavy metal standard concentration sequence distribution. The fusion features in the training set are processed using normalized parameters consistent with those used in constructing the dose-response relationship matrix to avoid inconsistencies between training and application. The support vector regression model is trained using a linear kernel function or a radial basis function kernel function, with the kernel function type determined by error comparison in the validation set. During training, the penalty factor and regression tolerance interval parameters are selected through the validation set to ensure consistent predictive stability of the support vector regression model across different concentration ranges. After training, the model parameters and normalized parameters of the support vector regression model are solidified into a heavy metal concentration mapping model. During the runtime phase, the same normalization processing is applied to the fusion features to be tested before inputting them into the heavy metal concentration mapping model to output predicted heavy metal concentration values. By using fusion features as a unified input and introducing a biological signal baseline as a measurement benchmark, the impact of soil matrix differences and sensor batch differences on concentration prediction can be reduced, providing a stable data foundation for subsequent pollution level determination and ecological risk level assessment.

[0082] Microbial sensor signals from soil samples to be tested are acquired, biotoxicity response features are extracted, the biotoxicity response features are input into a support vector machine model to output prediction confidence, the information fusion weights are corrected, the biotoxicity response features are fused to obtain the fused features to be tested, the fused features to be tested are input into a heavy metal concentration mapping model to output the predicted heavy metal concentration, the predicted heavy metal concentration is compared with the pollution level threshold sequence to determine the pollution level, and the ecological risk level is determined based on the pollution level.

[0083] Soil extracts were prepared from the soil samples to be tested, and microbial sensor signals were collected. After denoising and baseline drift compensation, biotoxicity response features were extracted. These biotoxicity response features were input into a support vector machine model to obtain prediction confidence scores. The prediction confidence scores were converted into weight correction coefficients, and the information fusion weights and uniform weights were linearly mixed and normalized to obtain corrected information fusion weights. The biotoxicity response features were then fused according to the corrected information fusion weights to obtain the fused features to be tested. The biosignal deviation was determined by comparing the fused features to be tested with the biosignal baseline. The fused features to be tested were input into a heavy metal concentration mapping model to obtain predicted heavy metal concentrations. The pollution level was determined based on a pollution level threshold sequence, and the ecological risk level was determined based on the pollution level.

[0084] The prediction confidence is determined by the classification confidence of the output of the soil sample to be tested by the support vector machine model. The corrected information fusion weights include weighting and updating the information fusion weights according to the prediction confidence. The biological signal bias is determined by the difference between the fusion feature to be tested and the biological signal baseline. The pollution level is determined by comparing the predicted heavy metal concentration with the pollution level threshold sequence. A correspondence is established between the ecological risk level interval threshold sequence and the pollution level. The potential ecological risk index is calculated by the predicted heavy metal concentration, the background heavy metal concentration and the heavy metal toxicity coefficient, and the potential ecological risk index is compared with the ecological risk level interval threshold sequence to determine the level.

[0085] In this embodiment, after preparing a soil extract from the soil sample to be tested, microbial sensor signals are collected and biotoxicity response features are extracted. The calculation caliber of the biotoxicity response features is consistent with that of the biotoxicity response spectrum dataset. The biotoxicity response features of the soil sample to be tested are then processed according to the normalization parameters from the training phase and input into a support vector machine model. The support vector machine model outputs decision values, and the prediction confidence is determined by these decision values. The prediction confidence is used to characterize the class separability and discrimination reliability of the current soil sample under the existing dose-effect relationship matrix.

[0086] In this embodiment, the prediction confidence level is determined using the following calculation expression:

[0087]

[0088] To predict confidence levels; The decision value output by the support vector machine model for the soil sample to be tested.

[0089] When correcting the information fusion weights, the weight update coefficients are first determined based on the prediction confidence, so that the information fusion weights converge towards a more uniform distribution when the prediction confidence decreases, thereby suppressing the fusion bias caused by a single feature anomaly. In this embodiment, the corrected information fusion weights are determined using the following calculation expression:

[0090]

[0091] For the first Corrected information fusion weights for each biotoxicity response feature; For the first Information fusion weights for individual biological toxicity response characteristics; Update the coefficients for the weights; This represents the number of biotoxicity response characteristics.

[0092] Among them, the weight update coefficient is from Confirmed. After completing the corrected information fusion weights, the fusion feature to be tested is obtained by weighted summation of the aforementioned fusion features, and the biological signal deviation is determined by comparing the fusion feature to be tested with the biological signal baseline.

[0093]

[0094] Biological signal bias; The fusion feature to be tested; This serves as the baseline for biological signals.

[0095] The heavy metal concentration mapping model uses the fusion features to be tested as input and outputs predicted heavy metal concentrations. When determining the pollution level, the predicted heavy metal concentrations are compared with a pollution level threshold sequence from low to high to determine the threshold interval into which the predicted heavy metal concentration falls, and the level corresponding to this threshold interval is taken as the pollution level. When determining the ecological risk level, firstly, a set of corresponding ecological risk level intervals is selected based on the pollution level. Then, a potential ecological risk index is calculated using the predicted heavy metal concentrations, background heavy metal concentrations, and heavy metal toxicity coefficients. The potential ecological risk index is then mapped to the ecological risk level intervals to output the ecological risk level. In this embodiment, the potential ecological risk index is determined using the following calculation expression:

[0096]

[0097] As a potential ecological risk index; The heavy metal toxicity coefficient; These are predicted values ​​for heavy metal concentrations. The background concentration of heavy metals is represented by this parameter. By introducing a weight update driven by prediction confidence and an interval mapping constrained by pollution level, the stability of the fusion features can be maintained when the reliability of sample discrimination fluctuates, and the output of ecological risk level can be kept consistent with the concentration prediction value and pollution level.

[0098] like Figure 2 As shown, a rapid detection system for soil heavy metal ecological risk microorganisms is used to implement the rapid detection method for soil heavy metal ecological risk microorganisms. The system includes:

[0099] The biosignal acquisition module is used to acquire microbial sensor signals from soil samples, extract biotoxicity response characteristics, and construct a biotoxicity response spectrum dataset. Hardware-wise, the module consists of a microbial sensor unit and an electrochemical measurement front-end. The microbial sensor unit includes an immobilized microbial electrode, a counter electrode, a reference electrode, and a matching reaction chamber and fluid interface to ensure stable contact between the soil extract and the immobilized microbial electrode. The electrochemical measurement front-end includes a constant potential control circuit, a transimpedance amplifier circuit, an excitation and sampling circuit, and an analog-to-digital conversion circuit to synchronously output electrochemical current and electrochemical impedance signals, and employs shielding and grounding structures to reduce power frequency interference. The module has a built-in clock synchronization and storage unit to record microbial sensor signals according to a unified sampling sequence, providing the raw data foundation for subsequent extraction of biotoxicity response characteristics and construction of the biotoxicity response spectrum dataset.

[0100] The model building module is used to establish a dose-response matrix from the biotoxicity response spectrum dataset, train a support vector machine model using the dose-response matrix to obtain a stability index, determine information fusion weights based on the stability index, fuse biotoxicity response features to obtain fusion features, statistically analyze the fusion features to obtain a biological signal baseline, and establish a heavy metal concentration mapping model according to the correspondence between heavy metal standard concentrations and fusion features. The model building module consists of a data processing unit and a storage unit in hardware. The data processing unit can use an embedded processor or industrial computing unit, possessing floating-point operation capabilities and vector operation acceleration capabilities, used to organize and manage the biotoxicity response spectrum dataset and establish a dose-response matrix, execute support vector machine model training to obtain a stability index, and complete the calculation of information fusion weights and generation of fusion features. The storage unit is used to store the biotoxicity response spectrum dataset, dose-response matrix, support vector machine model parameters, information fusion weights, biological signal baseline, and heavy metal concentration mapping model parameters. The module exchanges data with the biological signal acquisition module through a communication interface and provides an external interface for importing heavy metal standard concentration sequences and outputting modeling results.

[0101] The risk assessment module acquires microbial sensor signals from the soil sample to be tested, extracts biotoxicity response features, inputs these features into a support vector machine model to output predicted confidence levels, corrects the information fusion weights, fuses the biotoxicity response features to obtain the fused features to be tested, inputs these fused features into a heavy metal concentration mapping model to output predicted heavy metal concentrations, compares the predicted heavy metal concentrations with a pollution level threshold sequence to determine the pollution level, and determines the ecological risk level based on the pollution level. The risk assessment module consists of a real-time inference unit, input / output interfaces, and a human-computer interaction unit. The real-time inference unit can share the same data processing unit with the model building module or use a separate processor to receive data from the soil sample to be tested. The system extracts biotoxicity response features from microbial sensor signals, inputs these features into a support vector machine model to output predicted confidence levels, and adjusts the information fusion weights accordingly. It then generates the fusion features to be tested and calculates the biosignal deviation against the biological signal baseline. Finally, it calls a heavy metal concentration mapping model to output predicted heavy metal concentrations, thereby determining the pollution level and ecological risk level. The input / output interface is used to access the raw signal data from the acquisition module and retrieve model parameters from the storage unit. It also supports outputting predicted heavy metal concentrations, pollution levels, and ecological risk levels to a display terminal or host computer system. The human-computer interaction unit presents the judgment results in graphical or textual form on-site and indicates the data quality status.

[0102] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0103] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A rapid detection method for soil heavy metal ecological risk microorganisms, characterized in that, Includes the following steps: Acquire microbial sensor signals from soil samples, extract biotoxicity response features, and construct a biotoxicity response spectrum dataset; A dose-response matrix was established using the biotoxicity response spectrum dataset. A support vector machine model was trained using the dose-response matrix to obtain a stability index. The information fusion weights were determined based on the stability index. Biotoxicity response features were fused to obtain fusion features. The fusion features were statistically analyzed to obtain the biosignal baseline. A heavy metal concentration mapping model was established according to the correspondence between heavy metal standard concentrations and fusion features. The microbial sensor signal of the soil sample to be tested is acquired, the biotoxicity response features are extracted, the biotoxicity response features are input into the support vector machine model to output the prediction confidence, the information fusion weight is corrected, the biotoxicity response features are fused to obtain the fused features to be tested, the fused features to be tested are input into the heavy metal concentration mapping model to output the heavy metal concentration prediction value, the heavy metal concentration prediction value is compared with the pollution level threshold sequence to determine the pollution level, and the ecological risk level is determined based on the pollution level. Microbial sensor signals include electrochemical current signals and electrochemical impedance signals; Extracting biotoxicity response characteristics includes: determining the growth rate inhibition curve using electrochemical current signals and electrochemical impedance signals, and determining the enzyme activity inhibition gradient using electrochemical current signals and electrochemical impedance signals.

2. The rapid detection method for soil heavy metal ecological risk microorganisms according to claim 1, characterized in that, The construction of the biotoxicity response spectrum dataset includes: acquiring microbial sensor signals from soil samples with known heavy metal standard concentrations, extracting biotoxicity response features, and establishing a correspondence between the biotoxicity response features and heavy metal standard concentrations to construct the biotoxicity response spectrum dataset.

3. The rapid detection method for soil heavy metal ecological risk microorganisms according to claim 1, characterized in that, Establishing the dose-response matrix involves arranging the biotoxicity response characteristics according to different levels of heavy metal standard concentrations. The rows of the dose-response matrix correspond to the heavy metal standard concentrations, and the columns of the dose-response matrix correspond to the biotoxicity response characteristics.

4. The rapid detection method for soil heavy metal ecological risk microorganisms according to claim 1, characterized in that, Training a support vector machine model includes training with a linear support vector machine model. The stability metric is determined by the minimum classification margin between the classification hyperplane of the linear support vector machine model and the training samples.

5. The rapid detection method for soil heavy metal ecological risk microorganisms according to claim 4, characterized in that, Determining the information fusion weights based on stability indices includes obtaining the feature weights of the linear support vector machine model and performing normalization to obtain the information fusion weights. The fusion of biotoxicity response features to obtain fusion features includes weighting and summing the biotoxicity response features according to the information fusion weights.

6. The rapid detection method for soil heavy metal ecological risk microorganisms according to claim 1, characterized in that, The biological signal baseline is obtained by statistically analyzing the fusion features of the biotoxicity response spectrum dataset. The heavy metal concentration mapping model is established by training a support vector regression model according to the correspondence between the heavy metal standard concentration and the fusion features.

7. The rapid detection method for soil heavy metal ecological risk microorganisms according to claim 1, characterized in that, The prediction confidence is determined by the classification confidence of the output of the soil sample to be tested by the support vector machine model. The corrected information fusion weights include weighting and updating the information fusion weights according to the prediction confidence. The biological signal bias is determined by the difference between the fusion feature to be tested and the biological signal baseline. The pollution level is determined by comparing the predicted heavy metal concentration with the pollution level threshold sequence. A correspondence is established between the ecological risk level interval threshold sequence and the pollution level. The potential ecological risk index is calculated by the predicted heavy metal concentration, the background heavy metal concentration and the heavy metal toxicity coefficient, and the potential ecological risk index is compared with the ecological risk level interval threshold sequence to determine the level.

8. A rapid detection system for soil heavy metal ecological risk microorganisms, used to implement the rapid detection method for soil heavy metal ecological risk microorganisms according to any one of claims 1-7, characterized in that, The system includes: The biosignal acquisition module is used to acquire microbial sensor signals from soil samples, extract biotoxicity response characteristics, and construct a biotoxicity response spectrum dataset. The model building module is used to establish a dose-response relationship matrix from the biotoxicity response spectrum dataset, train a support vector machine model using the dose-response relationship matrix to obtain a stability index, determine the information fusion weights based on the stability index, fuse biotoxicity response features to obtain fusion features, statistically analyze the fusion features to obtain the biological signal baseline, and establish a heavy metal concentration mapping model according to the correspondence between heavy metal standard concentration and fusion features. The risk assessment module is used to acquire microbial sensor signals from the soil sample to be tested, extract biotoxicity response features, input the biotoxicity response features into the support vector machine model to output the prediction confidence, correct the information fusion weights, fuse the biotoxicity response features to obtain the fused features to be tested, input the fused features to be tested into the heavy metal concentration mapping model to output the predicted heavy metal concentration, compare the predicted heavy metal concentration with the pollution level threshold sequence to determine the pollution level, and determine the ecological risk level based on the pollution level.

Images