An AI-based large-area farmland soil antibiotic high-throughput detection analysis method, system and application

Abstract

The application belongs to the technical field of environmental pollutant detection and analysis, and discloses a large-area farmland soil antibiotic high-throughput detection and analysis method, system and application based on AI, which comprises the following steps: GIS grid-based sampling point layout, sample collection and preservation, sample pretreatment, antibiotic extraction, sample purification, high-throughput detection, detection verification, time series data construction and intelligent analysis. The application makes full use of existing laboratory resources, can simultaneously detect more than 30 kinds of antibiotics, constructs a time series-GIS grid-component-concentration multidimensional data set, and adopts various AI models including time series analysis, Kriging spatial interpolation and PCA component analysis to perform spatio-temporal analysis and risk early warning. The application can realize large-area, long-time series and high-throughput farmland soil antibiotic monitoring and intelligent analysis, and provides a scientific basis for soil new pollutant treatment and food safety guarantee.

Description

Technical Field

[0001] This invention belongs to the field of environmental pollutant detection and analysis technology, specifically relating to an AI-based method, system, and application for high-throughput detection and analysis of antibiotics in large-area farmland soil. Background Technology

[0002] Antibiotics, as an emerging class of pollutants, are widely present in farmland soils, mainly originating from irrigation with livestock and poultry wastewater, application of organic fertilizers, and the use of antibiotic pesticides. Antibiotics in farmland soils not only affect the structure and ecological functions of soil microbial communities but may also pose a potential threat to human health through the food chain. Therefore, monitoring and analyzing antibiotics in farmland soils is of great significance for ensuring the quality and safety of agricultural products and the health of the ecological environment.

[0003] Several methods for detecting antibiotics in soil have emerged in the existing technology, such as the method for determining antibiotics in environmental soil disclosed in CN119667060A; the rapid and efficient method for simultaneously detecting the content of 11 antibiotics in soil and sludge disclosed in CN106468691A; and the rapid high-throughput detection method for antibiotics in soil and sediment samples disclosed in CN113899836A. However, when addressing the need for high-throughput detection and analysis of antibiotics in large-area farmland soils, the following problems still exist: (1) Lack of standardized sampling schemes for large areas: When setting up sampling points for farmland soil in large areas, there is a lack of scientific basis, making it difficult to ensure the regional representativeness of sampling and data, and it is difficult to balance the cost and efficiency of sampling, transportation and testing. (2) Limited coverage of detection methods: Existing detection methods can usually only detect 10-20 antibiotics at the same time, and the detection speed is slow and the accuracy is low, which makes it difficult to meet the needs of large-scale, high-throughput detection of large areas, multiple time points, long time series and batch samples; (3) The pretreatment process is complicated and time-consuming: Traditional solid phase extraction methods, processes and equipment are complicated to operate, and it is difficult to automate each detection process, resulting in low efficiency. (4) Few data dimensions and scattered and isolated management: There is a lack of a unified platform for multi-dimensional spatiotemporal data collection, analysis and management, such as sampling and detection. Data management is not standardized, and it is difficult to collect and comprehensively apply multi-dimensional data for different crops and different planting areas over a long period of time and over a large area. (5) Insufficient existing laboratory and intelligent analysis capabilities: Traditional sampling, detection and statistical methods do not make full use of existing laboratory capabilities and their precision detection instruments. On the one hand, it is difficult to process a large number of samples quickly, efficiently and at low cost. On the other hand, it is difficult to effectively explore the spatiotemporal distribution patterns, risk characteristics and changing trends in massive multi-dimensional, large-area and multi-batch monitoring data. This results in long time consumption and high cost for large-scale and routine testing, making it difficult to balance analysis efficiency, accuracy, cost and timeliness. (6) Limited application areas and value: Currently, due to poor batch sampling and detection capabilities, low data dimensionality and insufficient accuracy, and low detection frequency, it is mainly used for sampling and detection in environmental supervision, making it difficult to achieve large-scale, regular, and continuous large-scale sampling and high-throughput, high-precision detection. It is also difficult to meet the various application needs in monitoring new pollutants in farmland soil, and it is also difficult to meet the application needs of building a basic database of antibiotics in farmland environment, which severely limits its application areas and data value.

[0004] In summary, there is an urgent need to develop a high-throughput and intelligent analysis technology system for antibiotics in large-area farmland soil that can simultaneously detect multiple antibiotics, has a scientific and reasonable sampling scheme, rapid and efficient pretreatment, standardized and unified data management, and intelligent and advanced analysis methods. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-based method, system, and application for high-throughput detection and analysis of antibiotics in large-area farmland soil. This invention fully utilizes existing laboratory resources and combines various technical means such as system planning of detection and sampling points, automation of sample processing procedures, high-throughput detection using UPLC-MS / MS+HPLC, and data processing and intelligent analysis based on AI models to solve the aforementioned technical problems in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A high-throughput detection and analysis method for antibiotics in large-area farmland soil based on AI includes the following steps: S1. GIS grid-based detection / sampling point layout: A combination of hierarchical and system-based point layout is adopted. The existing GIS of the testing laboratory is used as a distributed regional testing center. The large area of ​​farmland is divided into multiple planting areas, and then hierarchically according to the planting mode. Sampling points are deployed in each planting area using a grid system to ensure coverage of plots with different fertility levels, irrigation conditions and planting years. S2. Sample Collection and Preservation: Collect soil samples in batches according to the preset sampling depth and sampling method, number and record the GIS, planting information and sampling time information of the sampling points, and generate unique coding information; prepare two equal copies of the batch-collected samples, transport them separately under refrigerated and light-proof sealed conditions, and preserve them under refrigeration and light-proof conditions after returning to the testing laboratory. One copy is transported to the nearest testing laboratory for initial testing, and the other copy is transported to the adjacent testing laboratory for retesting. S3. Sample batch pretreatment: Each testing laboratory uses an automated sample pretreatment platform to automatically process batches of collected soil samples in parallel, including coding and identification, weighing, mixing, freeze drying, grinding and sieving, antibiotic extraction, and sample purification, to obtain multiple test solutions in batches. S4. High-throughput detection: Multiple test solutions are tested in parallel. Ultra-high performance liquid chromatography-triple quadrupole mass spectrometry is used to detect antibiotics in each test solution. More than 30 antibiotics are tested simultaneously. Quality control is performed. Samples that are determined to be positive are then verified by HPLC. The detection data of each test sample are obtained and linked with their coding information to form correlation data, which is then uploaded to the cloud server. S5. Multidimensional Data Construction and Intelligent Analysis: The cloud server receives, processes, and stores related data. After analysis and correction using a pre-trained AI model, it organizes and summarizes the test data from various planting areas, multiple batches, and multiple time points according to the composite dimensions of target antibiotic, planting area, crop type, GIS grid, sampling time, testing laboratory, and testing instrument code. This constructs a multidimensional dataset of antibiotic components, planting area, crop type, time series, GIS, antibiotic concentration, and testing laboratory. Finally, it analyzes the spatiotemporal distribution characteristics and variation patterns of target antibiotics in large-scale farmland, as well as the fluctuation range of test data from different laboratories, and outputs visualized analysis results according to user requests.

[0007] A high-throughput detection and analysis system for antibiotics in large-area farmland soil based on AI, used to implement the aforementioned method, includes: Multiple sampling terminals, automated sample pretreatment platform, UPLC-MS / MS and HPLC-UV / FLD detectors, detection terminals and cloud server; The sampling terminal has a built-in GIS positioning module and a timestamp module, which are used to record the GIS and time records of batch sampling of soil samples at each sampling point at the front end, and generate coded information for each soil sample, including the GIS of the collection point and date information. The automated sample pretreatment platform, UPLC-MS / MS, HPLC-UV / FLD detectors, and detection terminals are all located within the distributed regional detection center. The automated sample pretreatment platform handles the entire process of batch sample processing automatically, while the UPLC-MS / MS and HPLC-UV / FLD detectors are used for high-throughput batch detection and validation. The detection terminals have a built-in data import module for connecting with the UPLC-MS / MS and HPLC-UV / FLD detectors to acquire high-throughput UPLC-MS / MS and HPLC validation data for each sample, along with associated data formed by their coded information, and upload this data to the cloud server.

[0008] This invention also provides the application of the above-mentioned methods and systems in monitoring new pollutants in farmland soil, as well as their application in constructing a basic database of antibiotics in farmland environment.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) High-throughput detection capability: The method of the present invention can detect 39 antibiotics at the same time, and the detection throughput is much higher than that of existing methods (usually 10-22), which greatly improves the detection efficiency and coverage. (2) Scientific sampling scheme: adopt a combination of hierarchical and systematic sampling, and deploy points based on GIS grid to ensure the regional representativeness and spatial comparability of the data; (3) Rapid and efficient pretreatment: Dispersive solid phase extraction (d-SPE) technology is used to replace traditional solid phase extraction. It is simple to operate, time-saving, and low-cost, and is suitable for large-scale sample processing. (4) Standardized time series data management: Construct multidimensional datasets to realize integrated management of spatiotemporal data, laying the foundation for comprehensive analysis of long time series and large areas; (5) Intelligent analysis capability: Using AI models for spatiotemporal analysis and risk warning, it can automatically mine data patterns, identify pollution hotspots, and predict pollution trends, providing a scientific basis for decision-making; (6) Excellent detection performance: the method recovery rate is 58%-126%, and the limit of quantitation is 0.05-2.50 μg / kg, which meets the sensitivity requirements for the detection of antibiotics in farmland soil. Attached Figure Description

[0010] Figure 1 This is a schematic diagram illustrating the composition and logical relationship of the high-throughput detection and analysis system according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the high-throughput detection and analysis method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the multidimensional data construction process in the high-throughput detection and analysis method of this invention. Figure 4This is a flowchart illustrating the intelligent early warning process in the high-throughput detection and analysis method of this invention. Figure 5 A schematic diagram illustrating the spatial distribution of antibiotic concentrations in farmland soil within the detection area according to an embodiment of the present invention. Detailed Implementation

[0011] The following is in conjunction with the appendix Figure 1-5 The present invention provides a clear and complete description of the technical solutions in its embodiments, including multiple examples.

[0012] Example 1 See Figures 1-4 The AI-based high-throughput detection and analysis method for antibiotics in large-area farmland soil provided in this embodiment includes the following steps: S1. GIS-based grid-based detection / sampling point layout: In accordance with the requirements of HJ / T 166 and GB / T 36197, a combination of hierarchical and system-based point layout is adopted. The existing GIS of the testing laboratory is used as a distributed regional testing center. The large area of ​​farmland is divided into multiple planting areas, and then hierarchically according to the planting pattern. Sampling points are laid out in each planting area using a grid system to ensure coverage of plots with different fertility levels, irrigation conditions and planting years.

[0013] Specifically, the stratified sampling points include: S1-1, using the GIS locations of multiple testing laboratories with UPLC-MS / MS and HPLC capabilities as distributed regional testing centers, dividing large areas of farmland into multiple planting areas based on the principle of proximity-based initial testing and cross-testing, and setting up at least one distributed regional testing laboratory within each planting area or within 200 kilometers of that planting area. Multiple distributed regional testing centers cooperate with each other, each independently completing the proximity-based initial testing and cross-testing of batch samples and obtaining the corresponding testing data; S1-2, dividing each planting area into multiple sampling units; S1-3, stratifying each sampling unit according to more than 15 crop types, including sugarcane, rice, wheat, corn, vegetables, and fruit trees, with at least 3 sampling points set up for each planting type, and appropriately increasing the point density in key polluted areas.

[0014] S2. Sample Collection and Preservation: Soil samples are collected in batches according to the preset sampling depth and sampling method. The GIS, planting information, and sampling time information of the sampling points are recorded and a unique code is generated. The batch-collected samples are prepared into two equal portions and transported separately under refrigeration at 4℃ and light-proof sealed conditions. After being transported back to the testing laboratory, they are stored at 0-4℃ in the dark. One portion is transported back to the nearest testing laboratory for initial testing, and the other portion is transported back to the adjacent testing laboratory for retesting. The sampling depth is as follows: the soil sampling depth for cultivated layer is 0-20cm, and for planting fruit trees and forest crops, it is 0-60cm. For each sampling point, a multi-point mixing method is used to collect soil from 5-8 sub-sampling points and combine them into one mixed sample. The sampling amount is not less than 1kg.

[0015] S3. Sample Batch Pretreatment: Each testing laboratory uses an automated sample pretreatment platform (such as Gerstel MPS, Biotage Extrahera, etc., which can process 24-96 samples in parallel per batch) to automatically process the batch-collected soil samples in parallel, including: S3-1, coding and identification, weighing, mixing, freeze-drying, grinding and sieving to obtain multiple samples to be tested; S3-2, Antibiotic Extraction: Each sample to be tested is processed in parallel, using sodium citrate buffer with pH=3 and disodium EDTA as the extraction solution, combined with 5% formate acetonitrile for ultrasonic extraction, centrifugation to collect the supernatant, and repeating the extraction 1-2 times and combining the supernatants; S3-3, Sample Purification: Each sample to be tested is processed in parallel, using dispersion solid phase extraction for purification, using purification tubes pre-filled with PSA, C18 spherical packing material and anhydrous sodium sulfate for purification, nitrogen blowing and reconstitution, and filtration to obtain the test solution. Automated sample pretreatment platforms such as Gerstel MPS and Biotage Extrahera can be used to achieve batch parallel processing of samples, with a single batch capable of processing 24-96 samples.

[0016] S4. High-throughput detection: Multiple test solutions are detected in parallel. Ultra-high performance liquid chromatography-triple quadrupole mass spectrometry (UPLC-MS / MS) is used to detect antibiotics in each test solution. 39 antibiotics are detected simultaneously. Detection quality control is performed, and samples that are determined to be positive are verified by HPLC. The detection data of each test sample are obtained, and together with their coding information, they form correlation data, which is then uploaded to the cloud server.

[0017] The UPLC-MS / MS detection conditions were as follows: a 2.1 mm × 100 mm, 1.7 μm C18 column was used, with a column temperature of 30-40℃ (preferably 35℃), an injection volume of 2 μL, and a flow rate of 0.3 mL / min; mobile phase A was 0.1% formic acid aqueous solution, and mobile phase B was methanol solution, using a gradient elution program; the mass spectrometer was in positive ion mode with dynamic multiple reaction monitoring, a drying gas temperature of 325℃, a flow rate of 10 L / min, a nebulizer gas pressure of 50 psi, and a capillary voltage of +4000 V.

[0018] The steps for quality control of the test include: setting a standard curve for each batch of samples with a correlation coefficient r ≥ 0.990; setting at least one blank sample for each batch with the target component concentration below the detection limit; adding at least 5% of the matrix as a standard sample for each batch with a recovery rate ranging from 58% to 126%; and determining the midpoint of the standard curve for every 20 samples analyzed, with a relative deviation ≤ 20%.

[0019] The steps for HPLC detection validation include: for samples screened positive by UPLC-MS / MS (i.e., detecting any tetracycline compound with a LOQ ≥ 1), HPLC validation is initiated; the instrument is HPLC-UV / FLD, and the detection conditions are: a 250×4.6mm, 5μm C18 column, column temperature 35℃, mobile phase A is 0.05mol / L ammonium oxalate solution (pH 2.5), mobile phase B is acetonitrile, gradient is 0~5min 10%B→5~20min 30%B→20~25min 90%B→25~30min 10%B, flow rate 0.8mL / min, injection volume 20μL, UV 350nm.

[0020] The 39 antibiotics tested simultaneously covered four major categories: (1) Sulfonamides (19 types): sulfachlorpyridazine, sulfacetamide, sulfadoxine, sulfadiazine, sulfaguanidine, sulfamethoxypyrimidine, sulfamethoxypyrimidine, sulfamethoxazole, sulfadiazine, sulfapyridine, sulfaquinoxaline, sulfathiazole, benzoylsulfonamide, sulfadimethoxazole, sulfamethoxypyrimidine, sulfamethoxazole, sulfadiazine ... (1) Isopyrimidine; (2) Quinolones (14 types): Ciprofloxacin, Sinoxacin, Difloxacin, Enrofloxacin, Enrofloxacin, Flumethin, Flurofloxacin, Normefloxacin, Nasalidine, Norfloxacin, Ofloxacin, Oxyquinic acid, Sarafloxacin, Pefloxacin; (3) Macrolides (6 types): Azithromycin, Clarithromycin, Erythromycin, Lincomycin, Roxithromycin, Theylmycin; (4) Others (1 type): Trimethoprim.

[0021] S5. Multidimensional Data Construction and Intelligent Analysis: The cloud server receives, processes, and stores related data. After analysis and correction using a pre-trained AI model, the data is then organized and summarized according to a composite dimension of target antibiotic, planting area, crop type, GIS grid, sampling time, testing laboratory, and testing instrument code. This constructs a multidimensional dataset of antibiotic composition, planting area, crop type, time series, GIS, antibiotic concentration, and testing laboratory. Finally, the spatiotemporal distribution characteristics, variation patterns, and fluctuation range of different laboratory testing data of the target antibiotic in a large area of ​​farmland are analyzed. Visualized analysis results are output according to user requests. The specific steps include the following: S5-1. Obtaining multi-dimensional data for preliminary testing: The cloud server associates and stores the spatial location information, sampling time information, and test data obtained by the preliminary testing laboratory to form a multi-dimensional data structure containing time, space, component, and concentration dimensions. Each data point includes the planting area, sampling unit, sampling point GIS, sampling time, antibiotic component and its detection concentration value, thus obtaining multi-dimensional data for preliminary testing.

[0022] S5-2. Obtain the associated re-inspection multi-dimensional data: Repeat step S5-1 to process the data from the re-inspection laboratory, and then associate it with the associated initial inspection multi-dimensional data to obtain the associated re-inspection multi-dimensional data.

[0023] S5-3. AI Model Correction and Construction of Multi-Dimensional Dataset: The AI ​​model analyzes and corrects the associated initial and re-examination multi-dimensional data, identifies and cross-compares the detection data with deviations exceeding a preset threshold, and deletes or modifies the deviated data to obtain multi-dimensional data. The data is then summarized, sorted by time sequence, and a multi-dimensional dataset including antibiotic components, planting areas, crop types, time sequence, GIS, antibiotic concentration, and testing laboratory information is constructed.

[0024] S5-4. Based on this multidimensional dataset, intelligent analysis is performed using an AI model, including the following seven analytical components: (1) Time series analysis: The seasonal variation patterns and interannual variation trends of antibiotic concentrations were analyzed using the ARIMA model or LSTM neural network time series analysis model; (2) Spatial distribution analysis: Spatial interpolation algorithms such as Kriging interpolation or inverse distance weighted interpolation were used to construct a spatial distribution heat map of antibiotic concentration; (3) Component distribution analysis: Principal component analysis (PCA) or random forest model is used to analyze the differences in total antibiotic concentration in soil at different locations, reflecting the sources or accumulation of antibiotic pollution in soil in different regions; the contribution rate of major pollutants can be extracted by PCA dimensionality reduction, and the characteristic importance of each antibiotic component to the difference in total concentration can be evaluated by random forest model. (4) Hotspot identification: High-risk pollution areas are identified using machine learning classification models, and spatial hotspot areas are identified using the Getis-Ord Gi* statistical method; (5) Cluster analysis: Cluster analysis algorithms such as DBSCAN or K-means are used to classify polluted areas and identify farmland areas with similar pollution characteristics; (6) Analysis of fluctuation range of test data: The ANOVA method or the robust statistical (MAD) model is used to analyze the fluctuation range of test data of different laboratories and test instruments, and to determine or adjust the preset deviation threshold. The ANOVA method can compare the significance of inter-group differences in test data between different laboratories, and the MAD (median absolute deviation) method can robustly estimate the degree of data dispersion, thereby scientifically determining the correction threshold. (7) Risk warning: Based on historical and current data, predict pollution trends and generate warning information when the predicted value exceeds the safety threshold.

[0025] S5-5. Based on user requests, and according to the spatiotemporal distribution characteristics, change patterns, predicted pollution trends, and risk warning results of target antibiotics in large-area farmland obtained after intelligent analysis, output visualized analysis results, including GIS maps, time series charts, and statistical analysis reports.

[0026] A high-throughput detection and analysis system for antibiotics in large-area farmland soil based on AI, used to implement the aforementioned method, includes: Multiple sampling terminals, automated sample pretreatment platform, UPLC-MS / MS and HPLC-UV / FLD detectors, detection terminals and cloud server; The sampling terminal has a built-in GIS positioning module and a timestamp module, which are used to record the GIS and time records of batch sampling of soil samples at each sampling point at the front end, and generate coded information for each soil sample, including the GIS of the collection point and date information. The automated sample pretreatment platform, UPLC-MS / MS, HPLC-UV / FLD detectors, and detection terminals are all located within the distributed regional detection center. The automated sample pretreatment platform handles the entire process of batch sample processing automatically, while the UPLC-MS / MS and HPLC-UV / FLD detectors are used for high-throughput batch detection and validation. The detection terminals have a built-in data import module for connecting with the UPLC-MS / MS and HPLC-UV / FLD detectors to acquire high-throughput UPLC-MS / MS and HPLC validation data for each sample, along with associated data formed by their coded information, and upload this data to the cloud server.

[0027] The cloud server has a built-in detection data analysis program based on a pre-trained AI model, which includes the following eight modules: (1) Pre-trained AI model module: includes a variety of pre-trained AI models, covering time series models such as ARIMA, LSTM, and Prophet, spatial analysis models such as Kriging interpolation and inverse distance weighted interpolation, component analysis models such as principal component analysis (PCA) and random forest, clustering models such as DBSCAN and K-means, hotspot analysis models such as Getis-Ord Gi*, and statistical analysis models such as ANOVA and robust statistics (MAD), which are used to provide callable AI models for other modules to work.

[0028] (2) Basic data acquisition module: used to collect and store sampling location information, sampling time information and antibiotic detection data of farmland soil samples.

[0029] (3) Data processing module: Based on the pre-trained AI model, the module verifies, summarizes, processes, analyzes and corrects the uploaded related data; identifies and cross-compares the detection data with deviations greater than the preset threshold, and deletes or modifies the deviation data.

[0030] (4) Multidimensional Dataset Construction Module: Based on multiple composite dimensions including sampling time, space, detection object, and detection concentration, the detection data from various planting areas, multiple batches, and multiple time points are organized and summarized to construct a multidimensional dataset of antibiotic components, planting areas, crop types, time series, GIS, antibiotic concentration, and testing laboratories. The multidimensional dataset supports efficient querying based on spatial location and time range, including spatial index, time index, component index, concentration data table, and metadata table.

[0031] (5) Spatiotemporal Feature Analysis Module: Based on pre-trained AI models and multidimensional datasets, this module analyzes the temporal variation patterns and spatial distribution characteristics of antibiotic pollution. It employs time series analysis models to analyze the seasonal and interannual variation patterns of antibiotic concentrations; uses spatial interpolation algorithms to construct spatial distribution heatmaps of antibiotic concentrations; uses principal component analysis (PCA) or random forest models for component distribution analysis; uses machine learning classification models to identify high-risk pollution areas; uses the Getis-Ord Gi* statistical method to identify spatial hotspots; uses cluster analysis algorithms to identify farmland areas with similar pollution characteristics; and uses analysis of variance (ANOVA) or robust statistical (MAD) models to analyze the fluctuation range of detection data from different laboratories and testing instruments.

[0032] (6) Intelligent early warning module: Based on the pre-trained AI model, it analyzes the historical detection data and the current detection multidimensional dataset of the target planting area, predicts the antibiotic pollution trend of the target planting area and generates early warning information.

[0033] (7) Decision support module: Based on pre-trained AI models and multidimensional datasets, assess the degree and risk level of antibiotic pollution in farmland soil, generate targeted suggestions for the treatment of new soil pollutants and the governance of farmland ecology and food security based on pollution characteristic analysis, and evaluate the implementation effect of the treatment measures.

[0034] (8) Visualization output module: Based on pre-trained AI models and multidimensional datasets, according to user requests, output visualization analysis results in the form of maps, charts and reports, including: displaying sampling points and pollution distribution on GIS maps; displaying the changing trend of antibiotic concentration using time series charts; and generating various statistical analysis reports.

[0035] Figure 1 The three-tier architecture of the sampling terminal layer, the distributed regional detection center (including an automated sample pretreatment platform, UPLC-MS / MS, HPLC-UV / FLD and detection terminal), and the cloud server is demonstrated, as well as the data flow and control flow relationships between the modules.

[0036] Figure 2 The complete process of the five main steps S1 to S5 and their sub-steps is shown, including the entire chain from the layout of sampling points to the final visualization output.

[0037] Figure 3 It demonstrates the complete data flow from the input of data from the initial testing laboratory and the retesting laboratory, through the construction of a multidimensional dataset after AI model correction, to the seven intelligent analyses and the final visualization output.

[0038] Figure 4It demonstrates a complete closed-loop early warning process, from multidimensional dataset input, data preprocessing, AI model selection and training, time series analysis and trend prediction, risk level determination, early warning information generation and push, to decision support and governance recommendations.

[0039] The methods and systems of this invention can be applied to the monitoring of new pollutants in farmland soil, specifically including: large-area surveys of the current status of antibiotic pollution in farmland soil; tracing the sources of antibiotic pollution in farmland soil; risk assessment of antibiotic pollution in farmland soil; evaluation of the effectiveness of antibiotic pollution control in farmland soil; and early warning of food security risks. They can also be applied to the construction of a basic database of antibiotics in the farmland environment, which supports the formulation of policies for the control of new pollutants in farmland, the supervision of agricultural product quality and safety, and decision-making on sustainable agricultural development.

[0040] Example 2 This invention provides an AI-based method and system for high-throughput detection and analysis of antibiotics in farmland soil in a large-area rice-growing region of a province. It is a specific application of Example 1. Based on Example 1, the method specifically includes the following steps: 1. Sampling point layout The main rice-producing areas of Guangdong Province were selected as the testing area. Following the requirements of HJ / T 166 and GB / T 36197, a stratified sampling method based on a GIS grid was adopted. Firstly, the existing testing laboratories in Guangzhou (GIS: WGS84 POINT(113.500764,23.257302)) and Shenzhen were used as distributed regional testing centers. Based on the principle of initial testing at the nearest location and cross-testing, the testing area was divided into 12 sampling units, with one sampling point in each unit, for a total of 12 sampling points. The sampling points covered plots with different fertility levels (high, medium, low), irrigation conditions (river irrigation, well irrigation, rainfed), and planting ages (less than 5 years, 5-15 years, more than 15 years) to ensure regional representativeness. GPS was used to record the latitude and longitude coordinates of each sampling point, and the sampling terminal automatically generated unique coded information containing GIS coordinates and sampling time.

[0041] 2. Sample collection and preservation Sampling tools were stainless steel tools cleaned and dried with methanol to avoid cross-contamination. Following the sample collection requirements in HJ / T 166, section 6.2.3, samples were collected in March 2024 (before spring plowing). The topsoil layer was sampled at a depth of 0-20 cm. A multi-point mixing method was used at each sampling point, collecting soil from five sub-sampling points and combining them into one mixed sample, with a sample volume of approximately 1.5 kg. Immediately after collection, the samples were thoroughly mixed with clean stainless steel tools to remove foreign objects such as branches and stones. The samples were then placed in clean brown glass bottles, sealed with aluminum foil, and roughly divided into quarters for later use. Each sample was divided into two portions: one was transported to a nearby testing laboratory in Guangzhou for initial testing, and the other to a neighboring testing laboratory in Shenzhen for retesting. During transportation, the samples were kept refrigerated at 4℃, protected from light, and sealed tightly, avoiding violent shaking to prevent sample deterioration or loss of antibiotic components. The samples were returned to the laboratory within 6 hours. Store in the laboratory at 0-4℃ away from light, and complete the pretreatment within 3 days (or at the latest within 1 week, referring to section 8.1, "Sample Collection and Preservation," of T / GDSES 3—2022). 3. Sample pretreatment After removing foreign matter from fresh soil, it was manually mixed in a stainless steel pan and reduced to quarters using the quartering method. The reduced samples were then placed in a freeze dryer for at least 48 hours, with the freeze-dried volume being at least three times the analytical volume, to avoid sample contamination during the freeze-drying process. After freeze-drying, the samples were ground in an agate mortar and passed through a 60-mesh (0.25 mm) standard soil sieve. The sieved samples were then sealed and stored in a light-proof container until extraction and analysis (refer to section 8.2, Sample Preparation, of T / GDSES 3—2022).

[0042] 4. Antibiotic extraction and purification (a) Type of extract Main extraction buffer: sodium citrate buffer at pH 3, combined with disodium ethylenediaminetetraacetate (NaEDTA), used for preliminary extraction of antibiotics from samples and inhibition of metal ion interference.

[0043] Assisted extraction solution: 5% formate acetonitrile, used to enhance the extraction efficiency of the target antibiotic, in combination with sodium citrate buffer.

[0044] (b) Extraction time and conditions Extraction Procedure: Accurately weigh 5g of uniformly ground sample and place it in a centrifuge tube. Add 0.4g of disodium ethylenediaminetetraacetate (NaEDTA), 5mL of pH=3 sodium citrate buffer, and an internal standard mixture (concentration 100ppb). Vortex to mix and incubate overnight (approximately 12-24h) in a 4℃ refrigerator protected from light. The next day, add 10mL of 5% formate acetonitrile to the system and vortex for 1min to thoroughly mix the sample, followed by ultrasonic extraction for 15min. Add 4g of anhydrous sodium sulfate and 1g of sodium chloride to the extract and vortex for 3min to salt out. Centrifugation Conditions: Centrifuge at 4℃ and 18000rpm for 5min. Collect the supernatant and repeat the extraction procedure 1-2 times. Combine all supernatants for later use.

[0045] (c) Cleaning materials Dispersive solid-phase extraction was used for purification. The purification tube was pre-filled with 50 mg of ethylenediamine-N-propylsilane (PSA), 150 mg of C18 spherical packing material, and 900 mg of anhydrous sodium sulfate.

[0046] (d) Purification process The collected supernatant was carefully transferred to the purification tube and vortexed for 3 min for dispersion solid-phase extraction purification, followed by centrifugation at 4 °C and 5000 rpm for 5 min.

[0047] Transfer the supernatant after centrifugation to a clean glass test tube, blow it with nitrogen at 40°C until nearly dry, add 500 μL of ultrapure water-acetonitrile mixed solution (volume ratio 9:1) to reconstitute, and vortex to mix.

[0048] (e) Filtration After reconstitution, the sample was filtered through a 0.22 μm organic phase filter membrane and collected in a brown sample vial for analysis.

[0049] In this embodiment, the specific steps are as follows: Accurately weigh 5g of uniformly ground sample and place it in a 50mL centrifuge tube. Add 0.4g Na2EDTA, 5mL of pH=3 sodium citrate buffer, and an internal standard mixture (concentration 100ppb). Vortex to mix and incubate overnight (approximately 16h) in a 4°C refrigerator away from light. The next day, add 10mL of 5% formate acetonitrile, vortex for 1 min, and then sonicate for 15 min. Add 4g of anhydrous sodium sulfate and 1g of sodium chloride, vortex for 3 min for salting out. Centrifuge at 4°C and 18000rpm for 5 min and collect the supernatant. Repeat the extraction once and combine the supernatants. Transfer the supernatant to a pre-filled purification tube containing 50mg PSA, 150mg C18, and 900mg anhydrous sodium sulfate, vortex for 3 min, and centrifuge at 4°C and 5000rpm for 5 min. Transfer the supernatant to a glass test tube, blow it with nitrogen at 40°C until nearly dry, add 500 μL of ultrapure water-acetonitrile mixed solution (volume ratio 9:1) to redissolve, filter through a 0.22 μm organic phase filter membrane, and collect it into a brown sample vial for testing.

[0050] 5. UPLC-MS / MS detection (a) Detection method Ultra-high performance liquid chromatography-triple quadrupole mass spectrometry (UPLC-MS / MS) was used, and some tetracyclines could be verified by high performance liquid chromatography (HPLC).

[0051] (b) Instrument parameters Liquid chromatography: C18 column (2.1 mm × 100 mm, 1.7 μm), column temperature 30-40℃, injection volume 2 μL, flow rate 0.3 mL / min. Mobile phase A was formic acid aqueous solution, mobile phase B was methanol solution, and the gradient elution program is shown in Table 1 below: Table 1

[0052] Mass spectrometry: positive ion mode, dynamic multiple reaction monitoring (DMRM) or multiple reaction monitoring (MRM). Drying gas temperature 325℃, flow rate 10L / min, nebulizer gas pressure 50psi, capillary voltage +4000V.

[0053] (c) Types of pharmaceuticals Chromatographic grade reagents: methanol, acetonitrile, formic acid, ammonium acetate.

[0054] Analytical grade reagents: disodium ethylenediaminetetraacetate, anhydrous sodium sulfate, sodium chloride, disodium hydrogen phosphate, citric acid.

[0055] Standards: 39 antibiotic standards (sulfonamides, quinolones, etc.) and 8 internal standards (lincomycin-d3, sulfamethoxazole-d4, etc.), with a purity ≥93.0%.

[0056] (e) The limit of detection and quantification shall be determined according to conventional methods. (f) Quality control methods Standard curve: correlation coefficient r≥0.990 (some standards≥0.995), at least 5 concentration points (2-50μg / L), quantification using internal standard method.

[0057] Blank test: at least one blank test per batch (≤20 samples), with the concentration of the target component below the detection limit.

[0058] Spiked recovery: At least 5% of the matrix is ​​spiked in each batch, with a recovery rate ranging from 70% to 160%.

[0059] Intermediate concentration test: For every 20 samples analyzed, the midpoint of the standard curve is determined. If the relative deviation is ≤20%, the curve is redrawn.

[0060] In this embodiment, a Shimadzu LC-40D ultra-high performance liquid chromatography-triple quadrupole mass spectrometer was used for detection. Chromatographic conditions: C18 column (2.1 mm × 100 mm, 1.7 μm), column temperature 35 °C, injection volume 2 μL, flow rate 0.3 mL / min. Mobile phase A was 0.1% formic acid aqueous solution, mobile phase B was methanol, and the gradient elution program was: 0-2 min 85% A, 2-4 min 85%-60% A, 4-7 min 60% A, 7-9.1 min 60%-20% A, 9.1-10 min 20%-10% A, 10-11 min 10% A, 11-11.1 min 10%-85% A, 11.1-13 min 85% A. Mass spectrometry conditions: positive ion mode, DMRM monitoring, drying gas temperature 325℃, flow rate 10L / min, nebulizer gas pressure 50psi, capillary voltage +4000V.

[0061] A standard curve (r≥0.995) is set for each batch, with one blank sample and 5% matrix standard (recovery rate range 65%-120%). The midpoint of the standard curve is determined for every 20 samples analyzed (relative deviation ≤15%).

[0062] 6. Test Results See Figure 5Of the 12 samples, 11 tested positive for antibiotics, a detection rate of 91.7%. A total of 18 antibiotics were detected, covering sulfonamides (7), quinolones (6), macrolides (4), and trimethoprim, with an average of 3.2 antibiotics detected per sample. The five most frequently detected antibiotics were: flumethinyl (detected 8 times, 66.7%), norfloxacin (detected 6 times, 50.0%), sulfacetamide (detected 5 times, 41.7%), sulfachlorpyridazine (detected 4 times, 33.3%), and roxithromycin (detected 4 times, 33.3%). The total antibiotic concentration ranged from ND-1858.23 ng / g, with an average concentration of 285.6 ng / g.

[0063] 7. Cross-validation of initial and re-inspection data The test data of 12 paired samples from the initial testing laboratory in Guangzhou and the retesting laboratory in Shenzhen were uploaded to a cloud server and cross-compared using the data processing module. The results showed that the average relative deviation between the initial and retest data was 8.3% (range 3.2%-15.7%), all of which were below the preset deviation threshold (20%), indicating that the test data from the two laboratories were in good agreement and no data correction or deletion was required.

[0064] 8. Spatial Distribution and Intelligent Analysis Based on the detection data and GIS coordinates, a spatial distribution heatmap of total antibiotic concentration was constructed using the Kriging interpolation method. The results showed that the northern and eastern parts of the detection area had higher antibiotic concentrations, forming obvious pollution hotspots (Getis-Ord Gi* Z-score > 1.96, p < 0.05), which highly correlated with the distribution of surrounding livestock and poultry farms, indicating that livestock and poultry wastewater is likely the main source of antibiotic pollution in this area. Principal component analysis (PCA) was used to analyze the component distribution. The first two principal components contributed a cumulative 78.5% of the total, with PC1 mainly consisting of quinolones (load > 0.75) and PC2 mainly consisting of sulfonamides (load > 0.70), indicating a difference in the sources of the two types of antibiotics.

[0065] Example 3 This invention provides an AI-based high-throughput detection and analysis method and system for antibiotics in farmland soil in a large-area rice-growing area of ​​a province. It is a specific application of Example 1, and based on Examples 1-2, it further monitors the temporal changes of antibiotics in soil in a facility vegetable area. The specific steps include: 1. Scheme Design A greenhouse vegetable production area in Shandong Province was selected as the testing area, with three fixed monitoring points established. Existing testing laboratories in Jinan, Qingdao, and Yantai served as distributed regional testing centers. Soil samples were collected quarterly (January, April, July, and October) from January 2023 to December 2024, for a total of eight rounds of sampling, yielding 48 soil samples. The main crop type in this area is greenhouse vegetables (tomatoes, cucumbers, peppers, etc.), with groundwater irrigation as the primary method, and reclaimed water used in some areas.

[0066] 2. Sample collection and testing Sample collection, pretreatment, and UPLC-MS / MS detection were performed according to the method in Example 1. The sampling depth was 0-20 cm (topsoil layer), and 5-8 sub-sampling points were collected at each sampling point using a multi-point mixing method. Simultaneously, the GIS coordinates, sampling time, crop type, and irrigation method of each sampling point were recorded, and a unique code was automatically generated by the sampling terminal.

[0067] 3. Multidimensional data construction Data from eight rounds of sampling were organized to construct a time-series GIS grid multidimensional dataset. The dataset includes: spatial dimensions (latitude and longitude coordinates of 6 sampling points), temporal dimensions (8 sampling time points), component dimensions (detection results of 39 antibiotics), and concentration dimensions (quantitative concentration values ​​of each antibiotic), forming a 6×8×39 four-dimensional data structure. The multidimensional dataset also includes metadata information such as planting area, crop type, testing laboratory, and testing instrument code.

[0068] 4. Time-series variation analysis An LSTM neural network model was used to analyze the temporal variation of antibiotic concentrations. Model training parameters: input time window of four quarters, 64 hidden layer nodes, learning rate of 0.001, and 200 training epochs. Results showed: (1) Seasonal variation: The total concentration of antibiotics showed obvious seasonal fluctuations. The concentration was highest in summer (July), with an average concentration of 523.6 ng / g; and the concentration was lowest in winter (January), with an average concentration of 186.3 ng / g. This pattern was highly correlated with the planting cycle of greenhouse vegetables and the usage patterns of pesticides / veterinary drugs. The seasonal autoregression coefficient was significant (p < 0.01) when the ARIMA(2,1,1) model was used for verification.

[0069] (2) Interannual variation: The concentration of antibiotics in each quarter of 2024 increased compared with the same period in 2023, with an average increase of about 15% (range 8.6%-22.3%), indicating that antibiotic pollution in the region is on the rise. Among them, the average total concentration in July 2024 increased by 18.7% compared with July 2023, and the difference was statistically significant (paired t test, p=0.032).

[0070] (3) Prediction results: Based on the LSTM model, the antibiotic concentration in January 2025 was predicted to be 215.4 ng / g (95% confidence interval: 168.7-278.3 ng / g). The results showed that the predicted values ​​at 2 locations exceeded the yellow warning threshold (100 ng / g), and 1 location was close to the orange warning threshold (500 ng / g). It is recommended to strengthen monitoring and control.

[0071] 5. Cluster analysis K-means clustering (k=3) was used to classify the antibiotic composition characteristics of six monitoring sites. The clustering results showed that: Cluster 1 (2 sites) was dominated by sulfonamides, accounting for 62.3%; Cluster 2 (3 sites) was dominated by quinolones, accounting for 54.8%; and Cluster 3 (1 site) was a mixed pollution type with relatively uniform distribution across all clusters. The results from the DBSCAN algorithm were consistent with these findings.

[0072] 6. Risk Warning Based on time-series analysis results, the system automatically generates risk warning information. The warning level classification standards are: green (safe, concentration <100 ng / g), yellow (low risk, 100-500 ng / g), orange (medium risk, 500-1000 ng / g), and red (high risk, >1000 ng / g). Monitoring results in July 2024 showed that two monitoring sites reached the orange warning level, and one site reached the red warning level. The system has automatically pushed warning information to management departments through the intelligent warning module, including the warning level, site GIS coordinates, predicted concentration trend curves, and recommended control measures.

[0073] Example 4 This invention provides an AI-based high-throughput detection and analysis method and system for antibiotics in farmland soils of a large-area rice-growing region in a province. It is a specific application of Example 1, and based on Examples 1-3, further comparative analysis of antibiotics in farmland soils of multiple crop types is conducted. The method specifically includes the following steps: 1. Detection area and sampling Three provinces—Guangxi, Sichuan, and Shandong—were selected as the testing areas, covering five major crop types: sugarcane, citrus, greenhouse vegetables, rice, and wheat. Two or more existing testing laboratories in each region were designated as distributed regional testing centers. Three to five sampling points were set up for each crop type, resulting in a total of 63 soil samples collected. Testing laboratories in each province with UPLC-MS / MS capabilities served as distributed regional testing centers, with all sampling points located within 200 kilometers of each testing center. Sampling took place in April 2024, and sample collection, preservation, and transportation were carried out uniformly according to the method described in Example 1.

[0074] 2. High-throughput detection and quality control Sample pretreatment and UPLC-MS / MS detection were performed according to the method in Example 1, simultaneously detecting 39 antibiotics. For each batch (24 samples), a standard curve (r = 0.993-0.998), two blank samples (all below the limit of detection), and three matrix-spiked samples were set (recovery rates 62%-118%, mean 85.3%). The midpoint of the standard curve was determined for every 20 samples analyzed, with relative deviations ranging from 3.2% to 14.8% (all ≤20%). For the 11 samples that tested positive by UPLC-MS / MS, HPLC-UV / FLD validation was initiated, with a 100% validation concordance rate.

[0075] 3. Statistical analysis of test results Of the 63 samples, 60 tested positive for antibiotics, with an overall detection rate of 95.2%. The detection rates of antibiotics in farmland soils for different crop types are shown in Table 2 below. Table 2

[0076] This is closely related to its high-intensity use of pesticides and organic fertilizers.

[0077] 4. Cluster analysis and component distribution K-means clustering (k=3) was used to classify the antibiotic composition characteristics of 63 samples. The results showed that the samples could be divided into 3 clusters: Cluster 1 (High Sulfonamide Pollution): This cluster mainly includes samples from the greenhouse vegetable area (n=18), with sulfonamide antibiotics predominating, accounting for 62.5%±8.3%. Principal component analysis (PCA) results show that PC1 (contribution rate 52.3%) mainly represents sulfonamide pollution and is highly correlated with Cluster 1.

[0078] Cluster 2 (high quinolone contamination): mainly includes samples from citrus areas (n=15), with quinolone antibiotics predominating, accounting for 54.8%±7.6%. Random forest feature importance analysis showed that enrofloxacin and ciprofloxacin were key variables distinguishing Cluster 2 from other clusters (importance score >0.15).

[0079] Cluster 3 (mixed pollution type): mainly includes samples from sugarcane and rice fields (n=30). The distribution of various antibiotics is relatively uniform, with sulfonamides, quinolones and macrolides accounting for 35.2%, 33.8% and 31.0% respectively.

[0080] 5. Spatial Hotspot Identification Spatial hotspot analysis of total antibiotic concentrations at 63 sampling points was performed using the Getis-Ord Gi* statistical method. The results showed that some sampling points in the facility vegetable area of ​​Shandong Province (Z-score=2.85, p=0.004) and the citrus area of ​​Sichuan Province (Z-score=2.31, p=0.021) were significant spatial hotspots, indicating a high risk of antibiotic contamination in these areas.

[0081] 6. Analysis of data fluctuation range between laboratories Of the 63 samples, 31 underwent both initial and retesting. Analysis of variance (ANOVA) was used to compare the test data from the initial and retesting laboratories, and the results showed no significant difference between the two laboratories (F=1.23, p=0.273). The median absolute deviation (MAD) method was further used to calculate the data dispersion, with a MAD value of 12.6 ng / g. Based on this, a deviation threshold of 3 times the MAD (37.8 ng / g) was set for subsequent data correction.

[0082] 7. Governance Recommendations Based on cluster analysis and AI-powered intelligent analysis, the system automatically generates targeted governance recommendations through its decision support module: For greenhouse vegetable areas, the use of sulfonamide antibiotics should be strictly controlled, with recommendations to promote biopesticide alternatives and strengthen source control of organic fertilizers; for citrus areas, the use of quinolone antibiotics should be strictly controlled, with recommendations to strengthen veterinary drug management and ensure compliant discharge of livestock wastewater; for sugarcane and rice areas, comprehensive prevention and control measures should be implemented, with recommendations to optimize fertilization structures and irrigation methods, and reduce the frequency of antibiotic pesticide use. Simultaneously, the system assesses the risk level of each area based on a multidimensional dataset, generating a visualized GIS risk distribution map and statistical reports, providing a scientific basis for management departments to formulate differentiated governance plans.

[0083] The above embodiments of the present invention collectively construct a high-throughput detection and intelligent analysis technology system for antibiotics in large-area farmland soil, based on an AI model. This system features simultaneous detection of multiple antibiotics, a scientifically sound sampling scheme, rapid and efficient pretreatment, standardized and unified data management, and intelligent and advanced analysis methods. Antibiotic detection was performed on 63 soil samples from 21 regions, and the results of soil antibiotic pollution level classification are summarized in Table 3 (μg / kg) below. Table 3

[0084] The classification of antibiotic pollution levels in farmland soil in the above embodiments of this invention is mainly based on domestic field sample collection and existing literature research results. The average total concentration of antibiotics in the farmland soil tested in the embodiments of this invention is 16.9 ng / kg, with a wide range of concentrations (0.002-2260.64 μg / kg) and a detection rate of 100%, providing a large-scale (city, province, country) concentration background reference for classification. On the ecological risk benchmark, the internationally accepted 100 μg / kg is adopted as the ecotoxicity trigger value. Exceeding this value can easily lead to the spread of drug resistance genes, which is the core basis for the "mild to moderate" threshold in the classification of this invention. At the same time, Kong Weidong et al. pointed out that the concentration range of soil veterinary drug residues is from μg / kg to g / kg. In the existing literature, He Zhenxian, Tao Xueqin et al. measured the total antibiotic concentration in farmland soil in three coastal provinces as 0.40-396.8 μg / kg, and Guan Mingyin et al. monitored concentrations as high as 2.59-4.28 mg / kg in a polluted site in Handan, Hebei, providing a practical basis for the high concentration end of the classification.

[0085] The classification criteria for the degree of antibiotic contamination in farmland soil used in the various embodiments of this invention are as follows: - Clean: <10 ug / kg; - Mild: 10-100 ug / kg; - Moderate: 100-1000 ug / kg; - Severe: >1000 ug / kg; The classification results of antibiotic pollution levels in farmland soil, based on actual measurements and literature review, are as follows: - Severe: 3 (14.3%) - Moderate: 6 (28.6%); - Mild: 7 (33.3%); - Cleaning: 5 (23.8%).

[0086] The embodiments of this invention can simultaneously detect 39 antibiotics, with a detection throughput far exceeding that of existing methods (typically 10-22), significantly improving detection efficiency and coverage. A combination of hierarchical and systematic point deployment is employed, using a GIS grid for point placement. Distributed regional detection centers serve as hubs, with a 200-kilometer service radius for dividing planting areas, ensuring regional representativeness and spatial comparability of the data. A dual-track initial and re-inspection system effectively guarantees data quality. Dispersive solid-phase extraction (d-SPE) technology replaces traditional solid-phase extraction, combined with an automated parallel pre-processing platform (24-96 samples per batch), offering simple operation, short processing time, and low cost, suitable for large-scale sample processing. A seven-dimensional dataset (component-region-crop-time series-GIS-concentration-laboratory) is constructed to achieve integrated management and efficient indexing of spatiotemporal data, laying a data foundation for comprehensive analysis of long-term time series and large regions. Seven AI analysis methods are integrated—ARIMA / LSTM time series analysis, Kriging / IDW spatial interpolation, PCA / random forest component analysis, Getis-Ord... This invention utilizes Gi* hotspot identification, DBSCAN / K-means clustering analysis, ANOVA / MAD fluctuation range statistical analysis, and trend prediction and risk warning to automatically mine data patterns, identify pollution hotspots, analyze source characteristics, assess inter-laboratory data consistency, and predict pollution trends, providing a scientific basis for decision-making. The detection method has a recovery rate of 58%-126% and a quantitation limit of 0.05-2.50 μg / kg, meeting the sensitivity requirements for antibiotic detection in farmland soil. The initial-retest cross-validation mechanism effectively ensures the accuracy and reliability of the data. This invention is the first to systematically integrate a standardized sampling scheme based on GIS grids, a high-throughput detection method for 39 antibiotics, an initial-retest cross-validation mechanism, time-series-GIS grid-component-concentration multidimensional data construction, and seven AI-intelligent analyses, forming a complete large-area farmland soil antibiotic monitoring and analysis technology system. This solves several technical problems in existing technologies, such as low detection throughput, non-standardized sampling schemes, fragmented data management, and outdated analytical methods.

[0087] The above description is only a preferred embodiment of the present invention. Within the scope of the present invention, other technical solutions that are the same as or equivalent to them are all within the protection scope of the present invention.

Claims

1. A high-throughput detection and analysis method for antibiotics in large-area farmland soil based on AI, characterized in that, Includes the following steps: S1. GIS grid-based detection / sampling point layout: A combination of hierarchical and system-based point layout is adopted. The existing GIS of the testing laboratory is used as a distributed regional testing center. The large area of ​​farmland is divided into multiple planting areas, and then hierarchically according to the planting mode. Sampling points are deployed in each planting area using a grid system to ensure coverage of plots with different fertility levels, irrigation conditions and planting years. S2. Sample Collection and Preservation: Collect soil samples in batches according to the preset sampling depth and sampling method, number and record the GIS, planting information and sampling time information of the sampling points, and generate unique coding information; prepare two equal copies of the batch-collected samples, transport them separately under refrigerated and light-proof sealed conditions, and preserve them under refrigeration and light-proof conditions after returning to the testing laboratory. One copy is transported to the nearest testing laboratory for initial testing, and the other copy is transported to the adjacent testing laboratory for retesting. S3. Sample batch pretreatment: Each testing laboratory uses an automated sample pretreatment platform to automatically process batches of collected soil samples in parallel, including coding and identification, weighing, mixing, freeze drying, grinding and sieving, antibiotic extraction, and sample purification, to obtain multiple test solutions in batches. S4. High-throughput detection: Multiple test solutions are tested in parallel. Ultra-high performance liquid chromatography-triple quadrupole mass spectrometry is used to detect antibiotics in each test solution. More than 30 antibiotics are tested simultaneously. Quality control is performed. Samples that are determined to be positive are then verified by HPLC. The detection data of each test sample are obtained and linked with their coding information to form correlation data, which is then uploaded to the cloud server. S5. Multidimensional Data Construction and Intelligent Analysis: The cloud server receives, processes, and stores related data. After analysis and correction using a pre-trained AI model, it organizes and summarizes the test data from various planting areas, multiple batches, and multiple time points according to the composite dimensions of target antibiotic, planting area, crop type, GIS grid, sampling time, testing laboratory, and testing instrument code. This constructs a multidimensional dataset of antibiotic components, planting area, crop type, time series, GIS, antibiotic concentration, and testing laboratory. Finally, it analyzes the spatiotemporal distribution characteristics and variation patterns of target antibiotics in large-scale farmland, as well as the fluctuation range of test data from different laboratories, and outputs visualized analysis results according to user requests.

2. The method according to claim 1, characterized in that, The layered point arrangement in step S1 specifically refers to: S1-1. First, multiple testing laboratories with UPLC-MS / MS and HPLC capabilities are located in GIS locations as distributed regional testing centers. Based on the principle of proximity-based initial testing and cross-testing, large areas of farmland are divided into multiple planting areas. In each planting area or within 200 kilometers of that planting area, at least one distributed regional testing laboratory is set up. Multiple distributed regional testing centers cooperate with each other and independently complete the proximity-based initial testing and cross-testing of batch samples and obtain the corresponding test data. S1-2. Divide each planting area into multiple sampling units; S1-3. Each sampling unit is stratified according to more than 15 crop types, including sugarcane, rice, wheat, corn, vegetables, and fruit trees. At least 3 sampling points are set up for each planting type, and the sampling point density is appropriately increased in key pollution areas. The sampling depth in step S2 is as follows: the soil depth for the cultivated layer is 0-20cm, and for fruit trees and other crops, it is 0-60cm. For each sampling point, a multi-point mixing method is used to collect soil from 5-8 sub-sampling points, which are then combined into one mixed sample with a sampling volume of not less than 1kg.

3. The method according to claim 1, characterized in that, The automated parallel processing in step S3 includes: S3-1. Parallel coding, weighing, mixing, freeze-drying, grinding, and sieving are performed on the collected soil samples to obtain multiple samples to be tested. S3-2, Antibiotic extraction: Each sample to be tested was processed in parallel. Sodium citrate buffer with pH=3 was used in combination with disodium ethylenediaminetetraacetate as the extraction solution, and ultrasonic extraction was performed with 5% formic acidified acetonitrile. The supernatant was collected by centrifugation. The extraction was repeated 1-2 times and the supernatants were combined. S3-3 Sample purification: Each sample to be tested is processed in parallel, and is purified by dispersion solid phase extraction. Purification is carried out using purification tubes pre-filled with PSA, C18 spherical packing material and anhydrous sodium sulfate. After nitrogen blowing and reconstitution, the solution to be tested is obtained by filtration.

4. The method according to claim 1, characterized in that, The detection conditions for antibiotic detection in step S4 using UPLC-MS / MS are as follows: using a 2.1mm × 100mm, 1.7μm C5 microscope. 18 The chromatographic column was set at a temperature of 30-40℃, with an injection volume of 2 μL and a flow rate of 0.3 mL / min. Mobile phase A consisted of formic acid aqueous solution, and mobile phase B consisted of methanol solution, using a gradient elution program. The mass spectrometry was performed in positive ion mode with dynamic multiple reaction monitoring (DMRM), a drying gas temperature of 325℃, a flow rate of 10 L / min, a nebulizer gas pressure of 50 psi, and a capillary voltage of +4000 V. The steps for quality control in the detection include: setting a standard curve for each batch of samples with a correlation coefficient r ≥ 0.990; setting at least one blank sample for each batch with the target component concentration below the detection limit; adding at least 5% of the matrix as a standard sample for each batch, with a recovery rate ranging from 58% to 126%; and determining the midpoint of the standard curve for every 20 samples analyzed, with a relative deviation ≤ 20%. The steps for HPLC detection validation include: for samples that are positive for UPLC-MS / MS screening (i.e., detecting any tetracycline compound ≥ LOQ), HPLC validation is initiated; the instrument is HPLC-UV / FLD, and the detection conditions are: a 250×4.6mm, 5 μm column with C245mm diameter. 18 The column temperature was 35℃, and the mobile phase A was 0.05 mol / L ammonium oxalate solution; the mobile phase B was acetonitrile; the gradient was 0~5 min 10%B → 5~20 min 30%B → 20~25 min 90%B → 25~30 min 10%B; the flow rate was 0.8 mL / min; the injection volume was 20 μ; and the UV was 350 nm.

5. The method according to claim 1, characterized in that, The 30 or more antibiotics detected simultaneously in step S4 include: 19 sulfonamides: sulfachlorpyridazine, sulfacetyl, sulfadoxine, sulfadiazine, sulfaguanidine, sulfamethoxypyrimidine, sulfamethoxypyrimidine, sulfamethoxazole, sulfadiazine, sulfapyridine, sulfaquinoxaline, sulfathiazole, benzoylsulfonamide, sulfadimethoxazole, sulfamethylpyrimidine, sulfamethadiazole, sulfabenpyrazole, sulfadimethylisopyrimidine; 14 quinolones: ciprofloxacin, sinofloxacin, difloxacin, enrofloxacin, enoxacin, flumethylquine, fleroxacin, normefloxacin, nalidixic acid, norfloxacin, ofloxacin, oxaquinic acid, sarafloxacin, and pefloxacin; Six macrolides: azithromycin, clarithromycin, erythromycin dehydrate, lincomycin, roxithromycin, and tylosin; One other: trimethoprim.

6. The method according to claim 5, characterized in that, Step S5 specifically includes the following steps: S5-1. Obtaining multi-dimensional data for preliminary testing: The cloud server associates and stores the spatial location information, sampling time information, and test data of the samples obtained by the preliminary testing laboratory to form a multi-dimensional data structure containing time, space, component, and concentration dimensions. Each data point includes the planting area, sampling unit, sampling point GIS, sampling time, antibiotic component and its test concentration value, thus obtaining multi-dimensional data for preliminary testing. S5-2. Obtain the associated re-inspection multi-dimensional data: Repeat step S5-1 to process the data from the re-inspection laboratory, and then associate it with the associated initial inspection multi-dimensional data to obtain the associated re-inspection multi-dimensional data. S5-3. AI Model Correction and Construction of Multi-Dimensional Dataset: The AI ​​model analyzes and corrects the associated initial multi-dimensional data and associated re-examination multi-dimensional data, identifies and cross-compares the detection data with deviations greater than the preset threshold, and deletes or modifies the deviation data to obtain multi-dimensional data. The data is then summarized, sorted by time sequence, and a multi-dimensional dataset including antibiotic components, planting areas, crop types, time sequence, GIS, antibiotic concentration, and testing laboratory information is constructed. S5-4. Based on this multidimensional dataset, intelligent analysis is performed using an AI model, including: (1) Time series analysis: The seasonal variation patterns and interannual variation trends of antibiotic concentrations were analyzed using the ARIMA model or LSTM neural network time series analysis model; (2) Spatial distribution analysis: Spatial interpolation algorithms such as Kriging interpolation or inverse distance weighted interpolation were used to construct a spatial distribution heat map of antibiotic concentration; (3) Component distribution analysis: The principal component analysis method or random forest model is used to analyze the differences in total antibiotic concentration in soil at different locations, reflecting the source or accumulation of antibiotic pollution in soil in different regions; (4) Hotspot identification: High-risk pollution areas are identified using machine learning classification models, and spatial hotspot areas are identified using the Getis-Ord Gi* statistical method; (5) Cluster analysis: Cluster analysis algorithms such as DBSCAN or K-means are used to classify polluted areas and identify farmland areas with similar pollution characteristics; (6) Analysis of fluctuation range of test data: Using analysis of variance or robust statistical models, analyze the fluctuation range of test data of different laboratories and test instruments, and determine or adjust the preset deviation threshold. (7) Risk warning: Based on historical and current data, predict pollution trends and generate warning information when the predicted value exceeds the safety threshold; S5-5. Based on user requests, and according to the spatiotemporal distribution characteristics, change patterns, predicted pollution trends, and risk warning results of target antibiotics in large-area farmland obtained after intelligent analysis, output visualized analysis results.

7. A high-throughput detection and analysis system for antibiotics in large-area farmland soil based on AI, characterized in that, The method for implementing any one of claims 1 to 6 comprises: Multiple sampling terminals, automated sample pretreatment platform, UPLC-MS / MS and HPLC-UV / FLD detectors, detection terminals and cloud server; The sampling terminal has a built-in GIS positioning module and a timestamp module, which are used to record the GIS and time records of batch sampling of soil samples at each sampling point at the front end, and generate coded information for each soil sample, including the GIS of the collection point and date information. The automated sample pretreatment platform, UPLC-MS / MS, HPLC-UV / FLD detectors, and detection terminals are all located within the distributed regional detection center. The automated sample pretreatment platform is used for the fully automated processing of batch samples. The UPLC-MS / MS and HPLC-UV / FLD detectors are used for high-throughput batch detection and validation. The detection terminal has a built-in detection data import module for connecting with the UPLC-MS / MS and HPLC-UV / FLD detectors to acquire high-throughput UPLC-MS / MS and HPLC validation data of each sample, along with the associated data formed by their coding information, and uploads it to the cloud server. The cloud server has a built-in detection data analysis program based on a pre-trained AI model. The program includes: a pre-trained AI model module, a basic data acquisition module, a data processing module, a multidimensional dataset construction module, a spatiotemporal feature analysis module, an intelligent early warning module, a decision support module, and a visualization output module.

8. The AI-based high-throughput detection and analysis system for antibiotics in large-area farmland soil according to claim 7, characterized in that, The pre-trained AI model module includes a variety of pre-trained AI models, which are used to provide callable AI models for other modules to work. The basic data acquisition module is used to collect and store sampling location information, sampling time information, and antibiotic detection data of farmland soil samples; Data processing module: Based on a pre-trained AI model, it verifies, summarizes, processes, analyzes, and corrects the uploaded related data; Identify and cross-compare detection data with deviations exceeding a preset threshold, and delete or modify the deviating data. Multidimensional dataset construction module: Based on multiple composite dimensions including sampling time, space, detection object, and detection concentration, the detection data of various planting areas, multiple batches, and multiple time points are sorted and summarized to construct a multidimensional dataset of antibiotic components, planting areas, crop types, time series, GIS, antibiotic concentration, and testing laboratories. The multidimensional dataset supports efficient queries based on spatial location and time range, including spatial index, time index, component index, concentration data table and metadata table; The spatiotemporal feature analysis module analyzes the temporal variation patterns and spatial distribution characteristics of antibiotic pollution based on pre-trained AI models and multidimensional datasets; it uses time series analysis models to analyze the seasonal variation patterns and interannual trends of antibiotic concentrations; it constructs a spatial distribution heatmap of antibiotic concentrations using spatial interpolation algorithms; it performs component distribution analysis using principal component analysis or random forest models; it identifies high-risk pollution areas using machine learning classification models; it identifies spatial hotspots using the Getis-Ord Gi* statistical method; and it identifies farmland areas with similar pollution characteristics using clustering analysis algorithms. Analysis of variance or robust statistical models were used to analyze the fluctuation range of test data from different laboratories and testing instruments. Intelligent early warning module: Based on a pre-trained AI model, it analyzes historical detection data and current multidimensional datasets of the target planting area to predict antibiotic pollution trends in the target planting area and generate early warning information; Decision support module: Based on pre-trained AI models and multidimensional datasets, it assesses the degree and risk level of antibiotic pollution in farmland soil, generates targeted suggestions for the treatment of new soil pollutants and the management of farmland ecology and food security based on pollution characteristic analysis, and evaluates the implementation effect of the treatment measures. Visualization output module: Based on pre-trained AI models and multidimensional datasets, it outputs visualized analysis results in the form of maps, charts, and reports according to user requests, including: displaying sampling points and pollution distribution on GIS maps; displaying the changing trend of antibiotic concentrations using time-series charts; and generating various statistical analysis reports.

9. The application of the method according to any one of claims 1-6 or the system according to any one of claims 7-8 in the monitoring of new pollutants in farmland soil, characterized in that, It can be used for: large-scale surveys of antibiotic pollution status in farmland soil, tracing the sources of antibiotic pollution in farmland soil, risk assessment of antibiotic pollution in farmland soil, evaluation of the effectiveness of antibiotic pollution control in farmland soil, and early warning of food security risks.

10. The application of the method according to any one of claims 1-6 or the system according to any one of claims 7-8 in constructing a basic database of antibiotics in farmland environment, characterized in that, The basic database it constructs will be used to support the formulation of policies for the treatment of new pollutants in farmland, the supervision of agricultural product quality and safety, and decision-making on sustainable agricultural development.

Images