A water system sediment multi-parameter automatic detection system

Abstract

The application discloses a water system deposit multi-parameter automatic detection system, and relates to the technical field of automatic detection.The system comprises the following modules: a spectrum collection module that obtains non-destructive multi-dimensional spectrum data at a deposit; a sample collection module that pre-processes and separates trace deposit samples collected in situ; a detection and analysis module that is used for deep analysis of the multi-dimensional spectrum data, receives analysis results of a solution to be detected as a calibration reference, and synchronously outputs a plurality of preset target parameters and relative quantitative estimation results of unknown abnormalities; a simulation optimization module that is used for real-time simulation and prediction of the whole process of sample processing and analysis; and a data recording module that is used for trace storage of equipment data, operation logs and analysis results during system operation.The application realizes quantitative, automatic and high-throughput detection of a plurality of preset target parameters and unknown abnormalities of a deposit.

Description

Technical Field

[0001] This application relates to the field of automated detection technology, and in particular to an automated multi-parameter detection system for aquatic sediments. Background Technology

[0002] Aquatic sediments are loose particulate matter accumulated at the bottom of rivers, lakes, reservoirs, and other water bodies. They consist of mineral debris, organic residues, microbial communities, and pollutants adsorbed on them. As important "sinks" and "sources" in the aquatic environment system, sediments continuously enrich heavy metals, nutrients, and persistent organic pollutants from the overlying water bodies, and under certain conditions, release these substances into the water bodies, becoming endogenous pollution sources. Therefore, multi-parameter monitoring of aquatic sediments is an indispensable foundation for water quality assessment, pollution source tracing, and ecological risk assessment.

[0003] Currently, multi-parameter detection of sediments in aquatic systems mainly relies on traditional laboratory analytical methods. The typical process involves collecting sediment samples at target locations using a grab bucket or columnar sediment sampler, packaging them, transporting them to the laboratory, and undergoing pretreatment processes such as freeze-drying, grinding, sieving, and digestion extraction. Then, elemental analysis is used to determine total organic carbon and total nitrogen, atomic absorption spectrometry or inductively coupled plasma mass spectrometry is used to determine heavy metals, chemical methods or flow injection analysis is used to determine nutrients, and chromatography or chromatography-mass spectrometry is used to determine organic pollutants. When obtaining information on the microecological environment of the sediment surface is required, additional microscopic examination or biomarker analysis is necessary.

[0004] Therefore, there is an urgent need to develop an automated multi-parameter detection system for sediments in aquatic systems. Summary of the Invention

[0005] This invention provides an automated detection system for multiple parameters of aquatic sediments. Through in-situ acquisition of multimodal spectra, automatic pretreatment of trace samples, deep neural network fusion analysis, digital twin simulation scheduling, and full-process data traceability, it achieves quantitative, automated, and high-throughput detection of multiple preset target parameters and unknown anomalies in sediments.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an automated multi-parameter detection system for aquatic sediments, comprising: The spectral acquisition module is used to acquire non-destructive multidimensional spectral data from sediments. The sample collection module is used to pre-treat and separate trace sediment samples collected in situ to form a test solution; The detection and analysis module is used to perform in-depth analysis of multidimensional spectral data, receive the analysis results of the test solution as a calibration reference, and simultaneously output the relative quantitative estimation results of various preset target parameters and unknown anomalies. The simulation optimization module is used to perform real-time simulation and prediction of the entire sample processing and analysis process, and adaptively optimize the working sequence, task sequence and resource allocation of each module based on the simulation results. The data recording module is used to trace and store equipment data, operation logs, and analysis results during system operation.

[0007] The beneficial effects of the technical solution provided by this invention include at least the following: The method of this invention can directly acquire Raman, infrared and hyperspectral images, avoiding changes in the properties of samples during transportation and preservation, significantly improving the timeliness and in-situ representativeness of the detection results. It also uses a deep neural network to perform feature fusion analysis of multimodal spectra, and can simultaneously output multiple parameters such as mineral composition, total organic carbon, total nitrogen, organic debris and microbial mat area ratio in a single detection. At the same time, the anomaly identification unit automatically marks unknown signal peaks outside the known spectral library and performs semi-supervised tracing and relative quantification, breaking through the limitation of traditional methods that can only detect preset target objects.

[0008] This invention covers a complete pretreatment process from grading, centrifugation, digestion to filtration and volume determination. The simulation optimization module uses digital twin simulation and adaptive scheduling algorithms to dynamically optimize the sample processing sequence, resource allocation, and the working time of each module. Under the premise of meeting time and resource constraints, it minimizes the total completion time or maximizes equipment utilization, thereby improving system operating efficiency. Through data standardization, chain-linked storage, and a digital sample archive, it completely preserves the original spectral data, operation logs, analysis results, and model versions of each sample. It supports forward and reverse traceability along the entire process of sampling, pretreatment, detection, and analysis, providing verifiable technical assurance for the authenticity and integrity of environmental monitoring data. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart of the automated multi-parameter detection system for water sediments provided in an embodiment of the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0012] Example An automated multi-parameter detection system for sediments in aquatic systems.

[0013] Please refer to Figure 1 This is a flowchart of the automated multi-parameter detection system for water sediments provided in an embodiment of the present invention.

[0014] I. Spectrum Acquisition Module The spectral acquisition module integrates three spectral probes: a miniature Raman spectral probe, a near-infrared / mid-infrared spectral probe, and a hyperspectral imaging probe. These three probes are connected to the surface or profile of the sediment sample via a corrosion-resistant fiber bundle, enabling the acquisition of spectral information at different depths. The miniature Raman spectral probe uses a 785nm or 532nm excitation source to acquire Raman scattering spectra in the wavenumber range of 200-3200cm⁻¹, reflecting the characteristic peak positions of mineral lattice vibrations and organic functional groups. The near-infrared / mid-infrared spectral probe 102 operates at 4000-10000cm⁻¹ (near-infrared) or 400-4000cm⁻¹ (mid-infrared), sensitively detecting the total organic carbon (TOC) and total nitrogen (TN) content through characteristic absorption bands. The hyperspectral imaging probe performs area scanning in the visible-near-infrared (400-1000nm) or short-wave infrared (1000-2500nm) bands to acquire a hyperspectral image cube for each pixel, visually presenting the two-dimensional spatial distribution of organic debris, mineral particles, and microbial mats. All acquired raw spectral and image data undergo preprocessing steps, including: spatial registration of Raman, infrared, and hyperspectral images, unifying them to the same spatial coordinate system using a standard checkerboard and feature point matching algorithm; then, baseline correction (using adaptive iterative reweighted penalized least squares), noise filtering (using Savitzky-Golay smoothing or wavelet denoising), and normalization (such as standard normal variable transformation) are performed to eliminate instrument response differences. After preprocessing, three independent and spatially aligned data matrices are obtained: the Raman spectrum matrix, the infrared spectrum matrix, and the hyperspectral image data cube. These data will serve as input to the detection and analysis module.

[0015] II. Sample Collection Module The sample collection module comprises a sample grading unit, a centrifugation and dehydration unit, a sample digestion unit, and a filtration and volume adjustment unit connected in sequence. The sample grading unit features multi-stage stainless steel standard sieves with progressively smaller mesh sizes from top to bottom, continuously sprayed with pure water from a top spray system. This allows the trace sediment samples collected in situ to pass through each stage of the sieves sequentially under the water flow. Coarse particles of the corresponding particle size are retained on the sieves and collected for optional heavy mineral analysis or discharged. The fine particle suspension passing through the bottom sieve enters the collection container. The centrifugation and dehydration unit centrifuges the fine particle suspension at 3000-5000 rpm, discarding the supernatant to obtain a dehydrated sediment filter cake. The sample digestion unit uses a closed-loop microwave digester with a PTFE digestion vessel equipped with an automatic lid opening and closing mechanism and a pressure sensor. After the sediment filter cake is transferred to the digestion tank, digestion is automatically performed according to a preset heating program (e.g., first raising the temperature to 150°C and holding for 10 minutes, then raising it to 180°C and holding for 20 minutes). A pressure sensor monitors the tank pressure in real time; if overpressure occurs, the pressure is automatically released and an alarm is triggered. After digestion, the tank is opened after cooling. The filtration and volume adjustment unit uses a vacuum filtration device with a 0.22-micron or 0.45-micron pore size filter membrane to remove residue from the digestion solution. The filtrate is collected in a graduated volume adjustment bottle, and water is added to the mark using a level detection device and a peristaltic pump to complete the volume adjustment, forming the solution to be tested.

[0016] The test solution is transmitted to the chemical analysis unit in the detection and analysis module. The chemical analysis unit is an inductively coupled plasma optical emission spectrometer (ICP-OES) used to accurately determine the actual concentrations of total organic carbon, total nitrogen, and major and trace elements in the solution. These high-precision measurements play a dual role in this system: during the model training phase before system deployment, they are paired with the spectral data of the corresponding samples to form a labeled training dataset for training the deep neural network model; during the online operation phase, chemical analysis can be performed on one sample at a preset interval (e.g., every 20 samples), and the measured values ​​serve as a reference benchmark to verify the accuracy of the prediction results from the spectral analysis unit. If the deviation exceeds a threshold, it triggers an adaptive model update or an early warning.

[0017] III. Detection and Analysis Module The spectral analysis unit incorporates a pre-trained deep neural network model. This model employs a multi-branch feature extraction and attention fusion architecture. Specifically, the Raman spectroscopy branch is a first-dimensional convolutional neural network (1D-CNN), which performs convolution and pooling operations along the Raman shift axis to automatically extract mineral feature peaks and organic functional group feature peaks, outputting a Raman feature vector. The infrared spectroscopy branch is a second-dimensional convolutional neural network or a fully connected network, which extracts feature absorption bands related to TOC and TN along the wavenumber axis, outputting an infrared feature vector. The hyperspectral image branch is a three-dimensional convolutional neural network (3D-CNN), which simultaneously extracts the spectral features of a single pixel and the spatial texture features of adjacent pixels, outputting a hyperspectral feature vector. These three feature vectors are input to the attention fusion layer, which calculates the attention weight coefficients of each vector for the current prediction task and performs a weighted sum of the three vectors based on these weights to generate a unified fused feature vector. The fused feature vector is simultaneously fed into multiple parallel prediction heads: the first classification head (Softmax activation) outputs the probability distribution of mineral types; the second classification head (Softmax activation) outputs the probability distribution of organic matter functional group types; the first regression head (linear activation) outputs the TOC content; the second regression head (linear activation) outputs the TN content; and the third to fifth regression heads (all linear activation) output the area proportions of organic debris, mineral particles, and microbial debris, respectively. Thus, for each pixel in the hyperspectral image, the model can generate a complete set of predicted physical, chemical, and biological parameters. After traversing all pixels, a quantitative map of the chemical composition of sediment micro-regions with spatial distribution is formed.

[0018] The pre-training process of the deep neural network model is as follows: First, a large number of sediment samples from different regions and of different types are collected. Raman, infrared, and hyperspectral images are obtained using a spectral acquisition module as input data. Simultaneously, standard laboratory methods (such as combustion oxidation for TOC / TN determination, X-ray diffraction for mineral identification, and chemical digestion-ICP for elemental analysis) are used to determine the mineral type, organic matter functional group type, TOC content, TN content, and the true values ​​of each area percentage for the corresponding samples, serving as labels. All data pairs are divided into training, validation, and test sets according to the samples, with a recommended ratio of 70%:20%:10%. A weighted joint loss function is constructed: for two classification tasks, the cross-entropy loss function is used; for five regression tasks, the mean squared error loss function or the mean absolute error loss function is used; the total loss is the weighted sum of the losses from each task according to preset weights (such as equal weights or adaptively adjusted based on task uncertainty). During training, the training set is input into the model in batches. Forward propagation is used to calculate the predicted values, and backpropagation is used to calculate the gradient of the joint loss with respect to the model parameters. The Adam optimizer is then used to update the parameters. Repeat this process until the joint loss converges on the validation set, and select the model version with the lowest loss on the validation set as the final pre-trained model.

[0019] The anomaly identification unit is used to handle unknown anomalous signals beyond the conventional predictions of the spectral analysis unit. First, the characteristic peaks in the preprocessed Raman and infrared spectra are matched against a standard spectral library of known substances (e.g., using Pearson correlation coefficient or cosine similarity). Characteristic peaks with a matching degree below a preset threshold (e.g., 0.85) are marked as anomalous signal peaks, and their peak position, peak intensity, half-maximum width, and spectral source are recorded. Next, a joint feature space is constructed containing the feature vectors of labeled known substance peaks and the feature vectors of unlabeled anomalous signal peaks. Semi-supervised clustering algorithms (e.g., constrained seed K-means clustering combined with a small number of expert-labeled anomalous samples) are used for clustering. Based on the spatial distance relationship between each cluster and the known substance anchor cluster, the suspected source category of each anomalous signal peak cluster is inferred, and new temporary codes for unknown pollutants are assigned to independent clusters that cannot be associated with known categories. Multiple anomalous signal peaks that meet the spatiotemporal proximity condition (Raman and infrared acquisition points fall within the same hyperspectral pixel and the time difference is within a preset window) and their corresponding hyperspectral anomalous pixels are aggregated into an anomalous event, generating a structured anomalous event record. Finally, the characteristic peaks of substances with known and stable content in the spectrum (such as the 464 cm-1 Raman peak of quartz) are selected as internal standard peaks. The ratio of the peak area of ​​the abnormal signal peak to the peak area of ​​the internal standard peak is calculated, and the relative response ratio is output as the relative quantitative value of the abnormal substance.

[0020] The quantitative output of the spectral analysis unit and the abnormal event records of the anomaly identification unit 320 are entered into the confidence assessment and integration process together. For each predicted value, a score is given from three dimensions: model self-confidence (the classification result takes the maximum probability value of the Softmax layer, and the regression result uses the Monte Carlo dropout method, which is used during inference with the Dropout layer enabled in the network, 50 forward propagations performed, the variance of the predicted value is calculated, and the variance is normalized and mapped to the confidence score); cross-validation consistency (checking the difference between the Raman branch and the infrared branch for the predicted values ​​of common parameters such as TOC, the smaller the difference, the higher the consistency score); and physical boundary rationality (checking whether the sum of the proportions of each area is between 99% and 101%, and whether the content is non-negative and less than the theoretical maximum value). The three scores are weighted and summed to obtain a comprehensive confidence score, and the results are divided into high, medium, and low levels accordingly. High and medium confidence results are marked as passed or pending review, while low confidence results are spatiotemporally correlated and integrated with abnormal event records of the same sample and the same time window to form a standardized data package. The data packet contains both quantitative spectra of conventional parameters and descriptions and relative quantitative information of anomalous events.

[0021] IV. Simulation Optimization Module The simulation optimization module includes a data simulation unit and an adaptive scheduling unit. The data simulation unit establishes a digital twin model of the physical system. Based on the physical layout, operation sequence, historical operation logs, and equipment performance parameters of each unit in the sample collection and detection / analysis modules, it establishes state transition models and sample queue models for each unit. The state transition model defines units as having four states: idle, running, awaiting cleaning, and fault, and defines state switching conditions (such as the completion signal of the previous unit, the ready signal of the next unit, etc.). The sample queue model adopts a first-in-first-out (FIFO) rule. The current working status, sample location, task queue length, and equipment operating mode reported by each module are acquired in real time via the Internet of Things (IoT). This data is injected into the virtual mirror to maintain real-time synchronization between the digital twin and the physical system. Based on this, the discrete event simulation engine performs ultra-real-time simulation of the subsequent processing flow of all samples in the current queue at a simulation speed higher than physical time (e.g., 100 times faster), outputting the estimated entry and completion timestamps of each sample in each unit, the changing trend of the waiting queue length in each unit, and the estimated total time for the last sample to complete the entire process.

[0022] The adaptive scheduling unit receives the above prediction results and performs dynamic scheduling based on preset constraints and optimization objectives. Constraints include: timing constraints (samples must pass through each unit in the order of grading → centrifugation → digestion → filtration and volume adjustment → detection); resource constraints (only one sample is processed at a time in the same unit, and a single digestion unit can process a maximum number of samples per tank capacity); and integrity constraints (after a sample completes in a unit, its product must enter the next unit within a specified time limit, otherwise timeout reprocessing is triggered). The scheduling problem is formalized as an optimization problem with sample processing sequence, unit start-up timing, and resource allocation as decision variables, and a cost function is constructed. The cost function is a weighted sum of predicted performance indicators such as total completion time, average equipment utilization, average sample waiting time, and maximum waiting queue length. The weights are dynamically set according to the user-selected optimization objective (e.g., "shortest delivery time" or "highest output"). A genetic algorithm is used to generate multiple candidate scheduling schemes under the constraints, and each scheme is fed into a digital twin unit for parallel ultra-real-time simulation to obtain the predicted performance indicators for each scheme. The cost function values ​​of each scheme are compared, the optimal scheme is selected, and it is converted into executable operation instructions for each module and issued. For example, the sample collection module is sent with the specified sample insertion / removal sequence, and the detection and analysis module is sent with the signal to start the spectral acquisition sequence. The system continuously monitors the actual arrival / departure times recorded by the sample arrival sensors at the inlet and outlet of each unit. When the deviation between the actual execution time and the predicted time exceeds a preset threshold, or when a new high-priority sample (such as a sample that needs immediate re-inspection after anomaly identification) is inserted, rescheduling is triggered. Based on the latest state, the above optimization process is re-executed, an updated execution scheme is generated, and issued.

[0023] V. Data Recording Module The data recording module includes a data standardization unit, a storage management unit, and a sample archive. The data standardization unit connects to all modules via a pre-defined communication interface, receiving raw sensor data (such as temperature, pressure, and centrifugation speed), operation logs (such as "Sample #001 enters digestion unit"), analysis results (such as quantitative chromatograms and abnormal events), and model version information in real time. Based on the data source and field identifier, it automatically identifies the data category and calls the corresponding standardization template (such as JSON Schema), uniformly converting data of different formats into a pre-defined unified data exchange format (such as JSON-based structured records), and adding timestamps and module source markers.

[0024] After receiving standardized data records, the storage management unit calculates a data fingerprint for each record using the SHA-256 hash algorithm. Using the sample number as the primary association key and the timestamp as the secondary association key, the data fingerprint of each record is linked to the data fingerprint of the previous record for the same sample via pointers, forming a complete data traceability chain composed of multiple records linked end-to-end. The starting node of the traceability chain is the sample sampling record, and the ending node is the final analysis report. The original content, data fingerprints, and link pointers of all records are written to the distributed storage medium. When it is necessary to trace a sample, by traversing the chain structure from back to front or from front to back according to the sample number, all immutable data records of that sample from sampling, preprocessing, spectral detection to result output can be obtained completely.

[0025] The sample archive is automatically triggered upon completion of the entire sample processing flow. It retrieves all relevant data records for the sample by calling the traceability query interface of the storage management unit. This data is then structured and organized according to the detection stage (sampling, spectral acquisition, pretreatment, chemical analysis, spectral interpretation, anomaly identification, report generation) and data type (raw spectra, intermediate results, final report, logs), assembling into a complete digital archive containing basic sample information, raw spectrum sets, operation log summaries, analysis results, and model information. The archive is persistently stored in structured files (such as HDF5 or Parquet format) and indexed using the sample number and sampling timestamp as a combined primary key. The sample archive provides multi-dimensional search interfaces based on time range, spatial region, parameter value range, or anomaly event type, supporting rapid querying and comparative analysis by external systems.

[0026] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0027] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0028] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0029] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0030] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. An automated multi-parameter detection system for aquatic sediments, characterized in that, include: The spectral acquisition module is used to acquire non-destructive multidimensional spectral data from sediments. The sample collection module is used to pre-treat and separate trace sediment samples collected in situ to form a test solution; The detection and analysis module is used to perform in-depth analysis of multidimensional spectral data, receive the analysis results of the test solution as a calibration reference, and simultaneously output the relative quantitative estimation results of various preset target parameters and unknown anomalies. The simulation optimization module is used to perform real-time simulation and prediction of the entire sample processing and analysis process, and adaptively optimize the working sequence, task sequence and resource allocation of each module based on the simulation results. The data recording module is used to trace and store equipment data, operation logs, and analysis results during system operation.

2. The automated multi-parameter detection system for aquatic sediments as described in claim 1, characterized in that, The spectral acquisition module is used to acquire non-destructive multidimensional spectral data at the sediment, wherein: The spectral acquisition module integrates a miniature Raman spectral probe, a near-infrared / mid-infrared spectral probe, and a hyperspectral imaging probe. It acquires Raman, infrared, and hyperspectral images of sediments at different depths via optical fiber to form a non-destructive chemical fingerprint. The miniature Raman spectroscopy probe is used to acquire Raman characteristic spectra corresponding to functional groups of specific minerals and organic matter. The near-infrared / mid-infrared spectroscopy probe is used to detect the total organic carbon and total nitrogen content with high sensitivity. The hyperspectral imaging probe is used to perform area array scanning on the sediment surface to acquire two-dimensional distribution images of organic debris, mineral particles and microbial mats.

3. The automated multi-parameter detection system for aquatic sediments as described in claim 1, characterized in that, The sample collection module is used to pre-treat and separate trace sediment samples collected in situ to form a test solution, wherein: The sample collection module includes a sample grading unit, a centrifugation and dehydration unit, a sample digestion unit, and a filtration and volume determination unit. The sample grading unit is used to separate sediment samples into coarse-particle and fine-particle components according to particle size through a combination of multi-stage sieving and sedimentation. The centrifugal dehydration unit is used to separate the fine particles after classification into solid and liquid components to obtain a dehydrated sediment filter cake. The sample digestion unit is used to digest the transferred sediment samples under high temperature and high pressure conditions in a closed environment; The filtration and volume-regulating unit is used to filter the digestion solution to remove residues and adjust the volume to a preset level to form the solution to be tested.

4. The automated multi-parameter detection system for aquatic sediments as described in claim 1, characterized in that, The detection and analysis module is used for in-depth analysis of multidimensional spectral data, and receives the analysis results of the test solution as a calibration reference. It simultaneously outputs relative quantitative estimation results of various preset target parameters and unknown anomalies, including: The detection and analysis module includes a spectral analysis unit, a chemical analysis unit, and an anomaly detection unit; The spectral analysis unit has a built-in pre-trained deep neural network model, which is used to fuse features of pre-processed spectral data and simultaneously output the quantitative concentration of multiple preset target parameters. The chemical analysis unit is used to accurately determine the actual concentrations of total organic carbon, total nitrogen, and major and trace elements in the solution. The anomaly identification unit is used to identify abnormal signal peaks in the spectrum that deviate from the known feature library, and to perform source analysis and relative quantitative estimation of the abnormal signal peaks through semi-supervised learning.

5. The automated multi-parameter detection system for aquatic sediments as described in claim 4, characterized in that, The spectral analysis unit incorporates a pre-trained deep neural network model, wherein: The pre-training steps of the deep neural network model include: Construct a labeled training dataset, which includes Raman spectroscopy, infrared spectroscopy and hyperspectral images as input, and various preset target parameters determined by standard methods as labels; A joint loss function is constructed, which is a weighted combination of the cross-entropy loss function of the classification task and the mean squared error loss function of the regression task. The classification task corresponds to mineral type and organic matter functional group type, and the regression task corresponds to total organic carbon content, total nitrogen content, organic debris area ratio, mineral particle area ratio and microbial mat area ratio. The deep neural network model is iteratively optimized using the training dataset until the joint loss function converges; The model is optimized using a validation dataset to select the best one, thus obtaining a pre-trained model.

6. The automated multi-parameter detection system for aquatic sediments as described in claim 4, characterized in that, The output provides quantitative concentrations of various preset target parameters, among which: The quantitative concentration results output by the spectral analysis unit are evaluated for confidence level to determine the level of the results. The verified quantitative results are integrated with abnormal events and output as a standardized data packet. The confidence score is obtained by weighted summation of the model's own confidence score, cross-validation consistency score, and physical boundary reasonableness score.

7. The automated multi-parameter detection system for aquatic sediments as described in claim 1, characterized in that, The simulation optimization module is used to perform real-time simulation and prediction of the entire sample processing and analysis process, and adaptively optimizes the working sequence, task sequence, and resource allocation of each module based on the simulation results, wherein: The simulation optimization module includes a data simulation unit and an adaptive scheduling unit; The data simulation unit is used to construct a virtual image of the automated multi-parameter detection system for aquatic sediments, and to perform real-time dynamic simulation of the entire process from sample introduction, pretreatment, separation, detection to data output, and to predict the system behavior and sample processing progress at future moments based on the current state. The adaptive scheduling unit is used to adjust the working sequence, task sequence, resource allocation and running parameters of each module in real time under preset constraints based on simulation and prediction results and optimization algorithms.

8. The automated multi-parameter detection system for aquatic sediments as described in claim 1, characterized in that, The data recording module is used to trace and store equipment data, operation logs, and analysis results during system operation, wherein: The data recording module includes a data standardization unit, a storage management unit, and a sample archive. The data standardization unit is used to receive raw sensor data, operation logs, analysis results and model version information from various modules of the system in real time, and to convert data of different formats into a unified standard format. The storage management unit is used to store raw data and data fingerprints in a chain, establishing a traceability chain for the entire process from front-end sampling, sample pretreatment, spectral detection to result output; The sample archive is used to create an independent digital archive for each sample, fully preserving all spectral data, intermediate processing results, and final analysis reports generated at each testing stage.

Images