Method for tracing heavy metals in water body by converting ultraviolet spectrum gram angle difference field

Abstract

The application discloses a water heavy metal tracing method based on ultraviolet spectrum Gram angle difference field conversion, and belongs to the technical field of analytical chemistry and instrument analysis. The application comprises composite chemical probe screening, pollution source spectrum data acquisition, chemometric conversion driven Gram angle difference field atlas characterization, pollution source database construction, pollution tracing intelligent matching and source analysis decision and heavy metal quantitative tracing platform development. The application realizes rapid identification and accurate tracking of heavy metal pollutants in water bodies by constructing a characteristic correlation model and a standardized database of "spectrum fingerprint-pollution source". The method combines full-band characteristic acquisition of water quality fingerprint spectrum, atlas visual conversion technology and pattern matching algorithm of machine learning, and aims to overcome the limitations of traditional heavy metal tracing technology in spectrum overlap analysis, multi-source pollution differentiation, data fragmentation and tracing efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of analytical chemistry and instrumental analysis, in particular to a water heavy metal tracing method based on ultraviolet spectrum Gram angle difference field transformation. BACKGROUND

[0002] Heavy metal pollution, as a serious challenge to global ecological environment and public health, has various sources, including industrial wastewater discharge, mine leakage, agricultural non-point source pollution, etc. Moreover, actual pollution events are often characterized by multi-component heavy metal pollution. Traditional heavy metal tracing techniques have significant limitations in addressing such challenges: first, they rely on large laboratory instruments, which are costly and have complex operation procedures, making it difficult to meet the emergency response needs of sudden pollution; second, analysis techniques based on a single chemical probe are prone to cross interference due to spectral signal overlap in multi-metal mixed systems, and the ability to distinguish different pollution sources is insufficient; third, there is a lack of standardized management of tracing data, and the spectral characteristics and pollution source attributes are not systematically linked, making it difficult to reuse historical data; fourth, traditional spectral analysis only uses local features, lacks the ability to analyze global features of the full waveband spectrum, and is difficult to capture the differentiated characteristics of heavy metal pollution. SUMMARY

[0003] The purpose of the present application is to provide a water heavy metal tracing method based on ultraviolet spectrum Gram angle difference field transformation. By combining the ultraviolet-visible light spectrum Gram angle difference field transformation technology with machine learning, the difference expression of multi-component heavy metal spectral fingerprint signals is realized by introducing a composite chemical probe and a full-spectrum ultraviolet-visible light Gram angle difference field spectrum technology, the cross interference phenomenon is effectively suppressed, the detection sensitivity under low concentration conditions is improved, the pattern recognition ability of the machine learning model to global features is used to break through the limitations of traditional linear analysis, a mapping framework of "spectrum atlas-feature vector-pollution source attribute" is established, high-dimensional feature correlation in the Gram angle difference field atlas is accurately extracted, synchronous identification and matching of multi-component heavy metal pollution sources are realized, the accuracy and reliability of the tracing are improved, a multi-scale feature extraction strategy is adopted, different scale texture features and cross-wavelength dependence in the Gram angle difference field atlas are extracted by integrating multiple machine learning models, a composite feature vector containing global and local features is constructed, the adaptability of the model to complex pollution scenes is enhanced, the stability and consistency of the tracing results in the full concentration range are ensured, a lightweight end-to-end tracing platform is developed, the machine learning model and the low-cost Gram angle difference field fingerprint spectrometer are integrated, the whole process automation of "spectrum acquisition-feature extraction-database matching-result output" is realized, the equipment cost is reduced and the on-site one-key operation is supported, the portable and intelligent environmental monitoring technology is provided with technical support, through the closed-loop technical path of "multi-probe fingerprinting-multi-scale feature modeling-lightweight system integration", the chemical probe design and machine learning algorithm are deeply integrated, the discrimination and characterization dimension of the spectral signal are improved, the data dependence and analysis bottleneck of the traditional model are broken through, a rapid and high-precision heavy metal quantitative tracing system is formed, real-time and reliable technical support is provided for actual environmental monitoring and pollution emergency disposal, and the problems of multi-component tracing efficiency, spectral analysis accuracy, data management cost and equipment portability in the prior art are solved.

[0004] The present application is realized by the following technical solutions:

[0005] The present application is a water heavy metal tracing method based on ultraviolet spectrum Gram angle difference field transformation, comprising the following steps:

[0006] Composite chemical probe screening: according to the chemical properties of heavy metal ions, a specific color developing agent is selected, and the color developing agent is combined into a composite chemical probe through orthogonal experimental design.

[0007] Pollution source spectrum data acquisition: collect potential pollution source water samples, filter them, and then use an ultraviolet-visible light spectrophotometer to collect absorbance data in the full wavelength range, simultaneously record basic attribute information such as sampling site, time, and pollution source type, and form a standardized spectrum data set.

[0008] The stoichiometric conversion driven Gram angle difference field map characterization unifies the one-dimensional absorbance data through an interpolation algorithm, converts the one-dimensional absorbance data into two-dimensional map data by using a Gram angle difference field algorithm, and generates visual spectrum characterization by using a pseudo-color mapping technology.

[0009] The pollution source database is constructed by using a machine learning model to extract features from the two-dimensional map, fusing local and global features to generate a composite feature vector, integrating the feature vector and heavy metal component data to construct a traceability database containing multi-dimensional attributes, and establishing a dynamic updating mechanism.

[0010] The pollution traceability intelligent matching and source analysis decision is obtained by adding the composite chemical probe to the to-be-tested water sample after pretreatment, then using an ultraviolet-visible spectrophotometer to collect full-waveband absorbance data, and using a Gram angle difference field to convert the full-waveband absorbance data into a map and extract a feature vector, using a cosine similarity algorithm to calculate the matching degree with the database features, and outputting a suspected pollution source and a traceability report.

[0011] The heavy metal quantitative traceability platform is developed to ensure the operability and flexibility of the method. On the basis of the database and the pollution traceability intelligent matching, an interactive prediction platform is developed, which can directly perform pollution traceability by directly inputting the Gram angle difference field map of the pollution sample, and give a traceability report.

[0012] Further, the orthogonal experiment design includes investigating the effects of probe concentration ratio, reaction time and environmental conditions on the complexation reaction, monitoring the absorbance change by using ultraviolet-visible spectroscopy, and comparing and analyzing the selectivity and stability of the probe combination for metal ions.

[0013] Further, in the composite chemical probe screening, antimony, iron, nickel, cadmium and copper are selected as target analytes, and a plurality of composite chemical probes are selected according to their chemical properties. Then, the prepared composite chemical probes are mixed with single metal solutions, and the color development reaction effect is observed. By comparing and analyzing the color difference values of each composite chemical probe, the composite chemical probe 1 with the best color difference value is selected.

[0014] Further, in the collection of potential pollution source water samples, 100 mL of water samples from three pollution sources (S1-S3) in the upstream watershed of a river are collected, filtered through a 0.45 μm filter membrane, transferred to a 96-well plate, and then the composite chemical probe 1 is added to each well. Then, an enzyme-labeled instrument is used for optical characterization, a 230-780 nm wide band detection range (covering the ultraviolet-visible light characteristic absorption region) is set, high-resolution spectrum collection is performed at an interval of 2 nm, and 276 wavelength nodes of absorbance data are obtained by using an absorbance calculation formula during the scanning process of each sample well. A two-dimensional absorbance matrix (sample number x wavelength number) is constructed, and the absorbance calculation formula is as follows:

[0015] 1 (1)

[0016] wherein A(λ) is the absorbance at wavelength λ, is the molar absorption coefficient of the i-th heavy metal, is the concentration of the i-th metal, and l is the optical path length.

[0017] Further, the Gram angle difference field algorithm maps the absorbance sequence into a two-dimensional graph by data normalization and phase transformation operation, retains the time sequence correlation characteristics and spectral distribution rules between wavelengths, and converts the obtained absorbance data into a graph using the Gram angle difference field formula as follows:

[0018] (2)

[0019] wherein is the i-th sample point in the normalized sequence element, and similarly is the j-th sample point in the normalized sequence element.

[0020] Further, the machine learning model includes ResNet50 and InceptionV3 deep learning models, which perform multi-scale feature extraction on the converted GADF graph, capture local texture and global structure features of the graph at three scales of (224, 224), (336, 336), and (448, 448) through a multi-scale input strategy, extract feature vectors using ResNet50 and InceptionV3 at each scale, form a composite feature representation through feature splicing operation, and store the feature tensor in a preset path. The MD5 hash value of the calculation configuration parameter is generated to generate a cache identifier, and a dual verification mechanism of file modification timestamp and hash value is used to automatically trigger feature re-extraction and index update of new samples, thereby achieving dynamic maintenance and efficient retrieval of the database.

[0021] Further, the efficient retrieval realizes fast retrieval of the feature vector through the Annoy index structure, and the multi-scale input strategy formula, feature splicing operation formula, and hash value formula are as follows:

[0022] (3)

[0023] wherein S is a scale set, and is an image size parameter at the i-th scale.

[0024] (4)

[0025] wherein F is a composite feature vector, and are the feature vectors extracted by ResNet50 and InceptionV3 at the i-th scale, respectively.

[0026] (5)

[0027] wherein H is the hash value of the configuration parameter, ModelType is the model type list, ScaleConfig is the scale configuration parameter, and ImageSize is the input image size.

[0028] Further, after collecting the downstream pollution water sample, the previous steps are performed to obtain the GADF spectrum of the pollution water sample, then the model extracts a composite vector containing multi-scale features, the cosine similarity algorithm is used to calculate the matching degree with the database features, the top three suspected pollution sources are selected in descending order of similarity, and a traceability report is generated. In addition, the system optimizes and reuses the batch processing and caching mechanism to reduce the traceability time, and the cosine similarity algorithm calculation formula is as follows:

[0029] (6)

[0030] wherein, is the composite feature vector of the water sample to be traced, is the feature vector of the pollution source in the database, is the vector dot product, , are the lengths of the two vectors, respectively.

[0031] Further, in the development of the heavy metal quantitative traceability platform, the platform interface is a graphical operation platform of the pollution traceability analysis system, and the main functions include data selection, pollution traceability analysis and result visualization display. The user first specifies the database path and the absorbance data file to be analyzed through the file selection area, clicks "start analysis", and the system completes data preprocessing, pollution source analysis and other operations in the background thread. The progress bar and the log area feedback the processing state in real time. After the analysis is completed, the result tab page displays the suspected pollution sources and the similarity of each sample in the form of a table. Clicking on the table row can view the Gram angle difference field image of the corresponding sample in the GADF spectrum tab page, and the pollution characteristics are presented intuitively. The whole process realizes the full-process automatic analysis from data input to result visualization, avoids interface lag, and ensures the intuitive presentation of the analysis results.

[0032] The present application has the following advantages:

[0033] 1. The present application constructs a collaborative technology system of "tailored chemical probe-full waveband spectrum acquisition-spectrum feature transformation-machine learning matching": through the specific complexation reaction of composite probe and heavy metal, a differentiated "spectrum fingerprint" is generated in the full waveband of ultraviolet-visible light, combined with spectrum transformation technology to map one-dimensional absorbance curve to high-dimensional feature image, using machine learning model to extract multi-scale features of the spectrum, realize the analysis of cross-wavelength global dependence, and break through the limitations of traditional linear analysis method.

[0034] 2. The present application first deeply integrates digital water quality fingerprint spectrum technology and machine learning similarity algorithm, establishes a standardized correlation database of "spectrum features-pollution sources", and forms an end-to-end automated traceability process from spectrum acquisition to source identification.

[0035] 3. The present application realizes rapid spectrum acquisition, feature matching and pollution source identification after on-site sampling through lightweight hardware integration and embedded algorithm deployment, promotes the paradigm upgrade of heavy metal traceability technology from laboratory offline analysis to on-site real-time detection, and provides efficient technical support for environmental pollution emergency disposal and source supervision.

[0036] Of course, any product implementing the present application does not necessarily need to achieve all the advantages described above. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 Water body heavy metal traceability method flow chart;

[0038] Figure 2 Multi-component heavy metal dispersion system composite chemical probe screening chart;

[0039] Figure 3 Different pollution source Gram angle difference field spectrum;

[0040] Figure 4 Pollution traceability intelligent matching chart;

[0041] Figure 5 Heavy metal quantitative traceability platform chart. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0043] Please refer to Figures 1-5 , the present application provides a technical solution: a water body heavy metal traceability method based on ultraviolet spectrum Gram angle difference field transformation, comprising the following steps:

[0044] Complex chemical probe screening, according to the chemical properties of heavy metal ions, select the corresponding color developer, through orthogonal experimental design to combine them into a combination of chemical probes, orthogonal experimental design includes the influence of probe concentration ratio, reaction time and environmental conditions on complexation reaction, by monitoring the absorbance change by ultraviolet-visible spectroscopy, comparative analysis of the selectivity and stability of probe combination for metal ions, first select antimony, iron, nickel, cadmium and copper as the target analyte, and according to its chemical properties to select multiple complex chemical probes, then, the prepared complex chemical probes and single metal solution mixed, observe its color reaction effect. By comparing the color difference value of each complex chemical probe, the complex chemical probe 1 with the best color difference value is selected.

[0045] Through the characteristic complexation reaction of complex chemical probes and heavy metals, a differentiated "spectrum fingerprint cluster" is formed in the full wave band of ultraviolet-visible light. Combined with wide-band spectral scanning technology, the problem of signal overlap in traditional methods is effectively solved, and the discrimination degree of spectral characteristics of different pollution sources is greatly improved, providing a high-identification signal base for subsequent traceability matching.

[0046] Pollution source spectral data collection, collect potential pollution source water samples, filter them, then use ultraviolet-visible spectrophotometer to collect absorbance data in the full wave band range, record the sampling site, time, pollution source type and other basic attribute information at the same time, form a standardized spectral data set, collect 100 mL of water samples from 3 pollution sources (S1-S3) in the upstream watershed of the river, filter them through a 0.45 μm filter membrane, transfer them to a 96-well plate, and add complex chemical probe 1 to each well, then use an enzyme marker to perform optical characterization, set the 230-780 nm wide-band detection range (covering the ultraviolet-visible light characteristic absorption region), perform high-resolution spectral collection with 2 nm interval, calculate the absorbance data of 276 wavelength nodes for each sample well in the scanning process using the absorbance calculation formula, construct a two-dimensional absorbance matrix (sample number x wavelength number), the absorbance calculation formula is as follows:

[0047] 1 (1)

[0048] Where A(λ) is the absorbance at wavelength λ, is the molar absorption coefficient of the i th heavy metal, is the concentration of the i th metal, and l is the optical path.

[0049] The stoichiometric conversion driven Gram angle difference field mapping characterizes that one-dimensional absorbance data is uniformly sampled in dimension through an interpolation algorithm, is converted into a two-dimensional map by using a Gram angle difference field algorithm, and visualized spectrum characterization is generated by combining a pseudo-color mapping technology. The Gram angle difference field algorithm maps the absorbance sequence into a two-dimensional map by data normalization and phase transformation operations, retains the time sequence correlation characteristics and spectral distribution rules between wavelengths, and converts the obtained absorbance data into a map by using a Gram angle difference field formula, as follows:

[0050] (2)

[0051] wherein, is the i-th sample point in the normalized sequence element, and similarly, is the j-th sample point in the normalized sequence element.

[0052] By using the Gram angle difference field mapping conversion technology and the multi-model integrated feature extraction framework, automatic analysis of global features such as spectral texture and cross-wavelength correlation is realized, the limitation of traditional technology of only analyzing local features is broken through, and the synchronous recognition accuracy of multi-component heavy metal pollution sources is greatly improved. Different types of pollution sources such as industrial sources and agricultural sources can be accurately distinguished, and the misjudgment rate is significantly reduced.

[0053] Pollution source database construction, feature extraction of two-dimensional maps by machine learning model, fusion of local and global features to generate composite feature vectors, integration of feature vectors and heavy metal component data to construct a traceability database containing multi-dimensional attributes, and establishment of a dynamic updating mechanism. The machine learning model includes ResNet50 and InceptionV3 deep learning models, which perform multi-scale feature extraction on the converted GADF map. Through a multi-scale input strategy, local texture and global structure features of the map are captured at three scales of (224, 224), (336, 336), and (448, 448). The feature vectors of ResNet50 and InceptionV3 are extracted at each scale, respectively. Composite feature representation is formed by feature splicing operation, and feature tensor is stored in a preset path. The MD5 hash value of the calculation configuration parameter is generated to generate a cache identifier. By using the dual verification mechanism of file modification timestamp and hash value, the feature re-extraction and index update of new samples are automatically triggered to realize dynamic maintenance and efficient retrieval of the database. The efficient retrieval realizes fast retrieval of feature vectors by Annoy index structure. The multi-scale input strategy formula, feature splicing operation formula, and hash value formula are as follows:

[0054] (3)

[0055] wherein, S is a scale set, and is an image size parameter at the i-th scale.

[0056] (4)

[0057] where F is the composite feature vector, and are the feature vectors extracted by ResNet50 and InceptionV3 respectively at the i-th scale;

[0058] (5)

[0059] where H is the hash value of the configuration parameter, ModelType is the model type list, ScaleConfig is the scale configuration parameter, and ImageSize is the input image size.

[0060] A structured "spectrum fingerprint-pollution source" standardized database is constructed, integrating spectrum maps, heavy metal components, and multi-dimensional attributes of pollution sources. Machine learning similarity algorithms are used to achieve rapid matching of test samples, significantly improving the reuse rate of historical data. Compared with the traditional technology that requires several days to rebuild the model, the new scheme significantly shortens the model construction time and improves the response efficiency of pollution incidents.

[0061] After the pre-treatment of the test water sample, the composite chemical probe is added, then the ultraviolet-visible spectrophotometer is used to collect the full-band absorbance data, and the Gram angle difference field is converted into a spectrum and the feature vector is extracted. The cosine similarity algorithm is used to calculate the matching degree with the database features, and the suspected pollution source and the source tracing report are output. After collecting the downstream pollution water sample, the previous steps are performed to obtain the GADF spectrum of the pollution water sample. Then, the model extracts the composite vector containing multi-scale features, and the cosine similarity algorithm is used to calculate the matching degree with the database features. The top three suspected pollution sources are selected in descending order of similarity and a source tracing report is generated. In addition, the system optimizes and caches the reuse mechanism through batch processing to reduce the source tracing time. The cosine similarity algorithm calculation formula is as follows:

[0062] (6)

[0063] where is the composite feature vector of the water sample to be traced, is the feature vector of the pollution source in the database, is the vector dot product, , are the lengths of the two vectors respectively.

[0064] A light-weight portable detection system is developed, which integrates a miniature spectrometer and an embedded machine learning engine to realize the full-process automation of "sampling-spectrum conversion-source identification", and the single-source tracing time is greatly shortened compared with traditional technologies. The device is small in size and low in cost, supports one-key operation and wireless data transmission on site, and meets the real-time detection needs of river emergency monitoring, industrial wastewater online monitoring and other scenes.

[0065] By combining multi-model integrated deep feature extraction, the accuracy of distinguishing similar pollution sources is significantly improved, the confidence of the tracing result is high, and accurate basis can be provided for environmental protection law enforcement to solve the misjudgment problem caused by traditional single feature matching.

[0066] The heavy metal quantitative tracing platform is developed, in order to ensure the operability and flexibility of the method, on the basis of database and intelligent matching of pollution tracing, an interactive prediction platform is developed, which can directly input the gram angle difference field spectrum of the pollution sample, directly perform pollution tracing, and give a tracing report, the platform interface is a graphical operation platform of the pollution tracing analysis system, the main functions include data selection, pollution tracing analysis and result visualization display, the user first specifies the database path and the absorbance data file to be analyzed through the file selection area, clicks "start analysis", and the system completes data preprocessing, pollution source analysis and other operations in the background thread, the progress bar and the log area feedback the processing state in real time. After analysis, the result tab page displays the suspected pollution source and the similarity of each sample in the form of a table, and clicking the table row can view the gram angle difference field image of the corresponding sample in the GADF spectrum tab page, which intuitively presents the pollution characteristics. The whole process realizes the full-process automatic analysis from data input to result visualization, avoids interface lag, and ensures the intuitive presentation of the analysis result.

[0067] Based on low-cost spectrometer and light-weight algorithm deployment, the hardware cost of the system is greatly reduced compared with traditional devices, and no professional personnel is needed for operation, which can be widely applied to the on-site tracing of heavy metal pollution of river, groundwater and other environmental media, and promote the paradigm upgrade of detection technology from laboratory offline analysis to on-site real-time monitoring, and provide an efficient solution for environmental pollution emergency disposal and source control.

[0068] The present application solves the bottleneck problems of the prior art in multi-component tracing efficiency, spectral analysis accuracy, data management cost and device portability through the organic combination of ultraviolet-visible spectrum gram angle difference field conversion technology and machine learning. The specific goals include the following aspects:

[0069] In the analysis detection method, by introducing a composite chemical probe and full-spectrum ultraviolet-visible light Gram angular difference field spectrum technology, the differential expression of multi-component heavy metal spectral fingerprint signals is realized, the cross interference phenomenon is effectively suppressed, the detection sensitivity under low concentration conditions is improved, and the traceability limitation caused by signal overlap and harsh reaction conditions of traditional probes is solved.

[0070] At the model construction level, the pattern recognition ability of machine learning models for global features is used to break through the limitations of traditional linear analysis, establish a mapping framework of "spectrum - feature vector - pollution source attribute", accurately extract the high-dimensional feature correlation in the Gram angular difference field spectrum, realize the synchronous identification and matching of multi-component heavy metal pollution sources, and improve the accuracy and reliability of traceability.

[0071] In terms of data and algorithm optimization, a multi-scale feature extraction strategy is adopted, and multiple machine learning models are integrated to cooperatively extract texture features and cross-wavelength dependence of different scales in the Gram angular difference field spectrum, to construct a composite feature vector containing global and local features, enhance the adaptability of the model to complex pollution scenes, and ensure the stability and consistency of the traceability results in the whole concentration range.

[0072] In the application system design, a lightweight end-to-end traceability platform is developed, which integrates machine learning models and low-cost Gram angular difference field fingerprint spectrometers, realizes the full-process automation of "spectrum acquisition - feature extraction - database matching - result output", reduces the cost of equipment and supports one-key operation on site, and provides technical support for the portability and intelligentization of environmental monitoring technology.

[0073] Through the closed-loop technical path of "multi-probe fingerprinting - multi-scale feature modeling - lightweight system integration", the chemical probe design and machine learning algorithm are deeply integrated, which not only improves the discrimination and representation dimension of spectral signals, but also breaks through the data dependence and analysis bottleneck of traditional models, forming a rapid and high-precision heavy metal quantitative traceability system, providing real-time and reliable technical support for actual environmental monitoring and pollution emergency disposal.

[0074] Ultraviolet-visible light spectrum (UV-Vis Spectroscopy): a spectral technology for analyzing the composition and structure of a substance by measuring the absorbance of the substance in the ultraviolet light (200-400 nm) and visible light (400-780 nm) wavelength bands.

[0075] Gram angular difference field (Gram Angular Difference Field, GADF): a method of converting one-dimensional time series data (such as spectral absorbance sequence) into a two-dimensional image through polar coordinate mapping and trigonometric function coding, which preserves the time sequence correlation characteristics between data amplitude and wavelength.

[0076] Composite chemical probe: A probe system composed of multiple chromogenic agents in a specific ratio, capable of specifically complexing with multi-component heavy metal ions, generating differentiated color signals.

[0077] Spectral fingerprint cluster: A set of characteristic absorbance spectra formed by multi-component heavy metals reacting with composite probes across the entire ultraviolet-visible light wavelength range, which can serve as the "optical fingerprint" of pollution sources.

[0078] Broadband spectral scanning technology: A technology for collecting high-resolution (1-2 nm interval) spectral data across the entire ultraviolet-visible light wavelength range (e.g., 230-780 nm) to obtain full-waveband absorbance data.

[0079] Atomic Absorption Spectrometer (AAS): A laboratory instrument for quantitative analysis of element content by measuring the degree of absorption of specific wavelength light by the atomic vapor of the element to be measured.

[0080] Inductively Coupled Plasma Mass Spectrometer (ICP-MS): A high-sensitivity laboratory instrument for analyzing the elemental composition and content by ionizing the sample using inductively coupled plasma.

[0081] Colorimetric analysis technology: A method for quantitative analysis of substance concentration by comparing the color depth or spectral absorbance of colored substance solutions.

[0082] Interpolation algorithm: A mathematical method for inserting new values between known data points to convert discrete data sequences into continuous smooth data (such as linear interpolation, spline interpolation).

[0083] Pseudo-color mapping technology: A technique for mapping one-dimensional data (such as absorbance) to the color distribution of a two-dimensional image, visualizing data characteristics through color differences.

[0084] Cosine similarity algorithm: An algorithm for measuring the similarity of two vectors by calculating the cosine of the angle between them, with a value closer to 1 indicating higher similarity.

[0085] ResNet50: A deep residual neural network containing 50 layers of network structure, which solves the problem of training degradation in deep networks through residual connections, and is used for feature extraction.

[0086] InceptionV3: An efficient convolutional neural network architecture that processes features of different scales in parallel through "inception modules" to improve model representation ability.

[0087] Orthogonal experimental design: A method for efficiently screening the optimal experimental conditions (such as probe concentration, reaction time) by arranging multi-factor experiments using an orthogonal table.

[0088] Polar coordinate mapping: A mapping method that converts one-dimensional data in Cartesian coordinates to polar coordinates (radius, angle) representation to capture the temporal correlation of data.

[0089] Amplitude normalization: Scaling data to a specific range (such as [0, 1] or [-1, 1]) to eliminate dimensional differences and improve model training efficiency.

[0090] Incremental learning mechanism: A learning strategy that automatically updates model parameters when new data is input without retraining the entire data set, used for dynamic database updates.

[0091] Annoy index structure: An approximate nearest neighbor (ANN) index algorithm that constructs a binary tree structure to accelerate high-dimensional vector retrieval for fast database matching.

[0092] Microplate reader: An instrument for detecting the absorbance of samples in a microplate, supporting wide-band spectral acquisition, suitable for micro-sample analysis.

[0093] Embedded machine learning inference engine: A machine learning model inference module integrated in portable hardware, which can realize local spectral feature extraction and traceability matching.

[0094] Lightweight integration scheme: Embedding spectral conversion modules, databases and machine learning engines into portable hardware to realize "sampling-identification" automation and low-power system design.

[0095] Modular design: Split the system into independent functional modules (such as spectral acquisition, feature extraction), integrated through standardized interfaces to improve system flexibility and maintainability.

[0096] 96-well plate: An experimental plate with 96 micro-wells, which can handle multiple samples simultaneously, used for batch spectral acquisition and chemical reactions.

[0097] MD5 hash value: An encryption algorithm that converts data of any length into a 128-bit hash value, used to verify data integrity or generate cache identifiers.

[0098] Feature vector: A multi-dimensional vector composed of data features, used to represent sample properties (such as spectral features, pollution source attributes), supporting machine learning model training and matching.

[0099] The preferred embodiments of the application disclosed above are only to facilitate the elucidation of the application. The preferred embodiments do not describe all the details of the application and limit the application to the specific embodiments described. Obviously, many modifications and variations can be made in light of the teachings above. The description is chosen and described in order to best explain the principles of the application and its practical application to thereby enable others skilled in the art to best utilize the application and get the best results from the application. The application is only limited by the claims and their full scope and equivalents.

Claims

1. A method for tracing the source of heavy metals in water by transforming Gram angle difference field in ultraviolet spectroscopy, characterized in that, Includes the following steps: Composite chemical probe screening involves screening chromogenic agents based on the chemical characteristics of the target heavy metal ions and combining them into composite chemical probes through orthogonal experimental design. Pollution source spectral data acquisition involves collecting water samples from potential pollution sources, filtering them, and then using an ultraviolet-visible spectrophotometer to collect absorbance data across the entire wavelength range. The sampling location, time, and pollution source type attribute information are recorded to form a standardized spectral dataset. The Gram difference field spectral characterization driven by stoichiometry transforms one-dimensional absorbance data into a two-dimensional spectrum by unifying the dimension through an interpolation algorithm, and then uses the Gram difference field algorithm to generate a visual characterization by combining pseudo-color mapping technology. A pollution source database is constructed by using a deep learning model to extract features at multiple scales from the two-dimensional map, generating composite feature vectors, and integrating them with heavy metal component and pollution source attribute data to build a source tracing database, with a dynamic update mechanism. The deep learning model includes ResNet50 and InceptionV3, which capture the local texture and global structural features of the map at three scales (224,224), (336,336), and (448,448) through a multi-scale input strategy. Feature vectors are extracted for each scale using ResNet50 and InceptionV3 respectively, and composite feature representations are formed through feature concatenation. Intelligent matching and source apportionment decision-making for pollution source tracing: After adding the composite chemical probe to the water sample to be tested, absorbance data is collected. After Gram angle difference field conversion and feature extraction, the cosine similarity algorithm is used to match with the database and output suspected pollution sources and source tracing reports. A quantitative traceability platform for heavy metals was developed. Based on the aforementioned database and matching module, an interactive platform was developed that supports direct pollution source analysis by inputting Gram angle difference field maps.

2. The method for tracing the source of heavy metals in water bodies using ultraviolet spectral Gram difference field conversion according to claim 1, characterized in that, The orthogonal experimental design includes optimizing the probe concentration ratio, reaction time and environmental conditions, and monitoring absorbance changes through ultraviolet-visible spectroscopy to evaluate the selectivity and stability of the probe combination for metal ions.

3. The method for tracing the source of heavy metals in water bodies using ultraviolet spectral Gram difference field conversion according to claim 1, characterized in that, In the screening of the composite chemical probes, the target heavy metals include at least one of antimony, iron, nickel, cadmium, and copper; By comparing the color difference values ​​of each probe after reacting with the metal solution, the composite chemical probe with the best color difference value was selected.

4. The method for tracing the source of heavy metals in water through ultraviolet spectral Gram difference field conversion according to claim 3, characterized in that, The water samples collected from potential pollution sources included 100 mL samples from each of the three pollution sources in the upstream basin of the river. After filtration through a 0.45 μm filter membrane, a composite chemical probe was added, and absorbance data were collected at 2 nm intervals within the wavelength range of 230-780 nm using an ELISA reader. A two-dimensional absorbance matrix (number of samples × number of wavelengths) was constructed, and the absorbance was calculated using the following formula: 1 （1） Where i is the index of the component, and A(λ) is the absorbance at wavelength λ. Let be the molar absorptivity of the i-th heavy metal. Let be the concentration of the i-th metal, and l be the optical path length.

5. The method for tracing the source of heavy metals in water bodies using ultraviolet spectral Gram angle difference field conversion according to claim 1, characterized in that, The Gram angle difference field algorithm maps the absorbance sequence into a two-dimensional spectrum through data normalization and phase transformation operations, preserving the temporal correlation characteristics and spectral distribution patterns between wavelengths. The obtained absorbance data is then used to perform spectrum conversion using the Gram angle difference field formula, which is as follows: （2） in, For the i-th sample point in the normalized sequence elements, similarly... Let j be the j-th sample in the normalized sequence elements.

6. The method for tracing the source of heavy metals in water bodies using ultraviolet spectral Gram difference field conversion according to claim 1, characterized in that, The machine learning model includes ResNet50 and InceptionV3 deep learning models. It performs multi-scale feature extraction on the converted GADF map and adopts a multi-scale input strategy to capture the local texture and global structural features of the map at three scales: (224,224), (336,336), and (448,448). For each scale, feature vectors are extracted using ResNet50 and InceptionV3 respectively. Feature concatenation is performed to form a composite feature representation, and the feature tensor is stored in a preset path. A cache identifier is generated by calculating the MD5 hash value of the configuration parameters. Using a dual verification mechanism of file modification timestamp and hash value, the feature re-extraction and index update of new samples are automatically triggered, realizing dynamic maintenance and efficient retrieval of the database.

7. The method for tracing the source of heavy metals in water bodies using ultraviolet spectral Gram difference field conversion according to claim 6, characterized in that, The efficient retrieval achieves rapid feature vector retrieval through the Annoy index structure. The multi-scale input strategy formula, feature concatenation operation formula, and hash value formula are as follows: （3） Where S is the scale set, and represents the image size parameter of the i-th scale; （4） Where F is the composite feature vector, and These are the feature vectors extracted by ResNet50 and InceptionV3 at the i-th scale, respectively; （5） Where H is the hash value of the configuration parameter, ModelType is the list of model types, ScaleConfig is the scale configuration parameter, and ImageSize is the input image size.

8. The method for tracing the source of heavy metals in water bodies using ultraviolet spectral Gram angle difference field conversion according to claim 1, characterized in that, After collecting downstream polluted water samples in the pollution source tracing intelligent matching and source apportionment decision-making process, the GADF map of the polluted water samples is obtained. The composite vector containing multi-scale features is extracted through the model, and the matching degree with the database features is calculated using the cosine similarity algorithm. The top three suspected pollution sources are screened in descending order of similarity and a source tracing report is generated. The system reduces the source tracing time through batch processing optimization and cache reuse mechanism.

9. The method for tracing the source of heavy metals in water bodies using ultraviolet spectral Gram difference field conversion according to claim 8, characterized in that, The cosine similarity algorithm is calculated using the following formula: （6） in, The composite feature vector of the water sample to be traced. For the feature vectors of pollution sources in the database, It is the vector dot product. , These are the magnitudes of the two vectors, respectively.

10. The method for tracing the source of heavy metals in water by ultraviolet spectral Gram difference field conversion according to claim 1, characterized in that, The heavy metal quantitative traceability platform is a graphical interactive platform that allows users to upload absorbance data or Gram angle difference field maps, automatically perform traceability analysis, and output visual reports.

Images