Water quality anomaly identification method based on multi-modal spectrum fusion
By using multimodal spectral fusion technology, ultraviolet absorption, fluorescence emission, and Raman scattering spectra are acquired and processed simultaneously, solving the problems of insufficient accuracy and dynamic response capability in the detection of water quality anomalies in existing technologies, and realizing high-precision monitoring and early warning of complex water quality.
Patent Information
- Application Number
- CN202610736618.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-26
- Publication Date
- 2026-08-25
AI Technical Summary
Existing water quality anomaly detection technologies rely on single spectral modes or simple feature splicing, which makes it difficult to effectively characterize the intrinsic correlation of spectral responses under different physical mechanisms. This results in limited ability to identify complex water quality anomalies and a lack of dynamic response capability, making them susceptible to time asynchrony, baseline drift, and noise interference.
A multimodal spectral fusion method is adopted. By simultaneously acquiring ultraviolet absorption spectrum, fluorescence emission spectrum and Raman scattering spectrum, band alignment, noise suppression and baseline drift correction are performed to construct a standardized spectral matrix, extract abnormal response features, and generate water quality fusion feature vector through cross-modal coupling. Combined with the change rate in the time dimension, the abnormal disturbance index is calculated to realize the graded anomaly judgment.
It improves the accuracy and robustness of water quality anomaly identification, enhances the sensitivity to multi-source pollution and transient disturbances, improves the adaptability to multi-modal data, and realizes high-precision monitoring and early warning of complex water environments.
Smart Images
Figure CN122634438A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of water quality testing, and specifically to a method for identifying water quality anomalies based on multimodal spectral fusion. Background Technology
[0002] In the field of water quality monitoring, multimodal spectroscopy techniques such as ultraviolet absorption spectroscopy, fluorescence emission spectroscopy, and Raman scattering spectroscopy have been widely used for water composition analysis and pollution identification.
[0003] However, existing water quality anomaly detection technologies typically rely on single spectral modes or simple feature splicing for analysis, making it difficult to effectively characterize the intrinsic correlations between spectral responses under different physical mechanisms. This results in limited ability to identify complex water quality anomalies (such as multi-source pollution superposition and transient disturbances). Furthermore, different spectroscopic devices vary in sampling time, band range, and signal-to-noise characteristics, easily leading to problems such as time asynchrony, baseline drift, and noise interference, thus affecting the stability and consistency of feature extraction. On the other hand, traditional analytical methods often focus on static feature analysis, lacking the ability to dynamically model spectral changes over time, making it difficult to reflect the evolutionary trends of water quality anomalies.
[0004] Therefore, to solve the above problems, a water quality anomaly identification method based on multimodal spectral fusion is needed, which can improve the accuracy and dynamic response capability of multimodal water quality anomaly identification, and enhance the anti-interference and robustness of water quality identification. Summary of the Invention
[0005] In view of this, the purpose of this invention is to overcome the defects in the prior art and provide a water quality anomaly identification method based on multimodal spectral fusion, which can improve the accuracy and dynamic response capability of multimodal water quality anomaly identification, and enhance the anti-interference and robustness of water quality identification.
[0006] The water quality anomaly identification method based on multimodal spectral fusion of the present invention includes:
[0007] Collect multimodal spectral data of the water body to be tested, including ultraviolet absorption spectrum, fluorescence emission spectrum and Raman scattering spectrum;
[0008] Preprocessing of spectral data from different modes generates a standardized spectral matrix;
[0009] Extract anomalous response features from the normalized spectral matrix;
[0010] The abnormal response features in different modes are coupled to generate a water quality fusion feature vector;
[0011] Analyze the rate of change of the water quality fusion characteristic vector over time, and calculate the abnormal disturbance index of the water body to be tested;
[0012] Based on the abnormal disturbance index and the abnormal judgment threshold, the water quality abnormality identification result is output.
[0013] Furthermore, multimodal spectral data of the water body to be tested are collected, specifically including:
[0014] The ultraviolet spectroscopy acquisition module, fluorescence spectroscopy acquisition module, and Raman spectroscopy acquisition module are controlled to work synchronously based on a unified clock source.
[0015] The time offset of different mode acquisition times is compensated, and a unified spectral sequence is constructed based on the compensated timestamps.
[0016] Furthermore, the preprocessing includes band alignment, noise suppression, and baseline drift correction.
[0017] Baseline drift correction is performed according to the following formula:
[0018]
[0019] in, The corrected spectral intensity; The original spectral intensity; For spectral baseline function; Spectral wavelength; For the baseline curvature term; This is the band adaptive adjustment coefficient.
[0020] Furthermore, anomalous response features are extracted from the normalized spectral matrix, specifically including:
[0021] Extracting spectral lines from a normalized spectral matrix:
[0022]
[0023] in, For the first The normalized spectral line vector of the mode; For the first The mode in the first Standardized spectral intensity values at each wavelength sampling point; This represents the total number of wavelength sampling points.
[0024] Calculate the spectral gradient:
[0025]
[0026] in, For the first The mode in the first Spectral gradient at each wavelength position; Wavelength interval;
[0027] The spectral lines are enhanced according to the following formula:
[0028]
[0029] in, For the first The mode in the first Anomalous enhanced response values at each wavelength position; For the first The mode in the first Local spectral energy in the region near each wavelength; , The radius of the local window;
[0030] Anomaly enhancement response values exceeding the anomaly enhancement threshold are taken as target anomaly values, and the positions of the anomaly peaks and the half-width of the peaks corresponding to the target anomaly values are taken as anomaly response features.
[0031] Furthermore, the anomalous response features in different modalities are coupled to generate a water quality fusion feature vector, specifically including:
[0032] The abnormal response features in different modes are coupled according to the following formula:
[0033]
[0034] in, For the first Modality and the first Spectral coupling coefficients between modes; , The first Modality and the first Location of anomalous spectral peaks in the modality; , They are respectively , The corresponding peak width at half maximum (FWHM); It is a spectral width suppression factor;
[0035] Based on spectral line coupling coefficient and the corresponding abnormal peak intensity , Calculate the coupling response strength between different modes. :
[0036]
[0037] The coupling response intensities between different modes are combined to form a water quality fusion feature vector:
[0038]
[0039] in, for ; This represents the number of modes.
[0040] Furthermore, the anomalous disturbance index of the water body to be tested is calculated according to the following formula:
[0041]
[0042] in, This refers to the abnormal disturbance index; The characteristic rate of change; For characteristic change acceleration; It is an acceleration enhancement factor.
[0043] Furthermore, the anomaly detection threshold is determined according to the following formula:
[0044]
[0045] in, This is the threshold for anomaly detection; Basic threshold; This refers to the abnormal disturbance index; It represents the spectral residual energy; , All of these are disturbance adjustment parameters.
[0046] Furthermore, the output of water quality anomaly identification results includes:
[0047] If the abnormal disturbance index Greater than the anomaly detection threshold If so, the water quality will be abnormal:
[0048] like The water quality shows a slight abnormality.
[0049] like The water quality is then considered moderately abnormal.
[0050] like The water quality is severely abnormal.
[0051] in, as well as This is the grade boundary coefficient.
[0052] The beneficial effects of this invention are as follows: This invention discloses a water quality anomaly identification method based on multimodal spectral fusion. It constructs multi-source spectral data by simultaneously acquiring ultraviolet absorption, fluorescence emission, and Raman scattering spectra. Further, it performs band alignment, noise suppression, and baseline drift correction on the multimodal data to form a standardized spectral matrix. Based on this, it extracts anomaly response features such as spectral line gradients and local energy, and constructs a fusion feature vector through a cross-modal coupling mechanism. It calculates the anomaly disturbance index by combining the rate of change and acceleration in the time dimension, and achieves graded anomaly determination based on an adaptive threshold. This invention can effectively improve the accuracy and robustness of anomaly identification in complex aquatic environments, enhance sensitivity to multi-source pollution and transient disturbances, and improve adaptability to differences in multimodal data and noise interference, showing promising engineering application prospects. Attached Figure Description
[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments:
[0054] Figure 1 This is a schematic diagram of the water quality anomaly identification method of the present invention. Detailed Implementation
[0055] The present invention will be further described below with reference to the accompanying drawings, as shown in the figures:
[0056] This embodiment discloses a water quality anomaly identification method based on multimodal spectral fusion, including the following steps:
[0057] S1. Collect multimodal spectral data of the water body to be tested, including ultraviolet absorption spectrum, fluorescence emission spectrum and Raman scattering spectrum;
[0058] S2. Perform band alignment, noise suppression, and baseline drift correction on spectral data of different modes to generate a standardized spectral matrix;
[0059] S3. Extract anomalous response features from the standardized spectral matrix;
[0060] S4. Couple the abnormal response features in different modes to generate a water quality fusion feature vector;
[0061] S5. Analyze the rate of change of the water quality fusion characteristic vector over time, and calculate the abnormal disturbance index of the water body to be tested;
[0062] S6. Output the water quality anomaly identification results based on the anomaly disturbance index and the anomaly judgment threshold.
[0063] This invention provides a method for synchronous acquisition, unified correction, enhanced extraction of abnormal features, and cross-modal coupling analysis of multimodal spectral data, enabling more sensitive and robust identification of water quality anomalies, thereby improving monitoring accuracy and early warning capabilities in complex aquatic environments.
[0064] In this embodiment, in step S1, the multimodal spectral data represents the data types obtained for the same water body using different physical mechanisms, different response modes, or different sensing dimensions. Ultraviolet absorption spectroscopy, fluorescence emission spectroscopy, and Raman scattering spectroscopy represent three modal data types.
[0065] Collect multimodal spectral data of the water body to be tested, specifically including:
[0066] First, a unified clock source is constructed. This unified clock source can be a GPS timing module or a PTP high-precision network timing module, which is used to provide synchronous clock signals to the ultraviolet spectroscopy acquisition module, the fluorescence spectroscopy acquisition module, and the Raman spectroscopy acquisition module. This enables each acquisition module to trigger sampling operations under the same time reference, thereby reducing cross-device sampling timing deviations.
[0067] During the independent data acquisition process of each spectral acquisition module, slight time offsets may exist between different modes due to hardware response delays and differences in transmission links. Therefore, it is necessary to further obtain the original acquisition timestamps corresponding to each mode and align them based on a unified clock source. Specifically, time interpolation compensation or linear regression fitting can be used to correct the offset of the sampling times of ultraviolet, fluorescence, and Raman spectral data, thereby obtaining equivalent sampling times under a unified time reference.
[0068] After time offset compensation, the modal spectral data are sorted and reconstructed according to the corrected timestamps to form a unified spectral sequence. This unified spectral sequence is used to characterize the correspondence between spectral responses of different physical mechanisms at the same time point, thus providing a consistent data foundation for subsequent multimodal spectral fusion and anomaly feature extraction.
[0069] In this embodiment, step S2 includes preprocessing such as band alignment, noise suppression, and baseline drift correction. Band alignment is used to address inconsistencies in wavelength range, sampling interval, and resolution among different spectral modes. First, a unified reference band interval is determined based on the effective band ranges of ultraviolet absorption, fluorescence emission, and Raman scattering spectra. Then, an interpolation resampling method is used to perform scale unification processing on the spectral data of each mode. The interpolation method can be linear interpolation, cubic spline interpolation, or piecewise polynomial interpolation, so that different mode spectra are expressed in the same wavelength coordinate system with equal dimensions.
[0070] The noise suppression is used to reduce high-frequency disturbances caused by instrument electronic noise, ambient light interference, and random scattering fluctuations during spectral acquisition. Specifically, moving average filtering, median filtering, or Savitzky-Golay smoothing filtering can be used to perform preliminary denoising on the original spectrum to suppress random noise.
[0071] Baseline drift correction is performed according to the following formula:
[0072]
[0073] in, The corrected spectral intensity; The original spectral intensity; The spectral baseline function can be a polynomial fitting function or an Asymmetric Least Squares (ALS) baseline. Spectral wavelength; For the baseline curvature term; This is the band adaptive adjustment coefficient, which can be taken as 0.01 to 0.5, preferably 0.05 to 0.2, in order to balance the correction intensity and signal fidelity.
[0074] The standardized spectral matrix is essentially a unified response distribution map of water under various detection mechanisms. For three modes, the maximum spectral intensity, the minimum spectral intensity, and the original spectral intensity of the current mode are used for standardization to obtain the standardized spectral intensity (within the range of 0-1). The standardized spectral intensity corresponding to each mode is set to a row to form three rows of data values, thus obtaining a standardized spectral matrix with three rows and several columns.
[0075] In this embodiment, step S3, extracting the anomalous response features from the normalized spectral matrix, specifically includes:
[0076] Extracting spectral lines from a normalized spectral matrix:
[0077]
[0078] in, For the first The normalized spectral line vectors of the modes, This represents the spectral line vector corresponding to the ultraviolet absorption mode. This represents the spectral line vector corresponding to fluorescence emission. is the spectral line vector corresponding to the Raman scattering mode; For the first The mode in the first Standardized spectral intensity values at each wavelength sampling point; This represents the total number of wavelength sampling points.
[0079] Calculate the spectral gradient:
[0080]
[0081] in, For the first The mode in the first The spectral gradient at each wavelength position is used to characterize the rate of spectral change between adjacent wavelengths; The wavelength interval is set according to the spectral resolution: UV absorption: 0.1 nm to 1 nm; fluorescence spectrum: 0.5 nm to 2 nm; Raman spectrum: 1 cm⁻¹ to 5 cm⁻¹ equivalent interval.
[0082] The spectral lines are enhanced according to the following formula:
[0083]
[0084] in, For the first The mode in the first The abnormal enhancement response values at each wavelength position are used to comprehensively characterize the spectral line abruptness characteristics and local spectral energy accumulation characteristics. For the first The mode in the first The local spectral energy in the region near each wavelength is used to characterize the degree of continuous enhancement of local spectral peaks; , The local window radius, The value ranges from 3 to 10 sampling points.
[0085] Anomaly enhancement response values exceeding the anomaly enhancement threshold are taken as target anomaly values, and the positions of the anomaly peaks and the half-width of the peaks corresponding to the target anomaly values are taken as anomaly response features.
[0086] In this embodiment, step S4 involves coupling the abnormal response features in different modalities to generate a water quality fusion feature vector, specifically including:
[0087] The abnormal response features in different modes are coupled according to the following formula:
[0088]
[0089] in, For the first Modality and the first The spectral coupling coefficient between modes is used to characterize the degree of correlation between anomalous spectral lines in different modes; , The first Modality and the first Location of anomalous spectral peaks in the modality; , They are respectively , The corresponding peak width at half maximum (FWHM); This is a spectral width suppression factor used to adjust the influence of peak width differences, with a value range of 0.1 to 2.0, preferably 0.3 to 1.2;
[0090] Based on spectral line coupling coefficient and the corresponding abnormal peak intensity , Calculate the coupling response strength between different modes. Used to characterize cross-modal consistency enhancement response:
[0091]
[0092] The coupling response intensities between different modes are combined to form a water quality fusion feature vector:
[0093]
[0094] in, for ; The modality count corresponds to the modality type mentioned above and has a value of 3.
[0095] In this embodiment, in step S5, the abnormal disturbance index of the water body to be tested is calculated according to the following formula:
[0096]
[0097] in, This is an anomalous perturbation index, used to characterize the temporal evolution intensity of fusion features; The characteristic rate of change; For characteristic change acceleration; This is an acceleration enhancement factor used to strengthen the sensitivity to dynamic changes. Its value ranges from 0.1 to 1.0, with a preferred value of 0.2 to 0.6.
[0098] In this embodiment, in step S6, the anomaly detection threshold is determined according to the following formula:
[0099]
[0100] in, This is the threshold for anomaly detection; The basic threshold can be obtained by statistical analysis of historical stable water body data, and is generally taken as 0.1 to 1.0 (normalized) or the baseline mean value under the corresponding instrument unit. This refers to the abnormal disturbance index; is the spectral residual energy, used to characterize the degree of reconstruction error or anomalous deviation. It is usually normalized and its value ranges from 0 to 1 or from the original energy scale from 0 to 100. , These are all disturbance adjustment parameters, specifically, This is the trend sensitivity coefficient, and its value can range from 0.2 to 1.5. This is the residual sensitivity coefficient, and its value can range from 0.1 to 1.0.
[0101] The output includes water quality anomaly identification results, specifically:
[0102] If the abnormal disturbance index Greater than the anomaly detection threshold If so, the water quality will be abnormal:
[0103] like If so, the water quality shows a slight abnormality (possibly due to organic pollution).
[0104] like The water quality is moderately abnormal (there may be a persistent source of pollution).
[0105] like If this occurs, the water quality is severely abnormal (there may be a sudden pollution event).
[0106] in, as well as This is the grade boundary coefficient. The value can range from 1.2 to 1.7. The value can range from 1.8 to 3.0.
[0107] The above-mentioned classification method can achieve tiered early warning for different levels of pollution, improving the interpretability of decisions and the accuracy of responses in practical applications.
[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for identifying water quality anomalies based on multimodal spectral fusion, characterized in that: include: Collect multimodal spectral data of the water body to be tested, including ultraviolet absorption spectrum, fluorescence emission spectrum and Raman scattering spectrum; Preprocessing of spectral data from different modes generates a standardized spectral matrix; Extract anomalous response features from the normalized spectral matrix; The abnormal response features in different modes are coupled to generate a water quality fusion feature vector; Analyze the rate of change of the water quality fusion characteristic vector over time, and calculate the abnormal disturbance index of the water body to be tested; Based on the abnormal disturbance index and the abnormal judgment threshold, the water quality abnormality identification result is output.
2. The water quality anomaly identification method based on multimodal spectral fusion according to claim 1, characterized in that: Collect multimodal spectral data of the water body to be tested, specifically including: The ultraviolet spectroscopy acquisition module, fluorescence spectroscopy acquisition module, and Raman spectroscopy acquisition module are controlled to work synchronously based on a unified clock source. The time offset of different mode acquisition times is compensated, and a unified spectral sequence is constructed based on the compensated timestamps.
3. The water quality anomaly identification method based on multimodal spectral fusion according to claim 1, characterized in that: The preprocessing includes band alignment, noise suppression, and baseline drift correction. Baseline drift correction is performed according to the following formula: in, The corrected spectral intensity; The original spectral intensity; For spectral baseline function; Spectral wavelength; For the baseline curvature term; This is the band adaptive adjustment coefficient.
4. The water quality anomaly identification method based on multimodal spectral fusion according to claim 1, characterized in that: Extracting anomalous response features from the normalized spectral matrix, specifically including: Extracting spectral lines from a normalized spectral matrix: in, For the first The normalized spectral line vector of the mode; For the first The mode in the first Standardized spectral intensity values at each wavelength sampling point; This represents the total number of wavelength sampling points. Calculate the spectral gradient: in, For the first The mode in the first Spectral gradient at each wavelength position; Wavelength interval; The spectral lines are enhanced according to the following formula: in, For the first The mode in the first Anomalous enhanced response values at each wavelength position; For the first The mode in the first Local spectral energy in the region near each wavelength; , The radius of the local window; Anomaly enhancement response values exceeding the anomaly enhancement threshold are taken as target anomaly values, and the positions of the anomaly peaks and the half-width of the peaks corresponding to the target anomaly values are taken as anomaly response features.
5. The water quality anomaly identification method based on multimodal spectral fusion according to claim 1, characterized in that: Couple the anomalous response features in different modalities to generate a water quality fusion feature vector, specifically including: The abnormal response features in different modes are coupled according to the following formula: in, For the first Modality and the first Spectral coupling coefficients between modes; , The first Modality and the first Location of anomalous spectral peaks in the modality; , They are respectively , The corresponding peak width at half maximum (FWHM); It is a spectral width suppression factor; Based on spectral line coupling coefficient and the corresponding abnormal peak intensity , Calculate the coupling response strength between different modes. : The coupling response intensities between different modes are combined to form a water quality fusion feature vector: in, for ; This represents the number of modes.
6. The water quality anomaly identification method based on multimodal spectral fusion according to claim 5, characterized in that: The abnormal disturbance index of the water body to be tested is calculated according to the following formula: in, This refers to the abnormal disturbance index; The characteristic rate of change; For characteristic change acceleration; It is an acceleration enhancement factor.
7. The water quality anomaly identification method based on multimodal spectral fusion according to claim 1, characterized in that: The anomaly detection threshold is determined according to the following formula: in, This is the threshold for anomaly detection; Basic threshold; This refers to the abnormal disturbance index; It represents the spectral residual energy; , All of these are disturbance adjustment parameters.
8. The water quality anomaly identification method based on multimodal spectral fusion according to claim 7, characterized in that: The output includes water quality anomaly identification results, specifically: If the abnormal disturbance index Greater than the anomaly detection threshold If so, the water quality will be abnormal: like The water quality shows a slight abnormality. like The water quality is then considered moderately abnormal. like The water quality is then severely abnormal. in, as well as This is the grade boundary coefficient.