High-precision spectral signal peak detection method and system

Abstract

The invention relates to the technical field of spectral signal processing, in particular to a high-precision spectral signal peak detection method and system. According to the technical scheme, the method comprises the following steps: denoising and smoothing spectral data; dividing a spectrum curve by adopting a dynamic partitioning strategy; generating candidate partitioning points based on various local features, and generating feature blocks through optimization processing; according to the method, the influence of feature differences of different regions on a detection algorithm is reduced through a dynamic partitioning strategy, a multi-scale morphological feature system is constructed to comprehensively describe spectral curve morphological characteristics, potential peaks are gradually mined by applying a layered iteration peak detection algorithm, and local characteristics of spectral signals are adapted by means of an adaptive parameter adjustment mechanism, so that the spectral curve morphological characteristics are accurately detected. The peak value correction and fusion mechanism is utilized to improve the quality and reliability of a detection result, spectral data with a complex structure and a high noise level can be effectively dealt with, an efficient and reliable solution is provided for the field of spectral analysis, and the accuracy and robustness of peak value detection and the conciseness and reliability of the result are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral signal processing, and in particular to a high-precision spectral signal peak detection method and system. Background Art

[0002] Spectral signal peak detection is a key technical means for material composition analysis, structure characterization and concentration determination, and is widely used in environmental monitoring, biomedicine, materials science and other fields. Traditional spectral analysis methods generally have the following technical bottlenecks:

[0003] The spectral data collection process is easily affected by factors such as environmental noise and instrument response fluctuations, resulting in signal distortion. Although wavelet transform technology can separate signal and noise through multi-scale decomposition (signal wavelet coefficients are concentrated at a specific scale, and noise coefficients are dispersed), threshold processing of high-frequency details can easily cause weak peak features to be lost. The Savitzky-Golay (SG) smoothing algorithm suppresses noise through local polynomial fitting, but its parameter selection (window size / polynomial order) is highly sensitive, the edge processing capability is insufficient, and the stability is limited in complex spectral scenes.

[0004] Most existing methods treat spectral curves as a single data body for processing, ignoring the characteristic differences between different bands. For example, the gradient change rate difference between high curvature areas and flat areas can reach an order of magnitude. The unified processing strategy leads to the coexistence of peak miss detection (high dynamic area) and false peak misdetection (low change area), and it is difficult to deal with complex data with asymmetric peaks and shoulder peak structures such as Doppler broadened spectra.

[0005] More than 80% of traditional solutions rely only on intensity or differential features (such as first-order / second-order derivatives), and fail to effectively integrate key information such as frequency domain characteristics and local signal-to-noise ratio. The single feature dimension leads to insufficient resolution of overlapping peaks (peak spacing < 0.5 times the half-peak width), and the recognition accuracy of clustering algorithms (such as DBSCAN) in two-dimensional feature space is significantly reduced.

[0006] Detection algorithm parameters (such as thresholds and clustering radius) are usually set statically. When the noise level fluctuates by more than 20%, the detection accuracy of the fixed parameter solution decreases by more than 35%. Especially in a high dynamic noise environment (local signal-to-noise ratio fluctuation range of 5-50dB), it is difficult for existing technologies to balance sensitivity and specificity.

[0007] Preliminary detection results often have problems with peak position shift, residual pseudo peaks and redundant peaks. Traditional correction methods mostly rely on simple threshold screening, lack physical constraints (peak width / symmetry) and adaptive merging mechanisms, resulting in the final output results being unable to accurately reflect the true optical properties of the material.

[0008] In summary, there is an urgent need to develop a high-precision spectral peak detection scheme that integrates local feature analysis, multi-dimensional information fusion, parameter dynamic optimization, and intelligent result correction to break through the technical bottleneck of spectral data analysis with complex structures and high noise levels. Summary of the Invention

[0009] The object of the present invention is to address the problem in the background art that existing methods mostly treat spectra as a whole, ignoring the feature differences in different regions, resulting in limited peak detection accuracy and efficiency and difficulty in dealing with spectral data with complex structures and high noise levels. A high-precision spectral signal peak detection method and system are proposed.

[0010] On the one hand, the present application provides a high-precision spectral signal peak detection method, including the following steps:

[0011] S1. Perform denoising and smoothing preprocessing on the spectral data;

[0012] S2. Divide the spectral curve using a dynamic block strategy: generate candidate block points based on multiple local features and generate feature blocks through optimization processing;

[0013] S3. Construct a multi-scale morphological feature system: fuse derivative features, morphological gradient features, local extreme features, frequency domain features, and signal-to-noise ratio features, and perform standardization processing;

[0014] S4. Perform hierarchical iterative peak detection:

[0015] Locate candidate peaks through an adaptive threshold method;

[0016] Use a density clustering algorithm to analyze the feature matrix and identify potential peak regions;

[0017] Deduct the residual signal based on the model and perform iterative detection until the termination condition is met;

[0018] S5. Dynamically adjust the detection parameters according to the local characteristics of the spectrum;

[0019] S6. Perform correction and fusion processing on the detected peaks, and sequentially perform local search window correction, physical constraint condition screening, peak merging algorithm, and secondary merging processing. The peak correction and fusion mechanism of the present invention effectively improves the peak detection accuracy and result simplicity through local search window correction, physical constraint condition screening, peak merging algorithm, and secondary merging processing, ensuring that the detection results accurately reflect the true characteristics of the spectral curve.

[0020] Optionally, the step S2 includes:

[0021] Extract the gradient, Laplace response, and curvature of the spectral curve as local features;

[0022] Dynamically generate block thresholds based on feature distribution;

[0023] Blocks containing multiple peaks are subdivided according to the peak positions, and super-long blocks are divided according to size constraints.

[0024] The dynamic block strategy generation method includes:

[0025] The first-order derivative, second-order differential and curvature characteristics of the spectral curve are combined. Among them, the first-order derivative, namely the gradient, can reflect the change in the slope of the spectral curve at different points, providing key information for determining the upward and downward trends of the curve; the second-order differential is calculated with the help of the Laplace operator to capture the local acceleration change of the spectral curve and assist in identifying the change in the concave and convex characteristics of the curve; the curvature feature directly quantifies the degree of curvature of the curve.

[0026] Dynamic threshold optimization: Generates an adaptive threshold based on the 95% percentile of feature distribution to avoid over / under detection caused by fixed thresholds.

[0027] Based on the selected candidate block points, the initial block area is generated and optimized. For blocks containing multiple local peaks, they are further subdivided according to the specific locations of the local peaks, so that each sub-block covers a single peak feature more specifically; for blocks that are too long, they are evenly divided according to the preset maximum block size to ensure the rationality and processability of the blocks.

[0028] Optionally, in step S3:

[0029] The derivative features include first-order derivatives and second-order derivatives;

[0030] The standardization process adopts a robust normalization method based on statistical distribution.

[0031] It should be noted that the construction of multi-scale morphological features aims to fully mine the useful information of the spectral curve and improve the accuracy and reliability of peak detection. Its specific features are as follows:

[0032] Basic differential features: Calculate the first-order derivative and second-order derivative of the spectral curve to obtain basic differential features. The first-order derivative reflects the slope change of the spectral curve at different wavelengths, providing a quantitative basis for determining the rising or falling interval of the curve. The second-order derivative is used to capture the curvature change of the spectral curve, which helps to identify the local morphological transition point of the curve.

[0033] Morphological gradient verification: The morphological gradient is calculated by convolution operation between the structural element and the spectral curve. This feature can highlight the edge and mutation information in the spectral curve, enhance the contrast between the peak area and the surrounding area, and make the potential peak more prominent in the curve.

[0034] Local maximum value verification: By quickly identifying these local maximum value points, the search range can be effectively reduced, enabling subsequent analysis to more precisely focus on potential peak regions, thereby improving detection efficiency and accelerating the entire peak recognition process.

[0035] Frequency domain characteristics: The spectral signal is transformed from the time domain to the frequency domain through the fast Fourier transform, which reflects the periodic changes and frequency components of the spectral signal, and helps to identify whether there are periodically occurring peak patterns in the spectrum.

[0036] Local signal-to-noise ratio characteristics: Quantify the contrast between the signal and background noise at each point on the spectral curve, thereby effectively distinguishing real signal peaks from pseudo-peaks caused by noise.

[0037] Optionally, in the density clustering analysis of step S4:

[0038] The clustering radius parameter is dynamically calculated according to the distance distribution in the feature space;

[0039] The minimum sample number is adaptively adjusted based on the local signal-to-noise ratio and data scale;

[0040] In addition, in step S4, the hierarchical iterative detection algorithm is specifically used for peak detection of spectral curves, and its specific content includes:

[0041] Adaptive threshold method for rapid positioning: Analyze the statistical characteristics of the spectral signal in the local area, such as mean, variance, etc., and dynamically calculate a suitable threshold based on these characteristics. Subsequently, the spectral curve is preliminarily scanned through this threshold to quickly identify candidate peak positions that are significantly higher than the background noise.

[0042] Feature point analysis based on density clustering: The DBSCAN algorithm clusters according to the local density of feature points, and can effectively identify the clustering clusters corresponding to potential peak regions in the spectral curve. For each clustering cluster, the algorithm further calculates its morphological characteristics, such as the size, shape, density, etc. of the cluster. By analyzing these characteristics, true and reliable peaks are screened and confirmed.

[0043] Gaussian model subtraction and iterative processing: Subtract the constructed Gaussian model from the original spectral signal to obtain the residual signal. Then, the algorithm reapplies the detection process to the residual signal, that is, performs the adaptive threshold method and density clustering analysis again.

[0044] Optionally, the dynamic adjustment of detection parameters in step S5 includes:

[0045] Quantify the local signal-to-noise ratio to characterize the contrast relationship between the signal and background noise;

[0046] Dynamically adjust the sensitivity threshold of the detection algorithm according to the local signal-to-noise ratio.

[0047] Optionally, the local search window correction in step S6 includes:

[0048] An adaptive window is established with the initially detected peak position as the center, and the window size is dynamically adjusted according to the spectral resolution;

[0049] Perform a secondary extreme value search within the window to correct the peak position deviation.

[0050] Optionally, the adjacent peak merging in step S6 includes:

[0051] Merging is triggered when the peak spacing and peak height difference meet the preset conditions;

[0052] Optimize the peak parameters after merging using a weight factor.

[0053] On the other hand, the present application provides a high-precision spectral signal peak detection system, including:

[0054] A preprocessing module that performs spectral denoising and smoothing processing;

[0055] A dynamic block division module that adaptively divides the spectral curve based on local features;

[0056] A feature construction module that generates a multi-scale fusion standardized feature matrix;

[0057] A hierarchical detection module that integrates an adaptive threshold detection, density clustering, and residual iteration unit;

[0058] A parameter optimization module that dynamically configures detection parameters according to the local characteristics of the spectrum;

[0059] A peak optimization module that performs peak position correction, pseudo-peak elimination, and adjacent peak merging.

[0060] Optionally, the dynamic block division module is configured to:

[0061] Generate candidate block points through three-character edge detection;

[0062] Subdivide the multi-peak block according to the peak position, and evenly divide the ultra-long block according to the preset maximum size.

[0063] Optionally, the peak optimization module includes:

[0064] A physical constraint screening unit that eliminates pseudo-peaks based on the peak width threshold, peak height ratio, and peak shape symmetry;

[0065] A peak merging unit that merges adjacent similar peaks using a weight factor optimization algorithm.

[0066] Compared with the prior art, the present application includes at least one of the following beneficial technical effects:

[0067] Adaptive block division based on local characteristics of spectral signals (gradient, Laplace response, curvature) reduces the impact of feature differences in different regions on the detection algorithm and improves processing pertinence;

[0068] Integrate multi-dimensional features such as basic differential, morphological gradient, frequency domain, local signal-to-noise ratio, etc. to comprehensively describe the spectral curve morphology and enhance the integrity and noise resistance of feature expression;

[0069] Combining the adaptive threshold method, DBSCAN density clustering and Gaussian model subtraction iteration, the potential peaks are gradually mined, which effectively copes with complex structure spectral data and improves detection accuracy and robustness.

[0070] Dynamically optimize the sensitivity of algorithm parameters based on the local signal-to-noise ratio, so that the detection process can adapt to the local characteristics of the spectral signal, reduce the cost of manual parameter adjustment, and improve the generalization ability of the algorithm;

[0071] Through local search window correction, physical constraint screening and peak merging algorithm, peak position deviation is corrected, false peaks are removed, redundancy is eliminated, and the accuracy and reliability of detection results are improved.

[0072] The present invention reduces the influence of feature differences in different regions on the detection algorithm through a dynamic blocking strategy, constructs a multi-scale morphological feature system to comprehensively characterize the morphological characteristics of the spectral curve, uses a hierarchical iterative peak detection algorithm to gradually explore potential peaks, uses an adaptive parameter adjustment mechanism to adapt to the local characteristics of the spectral signal, and uses a peak correction and fusion mechanism to improve the quality and reliability of the detection results. It can effectively deal with spectral data with complex structures and high noise levels, provide an efficient and reliable solution for the field of spectral analysis, and improve the accuracy and robustness of peak detection and the simplicity and reliability of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 is a spectrum signal processing flow chart of the present invention;

[0074] Figure 2 It is the dynamic block result and feature extraction effect diagram of the present invention;

[0075] Figure 3 It is a multi-scale feature fusion feature matrix diagram of the present invention;

[0076] Figure 4 This is the final peak detection result diagram of the present invention. DETAILED DESCRIPTION

[0077] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0078] Example 1

[0079] A high-precision spectral signal peak detection method proposed by the present invention. The core architecture of the present invention includes five functional modules, as shown in Figure 1 System configuration diagram:

[0080] Signal preprocessing module: Perform data standardization, noise suppression, and signal enhancement;

[0081] Adaptive block engine: Dynamically divide the spectral interval based on multi-feature fusion;

[0082] Hierarchical detection unit: Integrate iterative threshold detection and density clustering analysis;

[0083] Result fusion mechanism: Realize the optimal integration of multi-level detection results;

[0084] Post-processing verification: Provide peak position correction and visual interaction interface.

[0085] I. Spectral data preprocessing

[0086] 1. Data loading and baseline correction

[0087] Parse the SIF standard format spectral file and automatically identify the orbit number and frame sequence information;

[0088] Perform differential signal processing: When data is detected, use the background subtraction algorithm to eliminate environmental noise;

[0089] Baseline drift correction: Intercept the first 50 data points to calculate the mean of the background noise. When the maximum signal deviation exceeds 1000 counts, perform global baseline zeroing.

[0090] 2. Wavelet threshold denoising

[0091] Select the Sym5 wavelet basis for three-level decomposition and retain the low-frequency approximation component;

[0092] High-frequency noise processing: Based on the improved Donoho threshold model, calculate the shrinkage threshold of the detail coefficients: , where 1.4826 is a constant correction coefficient, which is a calibration factor that correlates the median absolute deviation with the standard deviation of the normal distribution; MAD( ) is the median absolute deviation of the third-level detail coefficients, represents the detail coefficients of the third layer after three-level wavelet decomposition. MAD is a robust statistic, aiming to estimate the dispersion of noise in the third-level detail coefficients; N is the number of data points, referring to the total sample of the original data to be denoised;

[0093] Perform soft threshold shrinkage on the high-frequency components exceeding the threshold, and the signal-to-noise ratio after reconstruction is increased by more than 40%.

[0094] 3. Interpolation and smoothing processing

[0095] Dynamic selection of interpolation strategy: Cubic spline interpolation is enabled when the wavelength step size is less than 0.2 nm, otherwise linear interpolation is used;

[0096] Apply the Savitzky-Golay filter with a fixed window length of 25 data points and third-order polynomial fitting to effectively retain the peak shape characteristics.

[0097] II. Dynamic block processing

[0098] 1. Three-feature edge detection

[0099] As Figure 2 , extract three groups of feature vectors:

[0100] Gradient intensity: Calculate the absolute value of the first derivative of the signal;

[0101] Laplacian response: Identify the abrupt edges through the second-order differential operator;

[0102] Curvature feature: Quantify the degree of curvature of the spectral profile

[0103] Dynamic threshold setting: Extract the top 5% quantile of each feature as the edge determination threshold.

[0104] 2. Candidate point detection and screening

[0105] Detect the feature significant points and screen the candidate block points that meet the minimum spacing constraint. By comparing the feature values with the dynamic threshold, the key feature points in the spectral curve are identified. These feature points usually correspond to the local maxima, mutation points, or positions with significant curvature changes in the spectral curve. To ensure the rationality of the block, these candidate block points are screened to remove those that are too close. The screened candidate block points can effectively divide the spectral regions with different features, providing a basis for subsequent block generation.

[0106] 3. Block generation

[0107] Generate the initial blocks and adjust them according to the block length constraint. Using the screened candidate block points as boundaries, the spectral curve is divided into multiple initial block regions. During the block process, considering that there may be multiple overlapping peaks or complex structures in the spectral curve, the initial blocks are further optimized. For the blocks containing multiple local peaks, they are subdivided according to the positions of the local peaks to ensure that the spectral features within each block region are relatively concentrated; while for the overly long blocks, they are evenly divided according to the preset maximum block size to avoid a single block region being too large and resulting in reduced processing efficiency. The final block result can effectively divide the spectral curve into multiple feature blocks that are easy to process, laying a foundation for subsequent feature extraction and peak detection.

[0108] III. Multi-scale feature extraction

[0109] As Figure 3 , construct a multi-scale feature matrix.

[0110] 1. Basic feature calculation

[0111] Locate the spectral mutation region through the first derivative, capture the inflection point of curvature through the second derivative, and enhance the peak-valley contrast through the morphological gradient. The three respectively capture the slope, curvature, and mutation features, construct a multi-dimensional local change representation, and provide underlying data support for feature fusion.

[0112] 2. Frequency domain feature extraction

[0113] Perform a fast Fourier transform (FFT) on the spectral signal to extract its frequency domain features. Obtain the spectral frequency domain features through FFT, and analyze the periodic peak pattern and noise frequency. Identify the peaks associated with the material properties based on the frequency energy distribution, and provide frequency dimension information support for spectral analysis.

[0114] 3. Feature matrix construction and standardization

[0115] Integrate multi-dimensional features such as spectral derivative, morphological gradient, and signal-to-noise ratio to construct a point-to-point spectral feature matrix. Eliminate the influence of outliers through median standardization, and achieve data scale normalization based on the feature median and absolute deviation to form highly consistent input data to support subsequent clustering analysis.

[0116] IV. Hierarchical iterative peak detection

[0117] 1. Primary iterative detection layer

[0118] Set the initial peak threshold to 400 counts and perform 3 rounds of iterative detection with a decay coefficient of 0.6;

[0119] Gaussian residual subtraction model: , where is the remaining signal intensity after subtracting the Gaussian model, 0.8 is the weight coefficient, is the original signal intensity; is the abscissa of the independent variable corresponding to the current calculated residual, representing the wavelength of the spectrum; is the peak position of the i-th Gaussian peak, that is, the abscissa value corresponding to the maximum signal value of this peak; is the peak height of the i-th Gaussian peak, that is, the maximum signal intensity value of this peak at the center position ; is the original peak width of the i-th detected peak; is the exponential term of the Gaussian function, describing the attenuation law of the peak signal with the distance of the abscissa from the peak center;

[0120] Termination condition: The residual signal intensity is lower than 15% of the maximum value of the original signal.

[0121] 2. Secondary density clustering layer

[0122] Construct a 12-dimensional feature space, including normalized wavelength / intensity, morphological gradient, Fourier harmonic components, and local signal-to-noise ratio;

[0123] Dynamic DBSCAN parameters: ; where is the maximum neighborhood distance in the DBSCAN algorithm, representing the neighborhood range of samples; represents the distance from a single sample to its k-th nearest neighbor sample; 5.8 and 2.8 are scaling coefficients, corresponding to the core area and non-core area respectively. A larger is required in the core area to make it easier for samples in the high-density area to be connected to each other, while the in the non-core area is reduced to reduce the over-inclusion of low-density / noise areas and avoid misjudging noise as clustering members.

[0124] The minimum number of samples is adaptively adjusted according to the local signal-to-noise ratio, and the formula is: where is the minimum number of samples threshold of the DBSCAN algorithm, 0.015 is the proportionality coefficient, controlling the influence intensity of the total number of samples N and the signal-to-noise ratio SNR on min_samples; N is the total number of samples in the local area. The more samples there are, the more neighboring samples can be determined as core points; SNR is the local signal-to-noise ratio, measuring the ratio of the effective feature signal to noise in the local area; is the signal-to-noise ratio adjustment factor. The stronger the signal, the higher the SNR, and the more dense the effective samples in the area. ensures that the minimum number of samples is not less than 3 to ensure the basic reliability of clustering.

[0125] 3. Peak feature extraction

[0126] Locate the spectral maximum point in the clustering domain as the peak position, calculate the corresponding intensity value to determine the peak height, the half-peak width span represents the peak width, and the skewness is quantified based on the symmetry of the peak area morphology. The four-dimensional parameters (position / height / width / skewness) completely describe the peak shape characteristics. Among them, the skewness index can effectively identify asymmetric peaks caused by instrument response or sample characteristics, providing a criterion for the correlation analysis of material characteristics.

[0127] V. Peak correction and post-processing

[0128] 1. Local search window correction

[0129] An adaptive search window is established based on the initial inspection peak position, and the secondary extreme value positioning is performed. The window size is dynamically adjusted according to the spectral resolution: narrow window with high resolution for precise calibration (matching the characteristics of narrow peaks), wide window with low resolution for full coverage (adapting to the morphology of wide peaks), effectively eliminating the peak position offset error caused by noise interference.

[0130] 2. Screening of physical constraint conditions

[0131] Physical constraint conditions are established based on the peak width threshold, peak height ratio, and peak shape symmetry, eliminating too narrow / too wide peaks (noise pseudo-peaks) and abnormally asymmetric peaks (detection errors). The screening threshold supports dynamic parameter adjustment to adapt to different detection requirements, ensuring that the spectral characteristics conform to the laws of the optical properties of substances.

[0132] 3. Peak merging

[0133] Based on the double judgment criteria of peak distance and peak height difference, merging is triggered when the distance between adjacent peaks is less than the dynamic threshold and the height difference ≤ 20%. The merging parameters are optimized by using a weight factor (peak height weight 0.7, position weight 0.3). The dynamic threshold is adaptively calibrated according to the spectral resolution in steps of 0.01λ, eliminating redundant peaks while maintaining the integrity of the true multi-peak structure.

[0134] 4. Secondary merging process

[0135] After the above peak correction and screening steps, the detection results are merged again to further eliminate possible overlapping or similar peaks. By comprehensively considering the characteristic information such as the position, height, and width of the peaks, the detection results are comprehensively optimized and integrated. This step can ensure that the final peak list is concise and accurate, and can truly reflect the actual peak situation in the spectral curve. The peak list after the secondary merging process will be used as the final detection result, providing reliable data support for subsequent spectral analysis and data interpretation.

[0136] In this embodiment, the improved Donoho threshold model and Sym5 wavelet basis are adopted, the signal-to-noise ratio is improved, high-frequency noise is effectively suppressed, and at the same time, the low-frequency trend and detail features of the spectral signal are retained. Through the fixed window (25 points) of the Savitzky-Golay filter and the third-order polynomial fitting, the peak shape features are accurately retained while denoising, avoiding the damage to the peak shape by traditional smoothing methods. By dynamically subtracting the background noise and global baseline zeroing, environmental interference is eliminated; according to the wavelength step, the interpolation algorithm (cubic spline / linear) is adaptively selected to ensure the continuity and accuracy of the spectral data.

[0137] Among them, through multi-feature edge detection and dynamic thresholding, combining gradient intensity, Laplacian response, and curvature features to identify spectral mutation points and bending degrees, the threshold is dynamically set based on the 95th percentile to avoid over-detection or missed detection caused by a fixed threshold, and accurately locate the block boundaries. The blocks containing multiple peaks are subdivided according to the peak positions, and the overly long blocks are evenly divided to ensure that each sub-block contains only a single peak or a feature interval of reasonable length, reducing the interference of cross-regional feature mixing on the detection algorithm and improving the subsequent processing efficiency.

[0138] In addition, 12-dimensional features such as the first / second-order derivatives (slope and curvature changes), morphological gradients (edge enhancement), frequency-domain features (periodicity analysis), and local signal-to-noise ratios (distinguishing signals from noise) are integrated to comprehensively characterize the spatial morphology, frequency components, and noise distribution of the spectral curve, and improve the recognition of peak features.

[0139] Median normalization (based on the feature median and absolute deviation) is used to eliminate the influence of outliers, ensuring the stability and consistency of the feature matrix, providing a reliable input for subsequent analyses such as density clustering, gradually mining potential peaks, and adapting to complex spectral structures. By combining the adaptive threshold method with the Gaussian residual subtraction model, the strong peak signals are gradually stripped through multiple rounds of iteration to expose the hidden weak peaks, effectively dealing with the peak detection problems under overlapping peaks and noise backgrounds.

[0140] It should be noted that based on the DBSCAN algorithm, the 12-dimensional feature matrix is clustered, and the parameters (ε and min_samples) are dynamically adjusted to adapt to the local signal-to-noise ratio, identifying potential peak regions of any shape, and screening out real peaks through morphological features (peak width, skewness, etc.), suppressing noise interference and false peaks. Through the secondary detection of the residual signal, the weak peaks masked by strong peaks are mined, improving the detection ability for complex spectra (such as multi-component superimposed signals) and avoiding missed detection.

[0141] In this embodiment, the search window is dynamically adjusted based on the spectral resolution to re-locate the true maximum point near the initially detected peak position, correcting the peak position offset caused by noise or feature extraction errors, especially suitable for spectral data with mixed high and low resolutions. Pseudo-peaks that do not conform to the optical characteristics are removed through physical constraints such as peak width, peak height ratio, and symmetry; redundant peaks are dynamically merged based on the peak distance and peak height difference, eliminating repeated detections while retaining the true multi-peak structure, making the results conform to the laws of material characteristics, and improving the reliability and readability of the results.

[0142] The present invention forms a complete closed-loop through signal preprocessing, adaptive block division, hierarchical detection, result fusion, and post-processing verification. Each module can be independently optimized and adapted to different spectral data types (such as Doppler spectra). Through dynamic block division, parameter adjustment, and threshold optimization, it reduces the dependence on manual parameter tuning, improves the generalization ability of the algorithm for spectra with high noise and complex structures, and is applicable to automated spectral analysis scenarios. It systematically solves problems such as sensitivity to regional differences, poor noise robustness, and missed detection of complex peak patterns in traditional methods, providing an efficient and reliable engineering solution for high-precision spectral peak detection.

[0143] Embodiment 2

[0144] This embodiment provides a high-precision spectral signal peak detection system, including:

[0145] A preprocessing module that performs spectral denoising and smoothing processing;

[0146] A dynamic block division module that realizes adaptive block division of the spectral curve based on local features, specifically including:

[0147] Generating candidate block division points through three-feature edge detection;

[0148] Subdividing the multi-peak block according to the peak position and uniformly dividing the ultra-long block according to the preset maximum size

[0149] A feature construction module that generates a standardized feature matrix with multi-scale fusion;

[0150] A hierarchical detection module that integrates an adaptive threshold detection, density clustering, and residual iteration unit;

[0151] A parameter optimization module that dynamically configures detection parameters according to the local characteristics of the spectrum;

[0152] A peak optimization module that performs peak position correction, false peak elimination, and adjacent peak merging, specifically including:

[0153] A physical constraint screening unit that eliminates false peaks based on peak width threshold, peak height ratio, and peak shape symmetry;

[0154] A peak merging unit that merges adjacent similar peaks using a weight factor optimization algorithm.

[0155] The system of this embodiment improves the quality of spectral data through the preprocessing module, suppresses noise and retains the peak shape; the dynamic block division module reduces the interference of regional feature differences and improves the processing pertinence based on three-feature edge detection and adaptive block division strategy, such as Figure 2 ; the feature construction module generates a standardized feature matrix with multi-scale fusion, comprehensively depicts the spectral morphology, and enhances the noise resistance, such as Figure 3; The hierarchical detection module integrates multiple detection methods, gradually explores potential peaks, and effectively copes with complex spectral structures; the parameter optimization module dynamically configures parameters, reduces manual parameter adjustment, and improves generalization ability; the peak optimization module corrects peak positions, removes pseudo peaks, and eliminates redundancy through physical constraint screening and weight factor merging algorithms, ultimately achieving efficient and high-precision peak detection of complex high-noise spectral data. The results are accurate, reliable, and concise, such as Figure 4 The results show the accuracy of the peak detection results.

[0156] The above specific embodiments are only several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A high-precision spectral signal peak detection method, characterized in that It includes the following steps: S1. Perform denoising and smoothing preprocessing on spectral data; S2. Divide the spectral curve using a dynamic block strategy: generate candidate block points based on multiple local features and generate feature blocks through optimization processing; S3. Construct a multi-scale morphological feature system: fuse derivative features, morphological gradient features, local extreme value features, frequency domain features, and signal-to-noise ratio features, and perform standardization processing; S4. Perform hierarchical iterative peak detection: Locate candidate peaks by the adaptive threshold method; Analyze the feature matrix using the density clustering algorithm and identify potential peak regions; Deduct the residual signal based on the model and perform iterative detection until the termination condition is met; S5. Dynamically adjust the detection parameters according to the local characteristics of the spectrum; S6. Perform correction and fusion processing on the detected peaks, and sequentially perform local search window correction, physical constraint condition screening, peak merging algorithm, and secondary merging processing.

2. The high-precision spectral signal peak detection method according to claim 1, wherein The step S2 includes: Extract the gradient, Laplace response, and curvature of the spectral curve as local features; Dynamically generate block thresholds based on the feature distribution; Subdivide the multi-peak blocks according to the peak positions, and divide the ultra-long blocks according to the size constraints.

3. A high-precision spectral signal peak detection method according to claim 1, characterized in that In the step S3: The derivative features include the first derivative and the second derivative; The standardization processing uses a robust normalization method based on the statistical distribution.

4. A high-precision spectral signal peak detection method according to claim 1, characterized in that In the density clustering analysis of the step S4: The clustering radius parameter is dynamically calculated according to the distance distribution in the feature space; The minimum number of samples is adaptively adjusted based on the local signal-to-noise ratio and data scale.

5. A high-precision spectral signal peak detection method according to claim 1, characterized in that The dynamic adjustment of the detection parameters in the step S5 includes: Quantify the local signal-to-noise ratio to characterize the contrast relationship between the signal and background noise; Dynamically adjust the sensitivity threshold of the detection algorithm according to the local signal-to-noise ratio.

6. A high-precision spectral signal peak detection method according to claim 1, characterized in that The local search window correction in the step S6 includes: Establish an adaptive window centered on the initially detected peak position, and the window size is dynamically adjusted according to the spectral resolution; Perform a secondary extreme value search within the window to correct the peak position deviation.

7. A method for detecting the peak value of a high-precision spectral signal according to claim 1, characterized in that The adjacent peak merging in the step S6 includes: Trigger merging when the peak spacing and peak height difference meet the preset conditions; Optimize the peak parameters after merging using a weight factor.

8. A high-precision spectral signal peak detection system, characterized in that, It includes: A preprocessing module that performs spectral denoising and smoothing; A dynamic block module that adaptively divides the spectral curve based on local features; A feature construction module that generates a multi-scale fused standardized feature matrix; A hierarchical detection module that integrates an adaptive threshold detection, density clustering, and residual iteration unit; A parameter optimization module that dynamically configures the detection parameters according to the local characteristics of the spectrum; A peak optimization module that performs peak position correction, false peak elimination, and adjacent peak merging.

9. The high-precision spectral signal peak detection system according to claim 8, characterized in that The dynamic block module is configured to: Generate candidate block points through three-characteristic edge detection; Subdivide the multi-peak blocks according to the peak positions, and evenly divide the ultra-long blocks according to the preset maximum size.

10. A high-precision spectral signal peak detection system according to claim 8, characterized in that, The peak optimization module includes: A physical constraint screening unit that eliminates false peaks based on the peak width threshold, peak height ratio, and peak shape symmetry; A peak merging unit that optimizes the algorithm to merge adjacent similar peaks using a weight factor.

Images