A Microwave Rainfall Measurement Environmental Noise Identification Method Based on Feature Fingerprint Database Matching

Abstract

The present application relates to meteorological detection and environmental monitoring technical field, specifically to microwave rain measurement environmental noise recognition method based on feature fingerprint library matching, including obtaining aliasing signal containing rainfall echo and environmental noise and extracting radar detection metadata, then by introducing non-negative and sparse constraint, and taking system impulse response as prior information, applying deconvolution processing to separate aliasing signal into independent signal components with high precision; Parallel processing signal scaling, echo gating and space-time alignment, etc. to the separated signal components, construct rain measurement link view, and extract features from it to generate rain measurement binding fingerprints bound with metadata; By matching the fingerprint with the pre-constructed rain measurement environmental noise fingerprint library based on microwave propagation path perception and combining the rainfall physical mechanism constraint to weigh the multi-source evidence, the noise type is identified, and the pure rain measurement signal is obtained. The present application realizes the recognition of microwave rain measurement environmental noise through deconvolution processing.

Description

Technical Field

[0001] This invention relates to the field of meteorological detection and environmental monitoring technology, specifically to a microwave rainfall measurement environmental noise identification method based on feature fingerprint database matching. Background Technology

[0002] As urban refined disaster prevention and mitigation, sponge city operation and maintenance, flash flood warning and traffic management place increasing demands on the spatiotemporal resolution and reliability of rainfall monitoring, microwave rainfall measurement networks (including meteorological microwave radar, commercial microwave communication links and passive microwave) are being rapidly deployed and put into online operation.

[0003] However, complex surface and electromagnetic environments introduce multi-source non-target echoes such as ground clutter, side lobes and multipath, snow melt bright bands, wind turbines and biological micro-Doppler, solar interference, and equipment gain drift. Existing clutter suppression methods based on thresholds and empirical rules lack cross-site and cross-seasonal generalization, focus on attenuation rather than identification and naming, and lack auditable and reusable knowledge accumulation. At the same time, limited edge computing power and real-time processing time limits make it difficult for static parameters and single criteria to adapt to azimuth differences and environmental drift.

[0004] To address this, a microwave rain measurement environmental noise identification method based on feature fingerprint database matching is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying environmental noise in microwave rain measurement based on feature fingerprint database matching. By extracting feature fingerprints and searching in the fingerprint database, and combining physical constraint evidence trade-offs and prior multi-criteria fusion, a final noise label is generated to achieve the identification of rain measurement environmental noise types.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A microwave rainfall measurement environmental noise identification method based on feature fingerprint database matching includes:

[0008] The system acquires aliased microwave rain measurement signals containing rain echoes and environmental noise, and extracts radar detection metadata from them; it then applies aliased signal deconvolution processing to separate the aliased microwave rain measurement signals into signal components; the deconvolution processing introduces non-negativity and sparsity constraints, and uses the system impulse response as a priori to perform joint optimization in the time and frequency domains.

[0009] The signal components are processed in parallel to form a rain measurement link view. The parallel processing includes signal calibration, echo gating, and spatiotemporal alignment. Rain measurement echo feature groups are extracted based on the rain measurement link view to generate a rain measurement binding fingerprint that is bound to the radar detection metadata.

[0010] A rain measurement environmental noise fingerprint database is constructed, echo fingerprint matching based on microwave propagation path perception is performed, collaborative matching results are generated, and multi-source evidence is weighed and judged in combination with rainfall physical mechanism constraints to determine the noise type. Signal components identified as noise are removed to obtain pure rain measurement signals.

[0011] The process of acquiring the aliased microwave rain measurement signal containing rainfall echoes and environmental noise and extracting radar detection metadata from it is as follows:

[0012] Determine the observation sampling configuration, and set the center frequency, pulse repetition frequency, pulse width, and receiver gain;

[0013] Establish a unified spatiotemporal reference, acquire timestamps, site coordinates, antenna azimuth and elevation angles and bind them to the sampling task; perform aliased microwave rain measurement signal acquisition and front-end preprocessing, after down-conversion and bandpass filtering of the aliased microwave rain measurement signal, sample and quantize in-phase and quadrature components; analyze equipment and observation control records, extract transmit power, noise temperature, range gate index and scanning mode, and encapsulate them into radar detection metadata according to field specifications.

[0014] The process of applying aliased signal deconvolution processing to separate the aliased microwave rain measurement signal into signal components is as follows:

[0015] The acquired aliased microwave rain measurement signal is processed using a deconvolution algorithm, and N signal components are separated using multi-resolution analysis technology. The N signal components include at least one candidate rainfall signal component and at least one candidate noise signal component. The deconvolution algorithm analyzes the time and frequency domain characteristics of the signal, filters and enhances each signal component, and identifies the heterospectral characteristics of the rainfall echo and the environmental noise signal.

[0016] The deconvolution introduces non-negativity and sparsity constraints, and uses the system impulse response as a priori to achieve joint optimization in the time and frequency domains.

[0017] Preliminary signal classification and identification are performed on the deconvolutioned signal components to distinguish between rainfall echo components and noise signal components, and these components are marked as candidate signal components.

[0018] The process of constructing the rain measurement link view through parallel processing is as follows:

[0019] Read the distance gate index, site coordinates, antenna azimuth and elevation angles, determine the link crossing section and set echo gating to shield ground clutter;

[0020] A unified timeline is established using timestamps, and the calibrated echo signals are resampled and the echo delay is corrected. The range gate is projected onto the geographic coordinate system, a link unit index is established, and terrain occlusion is corrected. The calibrated echo signals are organized according to link units and time steps, and the echo amplitude, phase, and Doppler velocity are written. Radar detection metadata index is added to construct a rain measurement link view.

[0021] The process of extracting rainfall echo feature groups based on the rainfall link view and generating a rainfall binding fingerprint bound to the radar detection metadata is as follows:

[0022] Read the calibration echo signal according to the link unit and time step and locate the range gate index; extract the rain measurement echo feature group composed of echo amplitude, phase and Doppler velocity; align the feature group with the timestamp, station coordinates, antenna azimuth and elevation angles and scanning mode in the radar detection metadata; complete the binding with the range gate index, link unit index and time step as keys, and encapsulate it into a rain measurement binding fingerprint in a fixed field order.

[0023] The specific process of constructing a rainfall-based environmental noise fingerprint database is as follows:

[0024] The rainfall measurement environmental noise fingerprint database uses rainfall measurement bound fingerprints as the basic unit, defining fingerprint fields that include quantization encoding of rainfall echo feature groups, radar detection metadata index, link unit index, and geographic projection information; establishing a noise type encoding table and fingerprint dictionary, and generating fixed-length fingerprint vectors according to unified rules; constructing a primary key based on the link unit index, distance gate index, time step, and radar detection metadata index, establishing a path adjacency index and a time window index, and writing them into storage; setting the primary key to remove duplicates and retaining version snapshots to form the rainfall measurement environmental noise fingerprint database.

[0025] The specific process for generating collaborative matching results based on microwave propagation path awareness echo fingerprint matching is as follows:

[0026] The echo fingerprint matching based on microwave propagation path perception is pre-screened according to the link unit index, time step call path adjacency index, and time window index; the consistency of center frequency, pulse repetition frequency, scanning mode, antenna azimuth angle, and elevation angle is checked; the rain measurement bound fingerprint and the fingerprint in the rain measurement environmental noise fingerprint database are sequence aligned on the link unit sequence, and the matching score is calculated according to echo amplitude, phase, Doppler velocity, link unit index, and geographic projection information; candidates are truncated according to threshold and sorting, and the collaborative matching result is output.

[0027] The process of weighing multi-source evidence to constrain the physical mechanisms of rainfall is as follows:

[0028] Construct an evidence vector, including consistency between echo amplitude and path attenuation, consistency between phase evolution and propagation delay, consistency between Doppler velocity and scanning geometry, and time series coherence; normalize and weight the consistency scores of each piece of evidence under the constraint of the physical mechanism of rainfall to obtain a weighted score; output preliminary noise labels based on the weighted score threshold and judgment rules.

[0029] The process of multi-criteria fusion verification of preliminary noise labels and rainfall monitoring prior constraints is as follows:

[0030] A priori constraints for rainfall monitoring are established by combining historical rainfall data and meteorological observation records, and a priori criterion table is formed. Multiple criterion sets are set, including temporal continuity, consistency of spatial adjacent links, consistency of echo amplitude and rainfall intensity relationship, and verification of meteorological observation records. The preliminary noise labels are compared with the priori criterion table item by item and a pass mark is generated. The samples that fail are weighted and voted according to preset weights and rejection rules. The samples that fail are reviewed within the path adjacency index and time window index range, and the final noise label is output.

[0031] By removing signal components identified as noise, a pure rain measurement signal is obtained.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] 1. This invention introduces a deconvolution processing method for aliased signals with physical priors, using the system impulse response as prior information and combining non-negativity and sparsity constraints for joint optimization in the time-frequency domain, achieving high-precision separation of rainfall echoes and environmental noise signals. This method not only effectively distinguishes signals with different spectral characteristics but also handles complex cases of spectral overlap through multi-resolution analysis, fundamentally solving the problem of traditional filtering methods' difficulty in removing aliasing noise. This significantly improves the purity of the original rainfall signal, laying a high-quality data foundation for subsequent accurate feature extraction and recognition.

[0034] 2. This invention constructs a structured environmental noise fingerprint database for rainfall measurement and employs a sequence alignment and matching method based on microwave propagation path perception to achieve efficient and accurate identification of environmental noise types. This method is not limited to feature comparison of individual data points, but rather performs dynamic alignment and matching calculations on the fingerprints of the entire link unit sequence across geographic space and time dimensions. This enables more accurate capture of the spatiotemporal evolution characteristics of noise along the propagation path, effectively distinguishing between rainfall signals and noise signals with similar local features but different overall behavior patterns, significantly improving the accuracy and anti-interference capability of identification.

[0035] 3. This invention establishes a dual evidence balance and verification system that combines physical mechanism constraints with prior monitoring verification, ensuring the high reliability and robustness of noise discrimination results. The first layer utilizes the consistency of intrinsic physical quantities such as echo amplitude, phase, and Doppler velocity for strong constraints, eliminating discrimination results that violate the physical laws of rainfall. The second layer introduces historical rainfall data and external meteorological observation records for prior comparison and fusion verification, providing external evidence support for the discrimination. This combined internal and external, progressive verification mechanism greatly reduces the probability of misjudgment and missed judgment, ensuring that the final noise label output has extremely high confidence. Attached Figure Description

[0036] Figure 1 This is a flowchart of a microwave rain measurement environmental noise identification method based on feature fingerprint database matching.

[0037] Figure 2 This is a diagram illustrating the parallel processing structure of a microwave rain measurement environmental noise identification method based on feature fingerprint database matching, which performs parallel processing on aliased microwave rain measurement signals.

[0038] Figure 3 This is a flowchart illustrating the process of obtaining rainfall signals based on a rainfall environmental noise fingerprint database for a microwave rainfall measurement environmental noise identification method based on feature fingerprint database matching. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] This invention proposes a method for identifying ambient noise in microwave rainfall measurement based on feature fingerprint database matching. By extracting feature fingerprints and searching a fingerprint database, combined with physical constraint evidence trade-offs and prior multi-criteria fusion, a final noise label is generated, enabling the identification of ambient noise types in rainfall measurement. The effectiveness of this invention will be illustrated below with two embodiments.

[0041] Example 1:

[0042] In this embodiment, the method proposed in this invention is used to identify environmental noise during rainfall, with reference to... Figure 1 , Figure 2 and Figure 3 ,include:

[0043] The system acquires aliased microwave rain measurement signals containing rain echoes and environmental noise, and extracts radar detection metadata from them; it then applies aliased signal deconvolution processing to separate the aliased microwave rain measurement signals into signal components; the deconvolution processing introduces non-negativity and sparsity constraints, and uses the system impulse response as a priori to achieve joint optimization in the time and frequency domains.

[0044] The signal components are processed in parallel to form a rain measurement link view. The parallel processing includes signal calibration, echo gating and spatiotemporal alignment. Rain measurement echo feature groups are extracted based on the rain measurement link view to generate a rain measurement binding fingerprint that is bound to the radar detection metadata.

[0045] A rain measurement environmental noise fingerprint database is constructed. Echo fingerprint matching based on microwave propagation path perception is performed to generate collaborative matching results. Multi-source evidence is weighed and judged in combination with rainfall physical mechanism constraints to determine the noise type. Signal components identified as noise are removed to obtain pure rain measurement signals.

[0046] Furthermore, the process of acquiring the aliased microwave rain measurement signal containing rain echoes and environmental noise and extracting radar detection metadata from it is as follows: determine the observation sampling configuration, and set the center frequency, pulse repetition frequency, pulse width and receiving gain;

[0047] Establish a unified spatiotemporal reference, acquire timestamps, station coordinates, antenna azimuth and elevation angles and bind them to sampling tasks; perform aliased microwave rain measurement signal acquisition and front-end preprocessing, and sample and quantize in-phase and quadrature components after down-conversion and bandpass filtering of the aliased microwave rain measurement signals; analyze equipment and observation control records, extract transmit power, noise temperature, range gate index and scanning mode, and encapsulate them into radar detection metadata according to field specifications;

[0048] Specifically, during the observation and sampling configuration phase, the center frequency of the C-band (5.6-5.65GHz) or X-band (9.3-9.5GHz) is selected according to the detection requirements, and the pulse repetition frequency of 250-2000Hz and the pulse width of 0.5-2.0 microseconds are set. The receiving dynamic range is optimized through automatic gain control.

[0049] When establishing a unified spatiotemporal reference, the Global Positioning System is used to obtain centimeter-level station coordinates, atomic clocks provide nanosecond-level time synchronization, magnetic compasses and gyroscopes determine the antenna azimuth angle, tilt sensors monitor the elevation angle, and each is bound to a scanning task.

[0050] In the signal acquisition stage, the microwave echo is amplified with low noise and converted into an intermediate frequency signal by orthogonal mixing. After bandpass filtering, the I / Q components are synchronously sampled by a 14-bit ADC, with a sampling rate of 2-4 times the signal bandwidth.

[0051] During metadata parsing, key parameters such as transmit power, system noise temperature, range gate index, and scanning mode are extracted and encapsulated into standardized radar detection metadata in BUFR or HDF5 format.

[0052] Dynamic frequency selection and precise sampling configuration significantly improve detection accuracy. The use of GPS / BeiDou dual-mode positioning and atomic clock synchronization achieves centimeter-level spatial accuracy and nanosecond-level time accuracy. 14-bit high-precision ADC sampling and I / Q quadrature demodulation technology effectively improve the signal-to-noise ratio by 3dB-5dB. Encapsulation in strict accordance with international standard formats improves data compatibility. Automatic gain control and adaptive parameter adjustment reduce operation and maintenance costs by about 30%. The optimized signal processing flow controls the data output delay to within 5 minutes, meeting the real-time requirements of weather forecasting and early warning services.

[0053] The process of applying aliased signal deconvolution processing to separate the aliased microwave rain measurement signal into signal components is as follows:

[0054] The acquired aliased microwave rain measurement signals are processed using a deconvolution algorithm and transformed into a two-dimensional time-spectrum graph through methods such as short-time Fourier transform or wavelet transform. This graph uses time as one axis and frequency as the other, with each point representing the signal energy intensity at that time and frequency. On the time-spectrum graph, the rain echo signal and noise signal exhibit completely different morphological structures; the rain echo typically appears as a blocky structure with concentrated energy, a certain duration, and a specific frequency range. Environmental noise may appear as a low-energy, uniformly distributed background or as pulse interference with an extremely short duration and a very wide frequency range (represented by vertical lines on the graph). The different spectral characteristics of the rain echo and environmental noise signals are identified. The N signal components include at least one candidate rain signal component and at least one candidate noise signal component, and each signal component is filtered and enhanced.

[0055] The deconvolution introduces non-negativity and sparsity constraints, and uses the system impulse response as a priori to achieve joint optimization in the time and frequency domains. The pre-determined system impulse response is substituted into the algorithm, and the distortion represented by the model is reversed and removed when the signal is separated. During the iterative calculation process, all calculation results representing signal energy are forced to be greater than or equal to zero, and any possible solutions that produce negative values ​​are directly discarded. The sparse constraint-guided algorithm tends to generate a solution with energy concentrated in a few regions, which is consistent with the physical fact that rainfall signals are usually localized in time and space.

[0056] Specifically, to achieve the deconvolution processing, this invention employs the alternating direction multiplier method to find a set of separated signals. After this set of signals undergoes a mathematical transformation based on the known system impulse response (i.e., the inherent influence of the radar system itself on the signal), the result can approximate the actual aliased signal acquired to the greatest extent. The alternating direction multiplier method always follows two basic physical constraints: non-negativity constraint and sparsity constraint. The non-negativity constraint requires that the energy value of all separated signal components cannot be negative at any time, which conforms to the physical reality of signal energy. The sparsity constraint introduces a sparsity regularization term into the algorithm so that the real rain echo and most noise are concentrated in a specific region in time and frequency. The sparsity regularization term is optimized and selected based on historical data and experiments.

[0057] During the separation process, multi-resolution analysis is used to adjust the resolution parameters of deconvolution, separating overlapping signal components and reducing mutual interference between signals. The approximate range of signal activity is determined at low resolution, and deconvolution separation is re-executed only for these identified activity ranges. This process can be performed recursively.

[0058] Each independent signal component undergoes quantitative analysis to calculate a series of descriptive parameters. These parameters include: total energy, peak energy, signal duration, frequency center point, spectral width, and correlation coefficient with signals at adjacent spatiotemporal locations. The feature parameter set of each component is input into a pre-defined classifier to determine the source attribute of the component. The deconvolutioned signal components are then preliminarily classified and identified to distinguish between rain echo components and noise signal components, which are then labeled as signal components.

[0059] This processing method, based on constrained deconvolution, mathematically deconstructs the convolution and distortion processes of the signal to achieve accurate reconstruction of real rainfall signals. It fundamentally separates noise, corrects system distortion, and resolves overlapping signals, thereby significantly improving the accuracy, resolution, and fidelity of the detected data and providing highly reliable data input for subsequent meteorological analysis.

[0060] The process of constructing the rain measurement link view through parallel processing is as follows:

[0061] The parallel processing includes signal scaling, echo gating, and spatiotemporal alignment.

[0062] The signal calibration process, which generates a calibrated echo signal, is as follows: Based on the microwave rain measurement signal and combined with the transmit power and system noise temperature parameters in the radar detection metadata, the receiver channel gain, system noise reference, and analog-to-digital conversion quantization constant are estimated to obtain the channel calibration coefficients and baseline offset; the microwave rain measurement signal is amplitude calibrated using the channel calibration coefficients, and thermal noise subtraction, amplitude normalization, and power unit conversion are performed to form an amplitude calibrated echo sequence; phase and frequency calibration are performed to correct the local oscillator frequency offset, sampling clock deviation, and echo phase drift, and to restore the time base consistent with the in-phase and quadrature components; equalization and baseline correction are performed on the amplitude and phase response of the receiver link, and the calibrated echo signal is output.

[0063] Specifically, in the channel calibration coefficient estimation stage, an internal calibration signal with known power is injected, and the theoretical received power is calculated using radar equations in conjunction with the transmit power Pt and system noise temperature Ts. This calculation is then compared with the actual ADC output value to obtain the receive channel gain G and quantization constant K. The baseline bias is determined by analyzing the statistical characteristics of the noise floor during signal-free periods, establishing a linear mapping relationship between the ADC code value and the physical power.

[0064] During amplitude calibration, the thermal noise floor is first subtracted from the original ADC code value by subtracting the noise reference value; then, gain compensation is performed using channel calibration coefficients to convert the digital code value into linear power units; finally, amplitude normalization is performed to ensure that signals from different channels and time periods have a consistent dynamic range.

[0065] Phase and frequency calibration uses phase-locked loop technology to correct the local oscillator frequency deviation, interpolation algorithm to compensate for sampling clock deviation, stable ground object echo or internal reference signal to track phase drift, and performs time-domain alignment and phase correction on the I / Q signals respectively to restore orthogonality and time base consistency.

[0066] The receive link equalization compensates for the unevenness of the amplitude-frequency response through a frequency domain filter bank, eliminates multipath effects and group delay differences based on a predictive filtering algorithm, and finally outputs a standardized echo signal with precise calibration of amplitude, phase and frequency.

[0067] Read the range gate index, station coordinates, antenna azimuth and elevation angles to determine the link crossing section and set echo gating to shield against ground clutter; establish a unified time axis with timestamps, resample the calibration echo signal and correct the echo delay; project the range gate onto the geographic coordinate system, establish the link unit index and correct for terrain occlusion; organize the calibration echo signal according to link unit and time step, write the echo amplitude, phase, and Doppler velocity, add radar detection metadata index, and construct a rainfall measurement link view;

[0068] The echo gating is based on the distance gate index and station coordinates to calculate the geographical location of each distance gate, combined with the digital elevation model to identify the ground object blocking area, and automatically shields the echoes of fixed targets such as buildings and mountains by setting power thresholds and correlation tests. Dynamic gating is performed using ground object clutter maps to retain the effective echo range of meteorological targets.

[0069] The spatiotemporal alignment process involves establishing a unified timeline based on GPS timestamps, resampling the calibration echo signals at different scanning times to compensate for antenna rotation and signal propagation delays, and employing a cubic spline interpolation algorithm to ensure the continuity and smoothness of the time series, achieving sub-second time alignment accuracy.

[0070] When projecting the distance gate, the distance gate in the polar coordinate system is converted to the WGS84 geographic coordinate system. Taking into account the curvature of the earth and atmospheric refraction correction, the latitude and longitude index of the link unit is established. The occlusion analysis is performed in combination with the terrain elevation data, and invalid link segments blocked by the terrain are marked.

[0071] During the link view construction process, the calibration echo data is organized into a two-dimensional grid according to time steps and link units. The echo amplitude, phase difference and Doppler frequency shift information are written one by one. At the same time, the corresponding radar detection metadata index is associated to form a structured four-dimensional spatiotemporal echo data cube, which supports efficient spatiotemporal query and meteorological parameter inversion analysis.

[0072] This embodiment effectively shields ground clutter and fixed target interference through intelligent echo gating, significantly improving the purity and reliability of meteorological echo signals. Unified time axis establishment and delay correction ensure the spatiotemporal consistency of multi-scan data, eliminating time deviations caused by antenna rotation and signal propagation. Distance gate geographic projection and terrain occlusion analysis enable precise spatial positioning, improving the geographic accuracy of precipitation distribution estimation. The construction method of the four-dimensional spatiotemporal data cube optimizes the data storage structure and access efficiency, providing a high-quality basic data source for subsequent meteorological parameter inversion, precipitation intensity estimation, and weather analysis, thus comprehensively improving the data processing capabilities and application value of the microwave rain measurement radar system.

[0073] Furthermore, the process of extracting rain echo feature groups based on the rain measurement link view and binding them to radar detection metadata to generate a rain measurement binding fingerprint is as follows: read the calibration echo signal according to the link unit and time step and locate the range gate index; extract the rain echo feature group composed of echo amplitude, phase, and Doppler velocity; align the feature group with the timestamp, station coordinates, antenna azimuth and elevation angles, and scanning mode in the radar detection metadata; complete the binding using the range gate index, link unit index, and time step as keys, and encapsulate it into a rain measurement binding fingerprint according to a fixed field order;

[0074] Specifically, during the calibration echo signal reading stage, the rain measurement link view is traversed according to the link unit index and time step sequence. The specific spatial location is located by the distance gate index, and the corresponding I / Q complex echo sequence is extracted from the four-dimensional data cube. At the same time, the data integrity flag is verified to ensure the validity and continuity of the read signal.

[0075] When extracting the feature set of rainfall echoes, the echo amplitude is calculated from the complex echo signal as the reflectivity intensity index, and the phase information is extracted for differential reflectivity and correlation coefficient calculation. Doppler velocity components are obtained through spectral analysis to reflect the motion characteristics of precipitation particles. A multi-scale feature enhancement mechanism is introduced when extracting the feature set of rainfall echoes: wavelet decomposition and short-time Fourier transform are performed on the original time-series signal and spectral analysis to extract echo features at different time scales and frequency components, characterizing the differences between short-term abrupt changes and continuous precipitation processes. Enhanced features and basic features are written together into the feature set field to form a multi-scale composite feature description. Quality control is performed on the feature parameters, outliers are removed, and data quality levels are marked. Through the multi-scale feature enhancement mechanism, echo features at different scales are extracted in the time and frequency domains, taking into account the differences between short-term abrupt changes and continuous precipitation, thus improving the richness and detail of the feature representation.

[0076] During the field alignment process, the extracted feature groups are precisely matched with radar detection metadata according to a unified spatiotemporal index to ensure a one-to-one correspondence between key parameters such as timestamp, site coordinates, antenna pointing angle and scanning mode and echo features. A hash table fast lookup algorithm is used to improve matching efficiency.

[0077] In the binding and encapsulation stage, a three-dimensional composite key value is constructed using the distance gate index, link unit index, and time step. The echo feature group and metadata field are structurally encapsulated according to a predefined standard format to generate a rain measurement binding fingerprint containing complete observation information and echo features.

[0078] This embodiment ensures the integrity and spatial accuracy of echo signal readings through ordered traversal of a four-dimensional data cube and distance-gate indexing. The comprehensive extraction of multi-dimensional echo feature groups covers the intensity, phase, and motion characteristics of precipitation particles. The quality control mechanism effectively removes abnormal data and marks quality levels. The hash table fast matching algorithm significantly improves the alignment efficiency between feature groups and metadata, ensuring accurate spatiotemporal information correlation. The structured encapsulation of three-dimensional composite key values ​​realizes the standardized organization of observation data. The generated rainfall measurement binding fingerprint contains complete echo features and detection condition information, providing a high-quality standardized data foundation for subsequent precipitation classification and identification, intensity estimation, and meteorological analysis. Overall, it improves the data processing accuracy and application efficiency of the microwave rainfall measurement system.

[0079] Furthermore, the specific process of constructing the rainfall measurement environmental noise fingerprint database is as follows: taking the rainfall measurement bound fingerprint as the basic unit, defining fingerprint fields that include the quantization encoding of rainfall echo feature groups, radar detection metadata index, link unit index, and geographic projection information; establishing a noise type encoding table and fingerprint dictionary, and generating fixed-length fingerprint vectors according to unified rules; constructing a primary key based on the link unit index, distance gate index, time step, and radar detection metadata index, establishing a path adjacency index and a time window index, and writing them into storage; setting primary key deduplication and idempotent writing, and retaining version snapshots to form the rainfall measurement environmental noise fingerprint database;

[0080] Specifically, in the fingerprint field definition stage, the echo amplitude, phase and Doppler velocity in the rain echo feature group are encoded using 8-bit quantization, the radar detection metadata index is represented by a 32-bit hash value, and the link unit index and geographic projection information are stored using 16-bit and 64-bit encoding respectively, forming a standardized fingerprint field structure.

[0081] When establishing the noise type coding table, a hierarchical coding system is constructed based on the characteristics of different noise sources such as ground clutter, sea clutter, meteorological clutter, and electromagnetic interference. Each noise type is assigned a unique 8-bit coding identifier. The fingerprint dictionary stores feature templates in key-value pair format, supporting fast pattern matching and similarity calculation.

[0082] Fixed-length fingerprint vector generation follows field concatenation rules, encapsulating 128-bit or 256-bit vectors in a fixed order of feature encoding, metadata index, spatial index, and timestamp to ensure format consistency and comparability of fingerprints from different sources.

[0083] The primary key is constructed using a composite hash algorithm that combines link unit index, distance gate index, time step and radar detection metadata index. A B+ tree structure path adjacency index is built to support spatial queries, and the time window index uses a sliding window mechanism to achieve efficient time-series retrieval.

[0084] In storage management, deduplication control is achieved through primary key hash verification, idempotent write is guaranteed by copy-on-write technology, and a version snapshot mechanism records the historical state of the fingerprint database, supporting data rollback and incremental updates to ensure the integrity and consistency of the fingerprint database.

[0085] The constructed rain measurement environmental noise fingerprint database has a dynamic noise adaptive update mechanism: during the operation of the fingerprint database, updates are triggered according to time windows and the quality scores of new samples. Newly collected rain measurement fingerprints are compared with existing templates. If a similarity threshold is reached, the original template parameters are updated. If a new noise pattern is formed, a new template is added. The timeliness and environmental adaptability of the fingerprint database are maintained by using a sliding window and incremental snapshot method. The dynamic noise adaptive update mechanism realizes the incremental correction of fingerprint templates and the expansion of new noise patterns, maintaining the timeliness and environmental adaptability of the fingerprint database. The fingerprint dictionary adopts a key-value pair structure to support fast pattern matching and similarity calculation. The generated fixed-length fingerprint vectors are standardized and encapsulated, providing efficient and reliable data support for subsequent noise identification and feature retrieval.

[0086] This embodiment ensures the uniformity of fingerprint field structure and the comparability of cross-source data through feature parameter quantization encoding and index mapping; a hierarchical noise type encoding system enables unique identification and fine-grained management of different noise sources, avoiding category confusion; a composite hash primary key combined with a B+ tree path adjacency index and a sliding window mechanism effectively improves the efficiency of spatial positioning and temporal retrieval. Hash verification, copy-on-write, and version snapshot mechanisms are introduced into storage management to ensure deduplication control, idempotent writing, and historical traceability of fingerprint data.

[0087] Furthermore, the specific process of performing echo fingerprint matching based on microwave propagation path awareness is as follows: pre-screening is performed based on the link unit index and time step call path adjacency index and time window index; consistency verification is performed on center frequency, pulse repetition frequency, scanning mode, antenna azimuth angle and elevation angle; sequence alignment is performed on the rain measurement bound fingerprint and the fingerprint in the database on the link unit sequence, and the matching score is calculated according to echo amplitude, phase, Doppler velocity and link unit index and geographic projection information; candidates are truncated according to threshold and sorting, and the collaborative matching results are output.

[0088] Specifically, during the consistency verification stage, key observation parameters such as center frequency, pulse repetition frequency, scanning mode, antenna azimuth and elevation angles are compared to eliminate fingerprints in the database that do not match the input fingerprint conditions, ensuring that the candidate set is consistent with the physical constraints of the rainfall measurement scenario.

[0089] During the sequence alignment stage, candidate fingerprints are expanded according to the link unit sequence, and the input fingerprints are compared with fingerprints in the database one by one. The matching score is calculated by combining echo amplitude, phase, Doppler velocity, link unit index and geographic projection information, and an ordered matching list is formed.

[0090] During the candidate selection stage, a set of high-scoring fingerprints is extracted based on preset thresholds and sorting rules to generate a candidate set of rainfall noise. The candidate set is then further aggregated by combining spatial neighborhood consistency and temporal series continuity to eliminate isolated or unstable matching results and retain noise candidates with spatiotemporal continuity.

[0091] Finally, the verified and aggregated candidate set is output as a collaborative matching result, along with matching scores, parameter verification results, and time window information, providing a reliable input basis for subsequent noise discrimination and label generation.

[0092] This embodiment ensures the efficiency of the fingerprint retrieval process and the rationality of the range constraints through joint pre-screening using path adjacency index and time window index; the consistency verification stage compares the observation parameters item by item, realizing the physical credibility and constraint effectiveness of the candidate results; the sequence alignment process calculates the matching score by integrating multi-dimensional features, ensuring the refinement of fingerprint comparison and the accuracy of judgment; the candidate screening stage combines threshold, sorting and spatiotemporal aggregation strategies to improve the stability and continuity of the result set; the output rainfall noise candidate set is accompanied by structured scoring and time information, providing a standardized and scalable data input foundation for subsequent processing.

[0093] Furthermore, the process of weighing multi-source evidence under the constraint of rainfall physical mechanism is as follows: construct evidence vector, wherein the evidence includes consistency between echo amplitude and path attenuation, consistency between phase evolution and propagation delay, consistency between Doppler velocity and scanning geometry, and time series coherence; normalize and weight the consistency scores of each piece of evidence under the constraint of rainfall physical mechanism to obtain the weighting score; and output preliminary noise labels according to the weighting score threshold and judgment rules.

[0094] Specifically, in the evidence construction stage, various observation features are first extracted from the rain measurement fingerprint to generate corresponding physical consistency indices. For echo amplitude, the energy attenuation curve with propagation distance is calculated and compared with the link transmission path model to extract amplitude consistency scores. For phase, the evolution trajectory of phase over time is tracked, and a phase consistency metric is established by combining propagation delay and refractive index changes. For Doppler velocity, a scanning geometric model is established based on antenna azimuth and elevation parameters, and the velocity distribution is fitted with theoretical values ​​to obtain velocity consistency scores. For time series, a sliding window and autocorrelation calculation method are used to evaluate the coherence and stability of the fingerprint in the time dimension.

[0095] In the evidence fusion stage, the consistency scores are first normalized to ensure uniformity of dimensions. Then, weights are assigned to each indicator according to the constraints set by the precipitation physics mechanism. For example, the weight of Doppler velocity consistency is increased under severe convective weather, and the weight of temporal coherence is increased under continuous precipitation scenarios. The evidence fusion process introduces a credibility assessment and weighting mechanism, assigning credibility factors to different evidence sources. The credibility level is determined based on sensor status, data missing rate, and the degree of interference in the observation environment. The contribution of each piece of evidence is dynamically adjusted in the weight allocation to suppress the impact of low-credibility evidence on the fusion result. A comprehensive weighted score is generated through weighted synthesis to reflect the degree of conformity between candidate noise and precipitation physics. The credibility assessment and weighting mechanism dynamically corrects the contribution of each piece of evidence based on sensor status and data integrity, achieving a refined assessment of candidate noise. The preliminary noise labels output under the weighted score threshold and judgment rules are traceable and can retain evidence scores and weight parameters for subsequent correction.

[0096] During the label generation stage, candidate fingerprints are classified based on a weighted score threshold. Candidates that clearly do not meet the physical constraints are marked as interference noise, while candidates that meet most of the constraints are retained as pending samples. Preliminary noise labels are then output based on the judgment rules. Simultaneously, while generating labels, the corresponding evidence scores, weighting coefficients, and judgment results are recorded in a version snapshot as a basis for subsequent model updates and parameter corrections.

[0097] To improve the robustness of the judgment, the weighted score calculation process can be dynamically calibrated by combining multi-source information. For example, radar field observations, meteorological station rainfall records, or satellite precipitation products can be introduced as auxiliary evidence. Anomaly scores can be corrected through external correction mechanisms, thereby improving the stability and reliability of multi-source evidence fusion.

[0098] This embodiment ensures comprehensive physical constraints on candidate fingerprints by constructing a multi-source evidence vector that includes amplitude attenuation, phase evolution, velocity geometry, and temporal coherence. Through normalization and weighted synthesis mechanisms, indicators of different dimensions are unified and their weights are dynamically adjusted in conjunction with the rainfall scenario. External observation data is also introduced for dynamic calibration, which effectively enhances the robustness and reliability of multi-source evidence fusion and provides a standardized and stable data foundation for noise type identification and label generation.

[0099] The process of multi-criteria fusion verification of preliminary noise labels and rainfall monitoring prior constraints is as follows:

[0100] Based on prior knowledge, the system calls upon and integrates long-term historical rainfall data and observation records from ground meteorological stations within the monitoring area. Through standardized processing, quality verification, and statistical analysis of this data, it extracts the typical spatiotemporal characteristics and physical parameter patterns that real rainfall events in the local environment should possess, such as the correspondence between echo intensity and actual rainfall intensity, and the consistency threshold of adjacent microwave link signals. This refined and solidified knowledge is constructed into a standardized prior criterion table.

[0101] The initial noise label is used to automatically initiate multi-criteria fusion verification, comparing the candidate signal with each standard in the prior criterion table. The comparison covers four core dimensions: temporal continuity to determine whether the signal is an isolated instantaneous pulse rather than a continuous rainfall echo; spatial adjacent link consistency to check whether there are similar signal patterns on its surrounding links to distinguish between local interference and regional rainfall; consistency between echo amplitude and rainfall intensity to assess whether the signal strength conforms to the physical laws of local rainfall; and meteorological observation record verification to directly compare with the measured data from the nearest meteorological station. Each comparison generates a "pass" or "fail" label.

[0102] The comparison results are weighted and voted on according to the preset weights and veto rules. The preset weights and veto rules are that the weights of the four criteria are set to be primary and secondary. The verification of meteorological observation records is given the highest decisive weight, followed by the consistency of spatial adjacent links, then the consistency of the relationship between echo amplitude and rainfall intensity, and finally the temporal continuity.

[0103] This embodiment sets a strict rejection rule: if a candidate signal is analyzed and is considered to be rainfall, but the actual rainfall records of the ground meteorological station near its corresponding location and time point clearly show no rain or huge differences in rainfall intensity (for example, radar inversion shows heavy rain but actual measurement shows light rain), then regardless of the results of other criteria, the signal will be directly judged as noise. This rule is used to ensure that the identification results do not deviate significantly from the actual ground conditions.

[0104] The review mechanism prevents hasty conclusions from being drawn for samples whose scores in weighted voting are close to the decision threshold. Instead, it automatically expands the scope of analysis by utilizing path adjacency indexes and time window indexes. By examining the signal's performance over a longer time series and a wider spatial neighborhood, more comprehensive information can be obtained, allowing for the review of these uncertain samples and the making of a more reliable final judgment.

[0105] After all verification and review processes are completed, signal components identified as environmental noise are assigned a final noise label. All identified noise components are precisely removed from the original microwave rain measurement signal. After this purification process, the final output is a pure and reliable rain measurement signal.

[0106] By fusing the initially identified noise labels with prior constraints based on historical data and meteorological observations, the accuracy and reliability of noise identification are significantly improved. A comprehensive judgment is made using multiple dimensions, including temporal continuity, spatial consistency, physical relationship consistency, and external meteorological data verification, avoiding misjudgments that might arise from a single criterion. Furthermore, by setting up weighted voting, veto rules, and a review mechanism, the robustness of the algorithm is further enhanced, enabling secondary review of uncertain samples and ensuring higher credibility of the final output noise labels.

[0107] This embodiment, by introducing advanced deconvolution processing with non-negativity and sparsity constraints, can achieve efficient separation of rainfall echoes and environmental noise in the initial stage of signal processing, providing a clear foundation for subsequent identification. By constructing a feature fingerprint database bound to radar metadata and performing intelligent matching based on propagation path awareness, the specific type of noise can be accurately and efficiently identified. By introducing rainfall physical mechanisms as constraints for trade-off judgment, a closed-loop logic of separation-matching-verification is formed, which greatly improves the accuracy of noise removal, effectively avoids misjudging real rainfall signals as noise, and ultimately significantly improves the purity and reliability of microwave rainfall data.

[0108] Example 2:

[0109] The method of this invention is applied to a rainfall monitoring scenario along a highway in a city during a rainstorm. In this scenario, dual-polarized weather radar and microwave communication link equipment are deployed along the highway to acquire real-time signals during the rainstorm and to identify and label the environmental noise, thereby providing stable data support for traffic scheduling and meteorological services.

[0110] First, we acquire aliased microwave rain measurement signals containing real rain echoes and complex environmental noise caused by high-speed vehicles, communication base station signals, etc., and simultaneously extract radar detection metadata. We apply aliased signal deconvolution processing technology with non-negative and sparse constraints, and use the system impulse response as prior knowledge to jointly optimize in the time and frequency domains, so as to separate the aliased signal into multiple independent signal components with high precision.

[0111] Parallel signal calibration, echo gating, and spatiotemporal alignment are performed on the separated signal components to construct a detailed rain measurement link view. Based on the rain measurement link view, rain measurement echo feature groups composed of echo amplitude, phase, Doppler velocity, etc. are extracted and bound to the corresponding radar detection metadata to generate a structured rain measurement binding fingerprint.

[0112] A rain measurement environmental noise fingerprint database specifically designed for this highway environment is constructed. Echo fingerprint matching based on microwave propagation path perception is performed. The rain measurement binding fingerprints generated on-site are aligned with the noise fingerprints stored in the database, and the matching degree is calculated to generate a collaborative matching result.

[0113] By combining the physical mechanisms of rainfall with the multi-source evidence of the collaborative matching results, the specific type of noise is determined, and all signal components identified as noise are accurately removed, outputting a clean and reliable rainfall signal, providing a high-quality data foundation for traffic warnings and meteorological services for this road section.

[0114] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for microwave rain measurement environmental noise identification based on feature fingerprint library matching, characterized in that, include: The process involves acquiring aliased microwave rain measurement signals containing rainfall echoes and environmental noise, and extracting radar detection metadata from these signals. Then, deconvolution processing is applied to separate the aliased microwave rain measurement signals into signal components. This deconvolution process incorporates non-negativity and sparsity constraints, and uses the system impulse response as a priori for joint optimization in the time and frequency domains. The process of separating the aliased microwave rain measurement signals into signal components using deconvolution processing is as follows: The acquired aliased microwave rain measurement signal is processed using a deconvolution algorithm, and N signal components are separated using multi-resolution analysis technology. The N signal components include at least one candidate rainfall signal component and at least one candidate noise signal component. The deconvolution algorithm analyzes the time and frequency domain characteristics of the signal, filters and enhances each signal component, and identifies the heterospectral characteristics of the rainfall echo and the environmental noise signal. The deconvolution introduces non-negativity and sparsity constraints, and uses the system impulse response as a priori to achieve joint optimization in the time and frequency domains. Preliminary signal classification and identification are performed on the deconvolutioned signal components to distinguish between rain echo components and noise signal components, and these components are marked as candidate signal components. The signal components are processed in parallel to form a rain measurement link view. The parallel processing includes signal calibration, echo gating, and spatiotemporal alignment. Rain measurement echo feature groups are extracted based on the rain measurement link view to generate a rain measurement binding fingerprint that is bound to the radar detection metadata. A rain measurement environmental noise fingerprint database is constructed, echo fingerprint matching based on microwave propagation path perception is performed, collaborative matching results are generated, and multi-source evidence is weighed and judged in combination with rainfall physical mechanism constraints to determine the noise type. Signal components identified as noise are removed to obtain pure rain measurement signals.

2. The method for identifying environmental noise in a microwave rain measurement according to claim 1, wherein, The process of acquiring the aliased microwave rain measurement signal containing rainfall echoes and environmental noise and extracting radar detection metadata from it is as follows: Determine the observation sampling configuration, and set the center frequency, pulse repetition frequency, pulse width, and receiver gain; Establish a unified spatiotemporal reference, obtain timestamps, site coordinates, antenna azimuth and elevation angles, and bind them to the sampling task; The system performs acquisition and preprocessing of aliased microwave rain measurement signals. After down-conversion and bandpass filtering, the aliased microwave rain measurement signals are sampled and quantized for in-phase and quadrature components. The system analyzes equipment and observation control records to extract transmit power, noise temperature, range gate index and scanning mode, and encapsulates them into radar detection metadata according to field specifications.

3. The method of claim 1, wherein, The process of constructing the rain measurement link view through parallel processing is as follows: Read the distance gate index, site coordinates, antenna azimuth and elevation angles, determine the link crossing section and set echo gating to shield ground clutter; A unified timeline is established using timestamps, and the calibrated echo signals are resampled and the echo delay is corrected. The range gate is projected onto the geographic coordinate system, a link unit index is established, and terrain occlusion is corrected. The calibrated echo signals are organized according to link units and time steps, and the echo amplitude, phase, and Doppler velocity are written. Radar detection metadata index is added to construct a rain measurement link view.

4. The method for identifying environmental noise in a microwave rain gauge according to claim 1, wherein, The process of extracting rainfall echo feature groups based on the rainfall link view and generating a rainfall binding fingerprint bound to the radar detection metadata is as follows: The echo signal is read in link unit and time step, and the distance gate index is located; a rain measurement echo feature group composed of echo amplitude, phase and Doppler velocity is extracted; the feature group is field-aligned with the timestamp, site coordinates, antenna azimuth and elevation angle and scanning mode in the radar detection metadata; the binding is completed with the distance gate index, link unit index and time step as the key, and is packaged into a rain measurement binding fingerprint in a fixed field order.

5. The method for identifying environmental noise in a microwave rain gauge according to claim 1, wherein, The specific process of constructing the rain measurement environmental noise fingerprint library is as follows: The rain measurement environmental noise fingerprint library takes the rain measurement binding fingerprint as a basic unit, defines a fingerprint field containing the quantized encoding of the rain measurement echo feature group, the radar detection metadata index, the link unit index and the geographic projection information; a noise type coding table and a fingerprint dictionary are established, a fixed length fingerprint vector is generated according to a unified rule; a primary key is constructed according to the link unit index, the distance gate index, the time step and the radar detection metadata index, a path adjacency index and a time window index are established and written into storage; the primary key is de-duplicated and the version snapshot is retained to form the rain measurement environmental noise fingerprint library.

6. The method for identifying environmental noise in a microwave rain gauge according to claim 1, wherein, The specific process of performing echo fingerprint matching based on microwave propagation path perception to generate a collaborative matching result is as follows: The echo fingerprint matching based on microwave propagation path perception performs pre-screening according to the link unit index and the time step by calling the path adjacency index and the time window index; the consistency of the center frequency, the pulse repetition frequency, the scanning mode, the antenna azimuth and the elevation angle is checked; the rain measurement binding fingerprint and the rain measurement environmental noise fingerprint in the fingerprint library are sequentially aligned on the link unit sequence, and the matching score is calculated according to the echo amplitude, the phase, the Doppler velocity, the link unit index and the geographic projection information; the threshold and the sorting are used to intercept the candidates, and the collaborative matching result is output.

7. The method of claim 1, wherein, The process of weighting multiple sources of evidence under the constraint of the physical mechanism of rainfall is as follows: An evidence vector is constructed, including the consistency of echo amplitude and path attenuation, the consistency of phase evolution and propagation time delay, the consistency of Doppler velocity and scanning geometry, and the time sequence continuity; under the constraint of the physical mechanism of rainfall, the consistency of each evidence is normalized and weighted to obtain a weighting score; the threshold of the weighting score and the judgment rule are used to output a preliminary noise label.

8. The method for identifying environmental noise in a microwave rain gauge according to claim 1, wherein, The process of multi-criteria fusion verification of the preliminary noise label and the rainfall monitoring prior constraint is as follows: A rainfall monitoring prior constraint is established by combining historical rainfall data and meteorological observation records to form a prior criterion table; A multi-criteria set is set, including time sequence continuity, spatial adjacent link consistency, echo amplitude and rain intensity relationship consistency, and meteorological observation record checking; the preliminary noise label is compared with the prior criterion table item by item to generate a pass identification; the pre-set weight and the veto rule are used for weighted voting, the samples that do not pass are rechecked in the range of path adjacency index and time window index, and the final noise label is output; The signal components confirmed as noise are removed to obtain pure rain measurement signals.

Images