Particle morphology analysis system based on high-resolution airborne cloud particle imager

Abstract

The present application relates to the technical field of particle imaging, in particular to a particle morphology analysis system based on a high-resolution airborne cloud particle imager, comprising a particle motion state sensing unit, a multi-frame imaging trigger acquisition unit, a particle motion-morphology coupling reconstruction unit, the particle motion-morphology coupling reconstruction unit adopts an embedded processor, integrates acceleration data output by the IMU to obtain particle relative speed, restores particle motion path in combination with multi-frame images, and a morphology data integration output unit.The present application identifies non-spherical particle types such as ice crystals and graupel based on the profile difference of real morphology characteristics through a posture compensation module, extracts a core morphology axis as a correction reference, and finally performs directional rotation correction according to the particle motion direction, thereby improving the adaptability of non-spherical particle morphology and motion posture and solving the morphology misjudgment problem caused by the lack of non-spherical particle posture adaptation correction mechanism.

Description

Technical Field

[0001] This invention relates to the field of particle imaging technology, and more specifically, to a particle morphology analysis system based on a high-resolution airborne cloud particle imager. Background Technology

[0002] Cloud particle imagers are key devices for capturing cloud particle images and acquiring basic information about particle morphology. High-resolution airborne cloud particle imagers, with their scene adaptability, have become a core tool for acquiring cloud particle data in aviation meteorological observations. The morphological characteristics of cloud particles are core data support for atmospheric physics research, precipitation mechanism analysis, and aviation safety assurance. In airborne environments, it is necessary to accurately capture the three-dimensional structure of particles in dynamic scenes. However, during high-resolution imaging, interference caused by particle motion can easily lead to morphological distortion. How to integrate particle motion state with imaging data to achieve efficient reconstruction of the true morphology is a core research and development need in this field.

[0003] In the existing technology, relevant patents have explored the field of particle morphology analysis. For example, Chinese patent CN202010719660.0 discloses a method for determining the particle morphology of strongly extinct biomaterials in a target wavelength band. This method constructs geometric models of different particle morphologies of biomaterials, calculates the complex refractive index and extinction cross sections in each wavelength band, and filters the parameters corresponding to the maximum extinction cross section to determine the particle morphology of strongly extinct biomaterials in the target wavelength band with high accuracy. Another example is Chinese patent CN202310356432.5, which discloses a particle morphology recognition method based on YOLOv7 and interferometric imaging technology. This method uses an interferometric particle imaging system to obtain mixed particle field interferometric defocus images and create a dataset. By improving the YOLOv7 network (adding a collaborative attention module) for training and optimization, it achieves high-precision recognition of mixed particle field morphology, which can provide support for cloud particle field information acquisition.

[0004] Despite the design advantages of the aforementioned technical solutions, they also suffer from the following technical shortcomings: First, they are not adapted to motion interference in dynamic airborne scenarios: CN202010719660.0 focuses on determining the morphology of biological material particles based on their extinction characteristics, relying primarily on geometric models and extinction cross-section calculations. It does not consider the influence of particle motion parameters in the airborne environment, and therefore cannot cope with imaging blur and morphological deviations caused by dynamic motion, making it only suitable for static or near-static scenarios. Second, they lack a posture adaptation and correction mechanism for non-spherical particles: CN202310356432.5 While interferometric imaging and deep learning are used to achieve morphology recognition, they do not address the differences in motion and posture of non-spherical particles such as ice crystals and graupel. Relying solely on image feature training makes it difficult to eliminate morphology misjudgments caused by posture changes. Thirdly, a bidirectional coupling correction logic for motion and morphology is not established: CN202310356432.5 and CN202010719660.0 do not integrate particle motion and morphology data, failing to correct morphology distortion through motion parameters or optimize motion parameter calculations using morphology features, thus failing to simultaneously address the mutual influence of motion drift and morphology distortion. Therefore, we propose a particle morphology analysis system based on a high-resolution airborne cloud particle imager. Summary of the Invention

[0005] The purpose of this invention is to provide a particle morphology analysis system based on a high-resolution airborne cloud particle imager, in order to solve the problems mentioned in the background art, such as motion interference not adapted to airborne dynamic scenes, lack of attitude adaptation and correction mechanism for non-spherical particles, and lack of bidirectional coupling correction logic between motion and morphology.

[0006] To address the aforementioned technical problems, the present invention aims to provide a particle morphology analysis system based on a high-resolution airborne cloud particle imager, comprising:

[0007] The particle motion state sensing unit uses a high-precision IMU to collect the raw acceleration and angular velocity data of the airborne cloud particle imager in real time; after Kalman filtering for noise reduction and quaternion method attitude angle calculation, it analyzes the acceleration and angular velocity parameters of the particles in the airborne environment and outputs the real-time motion state of the particles.

[0008] A multi-frame imaging trigger acquisition unit is provided, which uses a micro-array shutter CMOS and an FPGA. The FPGA synchronously triggers the CMOS to expose the CMOS in separate regions, and acquires and buffers multiple frames of particle images.

[0009] The particle motion-morphology coupling reconstruction unit employs an embedded processor to integrate the acceleration data output by the IMU to obtain the relative velocity of the particles, and reconstructs the particle motion path by combining multiple frames of images. Motion stability is determined by the velocity change rate of the particles in consecutive frames. A motion-morphology coupling mapping model is established, using the fuzzy particle shape, relative velocity, motion direction, and motion stability determination results as inputs to generate realistic morphological features. For the motion posture of non-spherical particles, directional rotation compensation is performed based on the motion direction. A drift-morphology coupling correction algorithm is run, synchronously eliminating motion drift and morphological distortion through bidirectional feedback adjustment of motion data and morphological data, outputting realistic three-dimensional morphological data of the particles.

[0010] The morphological data integration and output unit stores the reconstructed particle morphological data locally and adapts to remote transmission protocols to output the morphological analysis results to external parties.

[0011] As a further improvement to this technical solution, the particle motion state sensing unit includes a data acquisition module, a data preprocessing module, a motion state calculation module, and a data output module, wherein:

[0012] The data acquisition module establishes a communication connection with the high-precision IMU to acquire the raw acceleration and angular velocity data of the airborne cloud particle imager;

[0013] The data preprocessing module performs outlier removal and noise suppression on the collected raw data, and uses the Kalman filter algorithm for noise suppression.

[0014] The motion state calculation module integrates the preprocessed acceleration data to obtain the carrier-related displacement, combines the angular velocity data to calculate the carrier attitude angle using the quaternion method, analyzes the particle motion parameters based on the relative motion relationship between the carrier motion and the particles, and fuses them to generate the real-time motion state of the particles.

[0015] The data output module transmits the real-time motion state of the particles to the particle motion-morphology coupling reconstruction unit.

[0016] As a further improvement to this technical solution, the multi-frame imaging trigger acquisition unit includes a trigger driving module, a zone exposure control module, an image acquisition module, and an image buffer module, wherein:

[0017] The triggering drive module generates synchronous trigger signals based on the FPGA;

[0018] The partitioned exposure control module receives the synchronous trigger signal and controls the microarray shutter CMOS to perform independent exposure of different regions.

[0019] The image acquisition module acquires particle images by performing regional exposure using a microarray shutter CMOS.

[0020] The image caching module caches the acquired multi-frame particle images.

[0021] As a further improvement to this technical solution, the particle motion-morphology coupling reconstruction unit includes a motion parameter calculation module, a motion state determination module, a morphology mapping modeling module, an attitude compensation module, a coupling correction module, and a morphology data output module, which are connected in sequence via communication.

[0022] The motion parameter calculation module integrates the acceleration data output by the IMU to obtain the relative velocity of the particles, and combines it with multiple frames of particle images to reconstruct the particle motion path.

[0023] The motion state determination module determines motion stability based on the velocity change rate of particles in consecutive frames.

[0024] The morphology mapping modeling module establishes a motion-morphology coupling mapping model, taking the fuzzy particle morphology, relative velocity, motion direction and motion stability determination results as inputs to generate real morphological features.

[0025] The attitude compensation module performs directional rotation compensation based on the direction of motion for the motion attitude of non-spherical particles.

[0026] The coupling correction module runs a drift-morphology coupling correction algorithm, which eliminates motion drift and morphological distortion synchronously through bidirectional feedback adjustment of motion data and morphological data.

[0027] The morphology data output module outputs three-dimensional real morphology data of the particles.

[0028] As a further improvement to this technical solution, the process of integrating the acceleration data to obtain the relative velocity of the particles and reconstructing the motion path in the motion parameter calculation module includes the following steps:

[0029] S31.1 Obtain the continuous acceleration data output by the IMU and record the acceleration value at each moment;

[0030] S31.2. The relative velocity of the particles is obtained by calculating the acceleration data using the trapezoidal integral method. ;

[0031] S31.3 Extract contour feature points from multi-frame particle images, and obtain the pixel displacement of particles between adjacent frames through an inter-frame feature point matching algorithm;

[0032] S31.4. Combining the calibration relationship between the image and the actual space, the pixel displacement is converted into actual displacement to help reconstruct the continuous motion path of the particles, making... It is consistent with the temporal changes of the motion path.

[0033] As a further improvement to this technical solution, the motion state determination module includes the following steps in determining motion stability:

[0034] S32.1 Obtain the relative particle velocities of two consecutive frames output by the motion parameter calculation module. Record the time interval between the acquisition of two frames;

[0035] S32.2, based on two frames The rate of change of velocity was calculated using the time interval. ;

[0036] S32.3, according to Determine the motion state within the numerical range: when When it is in the low range, it is judged as a change in motion; when When the motion is in the high range, it is determined to be stable, and the determination result is synchronously transmitted to the morphology mapping modeling module.

[0037] As a further improvement to this technical solution, the morphological mapping modeling module includes a feature extraction submodule, a weight configuration submodule, and a model calculation submodule that are sequentially connected in communication, wherein:

[0038] The feature extraction submodule extracts from fuzzy particle morphology Separate and extract edge features and texture features;

[0039] The weight configuration submodule is based on the output of the motion state determination module. Adjust the fusion weight of morphological features and motion parameters, increase the weight ratio of texture features when the motion is stable, and increase the weight ratio of edge features when the motion is variable speed.

[0040] The model computation submodule uses extracted morphological features and particle relative velocities. , direction of movement and Using this as input, perform fusion operations to generate realistic morphological features. .

[0041] As a further improvement to this technical solution, the attitude compensation module includes a particle type identification submodule, a shape axis extraction submodule, and a rotation correction submodule that are sequentially connected in communication, wherein:

[0042] The particle type recognition submodule is based on real morphological features. The differences in the contour features distinguish between two types of non-spherical particles: ice crystals and graupel.

[0043] The morphological axis extraction submodule targets different types of non-spherical particles, from... The core morphological axis is extracted as the calibration benchmark.

[0044] The rotation correction submodule performs directional rotation correction on the core shape axis based on the particle's motion direction, thereby improving the adaptability of the particle's shape and motion posture.

[0045] As a further improvement to this technical solution, the process of performing bidirectional feedback adjustment in the coupling correction module includes the following steps:

[0046] S35.1, Particle relative velocity output by the motion parameter calculation module Based on this, the true morphological characteristics were corrected. The distorted parts were used to obtain the morphological features after preliminary correction. ;

[0047] S35.2, with With fuzzy particle morphology Based on the characteristic differences, the integral benchmark of the motion parameter solution module is optimized in reverse.

[0048] S35.3 Repeat steps S35.1 to S35.2 until the amplitude of motion drift and shape distortion changes narrows synchronously, and output the final three-dimensional real shape data of the particles.

[0049] As a further improvement to this technical solution, the morphological data integration and output unit includes a data verification module, a local storage module, a transmission protocol adaptation module, and an output control module, which are connected in sequence via communication.

[0050] The data verification module receives the three-dimensional real shape data of the particles, verifies the integrity and format of the data, ensures the integrity of the data through the CRC32 verification algorithm, removes abnormal data and triggers retransmission;

[0051] The local storage module uses industrial-grade storage media to classify and store the verified data according to "collection time-particle type", establishes an index for easy retrieval, and cyclically overwrites and prioritizes the retention of the latest data when the storage is full.

[0052] The transmission protocol adaptation module dynamically adapts the communication protocol according to the transmission scenario. When transmitting to airborne internal equipment, it adapts to the CAN bus protocol, and when transmitting to ground stations remotely, it adapts to the airborne Ethernet protocol. It encapsulates the morphological data into protocol frames and adds an identity field.

[0053] The output control module coordinates the priorities of local storage and remote transmission. When there is a real-time transmission requirement, it prioritizes ensuring data output bandwidth. When only offline analysis is required, it controls the data to only perform local storage and synchronously outputs data processing status indicators.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] 1. This invention sets up a particle motion state sensing unit and uses a high-precision IMU to collect raw data of particle acceleration and angular velocity in an airborne environment. The data is then denoised using a Kalman filter algorithm and the attitude angle is calculated using a quaternion method. Simultaneously, it is combined with an FPGA synchronous trigger microarray shutter CMOS regional independent exposure technology in a multi-frame imaging trigger acquisition unit to achieve accurate capture of the real-time particle motion state and acquisition and buffering of multiple clear images. This effectively adapts to dynamic airborne scenarios and solves the problem that existing technologies cannot cope with the imaging blur and shape deviation caused by dynamic motion.

[0056] 2. This invention uses the attitude compensation module in the particle motion-morphology coupling reconstruction unit to first identify non-spherical particle types such as ice crystals and graupel based on the contour differences of real morphological features, then extract the core morphological axis as the correction benchmark, and finally perform directional rotation correction according to the particle motion direction. This invention specifically improves the adaptability of non-spherical particle morphology and motion attitude, and solves the problem of morphological misjudgment caused by the lack of non-spherical particle attitude adaptation correction mechanism in the prior art.

[0057] 3. This invention establishes a motion-morphology coupling mapping model, taking the fuzzy particle morphology, relative velocity, motion direction, and motion stability determination results as inputs. It dynamically adjusts the fusion weights of morphological features and motion parameters based on the velocity change rate quantification index, and simultaneously runs a drift-morphology coupling correction algorithm. Through bidirectional feedback adjustment of motion data and morphological data, it achieves the synchronous elimination of motion drift and morphological distortion, solving the defect of existing technologies that do not establish bidirectional motion-morphology coupling correction logic, and can output three-dimensional real morphological data of particles. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the system framework of the present invention;

[0059] The meanings of the labels in the diagram are as follows:

[0060] 1. Particle motion state sensing unit; 11. Data acquisition module; 12. Data preprocessing module; 13. Motion state calculation module; 14. Data output module;

[0061] 2. Multi-frame imaging trigger acquisition unit; 21. Trigger drive module; 22. Zone exposure control module; 23. Image acquisition module; 24. Image buffer module;

[0062] 3. Particle motion-morphology coupling reconstruction unit; 31. Motion parameter calculation module; 32. Motion state determination module; 33. Morphology mapping modeling module; 34. Attitude compensation module; 35. Coupling correction module; 36. Morphology data output module;

[0063] 4. Morphological data integration and output unit; 41. Data verification module; 42. Local storage module; 43. Transmission protocol adaptation module; 44. Output control module. Detailed Implementation

[0064] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0065] like Figure 1 As shown, this embodiment provides a particle morphology analysis system based on a high-resolution airborne cloud particle imager, including:

[0066] The particle motion state sensing unit 1 uses a high-precision IMU to collect the raw acceleration and angular velocity data of the airborne cloud particle imager in real time; after Kalman filtering for noise reduction and quaternion method attitude angle calculation, it analyzes the acceleration and angular velocity parameters of the particles in the airborne environment and outputs the real-time motion state of the particles.

[0067] In this embodiment, the particle motion state sensing unit 1 includes a data acquisition module 11, a data preprocessing module 12, a motion state calculation module 13, and a data output module 14, wherein:

[0068] The data acquisition module 11 establishes a communication connection with the high-precision IMU to acquire the raw acceleration and angular velocity data of the airborne cloud particle imager;

[0069] Specifically, the data acquisition module 11 is adapted to the airborne dynamic environment and uses a high-precision IMU integrating a triaxial accelerometer and a triaxial angular velocity meter. It possesses vibration resistance and wide temperature range characteristics to withstand mechanical shocks and temperature changes during flight. The data acquisition module 11 establishes a connection with the IMU via an SPI communication interface, with the communication clock frequency set to 1MHz. A standard frame structure is used to achieve data transmission and integrity verification. The acquisition frequency is configured to 1kHz, and the IMU range is set to acceleration ±2g and angular velocity ±500° / s, covering the dynamic range of cloud particle motion. Acquired data is stored in the order of "timestamp-triaxial acceleration-triaxial angular velocity," with the timestamp generated synchronously with the acquisition action. After power-on, the module first completes IMU self-test and initialization. If the self-test passes, it enters continuous acquisition mode; if it fails, a restart mechanism is triggered to ensure the reliability of the acquisition link.

[0070] The data preprocessing module 12 performs outlier removal and noise suppression on the collected raw data, and uses the Kalman filter algorithm for noise suppression;

[0071] Specifically, the data preprocessing module 12 first passes through 3 Outlier removal criteria: Using 50 consecutive sets of collected data as a window, calculate the mean of the data for each axis. with standard deviation Exceeding Outliers within a range are replaced using linear interpolation of two adjacent frames of valid data to avoid data gaps. Kalman filtering is then used for noise suppression, adapting to 6D data processing scenarios (three-axis acceleration + three-axis angular velocity), as detailed below:

[0072] ;

[0073] in for Time-state vector; , They are respectively time The true value of the axis acceleration; They are respectively time True value of shaft angular velocity; superscript " " indicates vector transpose;

[0074] Equations of state: ;

[0075] in: It is a 6×6 state transition matrix (identity matrix).

[0076] for Time-state vector;

[0077] for Time-process noise (Gaussian white noise).

[0078] Observation equation: ;

[0079] in: for Time-based observation vector (current frame data after outlier removal);

[0080] The observation matrix is ​​a 6×6 matrix (identity matrix). for Observe noise (Gaussian white noise) at all times.

[0081] In the above equation, , Adapts to the stable particle motion characteristics within a short period of time; and The covariance matrix is ​​set based on the noise characteristics of the IMU itself, and high signal-to-noise ratio data can be output by iterating according to the standard Kalman filtering process.

[0082] The motion state calculation module 13 integrates the preprocessed acceleration data to obtain the carrier-related displacement, combines the angular velocity data to calculate the carrier attitude angle using the quaternion method, analyzes the particle motion parameters based on the relative motion relationship between the carrier motion and the particles, and fuses them to generate the real-time motion state of the particles.

[0083] Specifically, the motion state calculation module 13 first executes a zero-bias calibration process: during the initialization phase, the system remains stationary for 3 seconds, and preprocessed triaxial acceleration data is collected during this period. The mean value of each axis is calculated as the zero-bias value. Before subsequent integration calculations, the acceleration data is subtracted from the corresponding axis's zero-bias value to eliminate the cumulative integration error caused by the zero-bias. Based on the preprocessed acceleration data, the trapezoidal integration method is used to calculate the particle's relative velocity and the carrier's related displacement. The integration time step is consistent with the data acquisition frequency (1ms), and the velocity calculation formula is:

[0084] ;

[0085] in: for time Relative velocity of axon particles; for time Relative velocity of axon particles; for After deducting zero bias at any time Axial acceleration; for After deducting zero bias at any time Axial acceleration; The integration time step is 1 ms.

[0086] Initial velocity set to ( , ) are respectively Relative velocity of axion particles, calculation logic and (Completely identical); the carrier-related displacement is obtained by the second integral of the velocity, and the calculation formula is:

[0087] ;

[0088] in: for time Shaft-borne related displacement; for time The displacement related to the shaft carrier; the meanings of the other symbols are consistent with the velocity calculation formula.

[0089] The initial displacement is set to ( , ) are respectively The relevant displacement of the shaft carrier, the calculation logic and completely consistent).

[0090] Specifically, the attitude angle calculation uses the quaternion method to avoid the Euler angle gimbal lock problem, and the initial quaternion is set as follows: ( (These are the four components of the quaternion). The quaternion is updated based on angular velocity data with a 1ms sampling interval. The update formula is:

[0091] ;

[0092] in: , , , for The four components of the time quaternion; , , , for The four components of the time quaternion; For the current frame Axial angular velocity; The sampling interval is 1ms.

[0093] Furthermore, after the quaternion is updated, it needs to be normalized and then converted into Euler angles (roll, pitch, and yaw) to accurately describe the real-time attitude of the carrier and provide an attitude reference for subsequent particle motion parameter analysis.

[0094] Furthermore, in the particle motion parameter analysis stage, two coordinate systems are first defined: the inertial coordinate system (i.e., the geodetic coordinate system), The axis points east. The axis points north. (axis perpendicular to the ground and upward) and the carrier coordinate system (i.e., the imager mounting coordinate system). Along the optical axis of the imager Lateral aspect of the imager (Axial along the longitudinal direction of the imager).

[0095] Using the attitude angles (roll, pitch, and yaw) obtained in the previous calculation, a rotation matrix is ​​constructed from the carrier coordinate system to the inertial coordinate system (this rotation matrix is ​​obtained by multiplying the three basic rotation matrices of roll, pitch, and yaw). Then, the absolute motion data of the carrier (including the absolute acceleration and absolute velocity of the carrier) provided by the airborne navigation system is fused together. Combining the relative motion relationship between the carrier and the particle, the absolute acceleration, absolute velocity, and direction of motion of the particle in the inertial coordinate system are obtained by reverse analysis. Finally, these parameters are fused and integrated to generate a complete real-time motion state dataset of the particle.

[0096] Data output module 14 transmits the real-time motion state of the particles to particle motion-morphology coupling reconstruction unit 3.

[0097] Specifically, the data output module 14 uses the LVDS interface to transmit data to the particle motion-morphology coupling reconstruction unit 3, with the transmission rate set to 1Mbps to adapt to the data acquisition frequency and subsequent unit processing requirements.

[0098] Meanwhile, the data output module 14 has a built-in 1024-byte FIFO buffer. When the buffer is half full, transmission is initiated to ensure continuous data output. A fixed-format data frame is designed, including a frame header, timestamp, particle motion parameters, and checksum. The timestamp is consistent with that of the data acquisition module 11, and a 1kHz synchronization clock signal is output to achieve precise timing alignment between motion data and subsequent image data. If a buffer overflow or checksum failure is detected, an alarm and retransmission mechanism is triggered (up to 3 retransmissions). If the retransmission still fails, a fault log is recorded to ensure data transmission reliability.

[0099] Multi-frame imaging trigger acquisition unit 2 adopts micro-array shutter CMOS and FPGA. The FPGA synchronously triggers CMOS to expose independently in different regions, and acquires and buffers multiple frames of particle images.

[0100] In this embodiment, the multi-frame imaging trigger acquisition unit 2 includes a trigger driving module 21, a zone exposure control module 22, an image acquisition module 23, and an image buffer module 24, wherein:

[0101] Trigger driver module 21 generates synchronous trigger signals based on FPGA;

[0102] Specifically, the trigger driver module 21 is based on an industrial-grade FPGA and is adapted to complex airborne operating conditions. The trigger driver module 21 receives a 1kHz synchronous clock signal output from the particle motion state sensing unit and performs phase calibration through the FPGA's internal phase-locked loop (PLL) to ensure that the timing deviation between the trigger signal and the motion data is ≤1μs, achieving frame-level alignment. The generated synchronous trigger signal is a 1kHz periodic pulse with a pulse width of 10μs, using a 3.3VTTL level, meeting the CMOS trigger level requirements of the microarray shutter. The FPGA has embedded trigger logic, supporting both normal acquisition (continuous output) and abnormal stop (when receiving alarm signals) modes, while simultaneously outputting a trigger status indicator for easy coordination with subsequent modules. Differential routing is used for trigger signal transmission to suppress signal jitter caused by electromagnetic interference.

[0103] The zone exposure control module 22 receives the synchronous trigger signal and controls the microarray shutter CMOS to perform independent zone exposure;

[0104] Specifically, the zone exposure control module 22 receives the synchronization signal from the trigger drive module to achieve independent exposure control for each zone. The effective imaging area of ​​the microarray shutter CMOS is divided into 16 uniformly distributed 4×4 sub-regions, with no overlap between sub-regions and full coverage of the imaging range. All sub-regions are exposed synchronously, and the exposure duration is dynamically adapted (10μs~100μs) according to the velocity parameters transmitted by the particle motion state sensing unit 1: when the particle velocity is high, a short exposure of 10~30μs is used to avoid motion blur, and when the velocity is low, a long exposure of 50~100μs is used to improve the signal-to-noise ratio.

[0105] Meanwhile, the switching duration of each sub-region is controlled by an independent shutter drive circuit, and the exposure sequence is precisely synchronized with the trigger signal with a timing accuracy of 1μs. It also receives CMOS shutter status feedback in real time, outputs an alarm and records the number when a fault is detected, and enables the backup strategy of adjacent sub-regions to ensure the continuity of acquisition.

[0106] Image acquisition module 23 acquires particle images by performing a microarray shutter CMOS with regional exposure;

[0107] Specifically, the image acquisition module 23 is based on an industrial-grade microarray shutter CMOS to acquire particle images. It uses a CMOS with a resolution of 1920×1080, a maximum frame rate of 200fps, and a pixel depth of 8 bits, and has built-in anti-halo technology to suppress strong light interference.

[0108] Simultaneously, the CMOS acquisition trigger is linked to the exposure end signal, and image readout is initiated immediately after exposure (readout rate 1Gbps). A single frame image is read out within 1ms, matching a 1kHz acquisition frequency. During the readout of each frame, a timestamp generated by the FPGA is synchronously recorded. This timestamp is based on the same reference clock as the timestamp of the particle motion state sensing unit, with an error ≤1μs, achieving precise data correlation. The acquired images are in RAW format, and the CMOS has a built-in bad pixel correction function. The image acquisition module 23 performs preliminary noise reduction using a 3×3 window mean filter via the FPGA, while also supporting automatic gain control (gain range 1~8x), dynamically adjusting according to changes in illumination to stabilize image contrast and sharpness.

[0109] The image caching module 24 caches the acquired multi-frame particle images.

[0110] Specifically, the image caching module 24 uses industrial-grade DDR3 memory chips as the caching medium, with a 4GB cache capacity and supports a 1600Mbps transmission rate to meet real-time storage requirements. It adopts a triplet storage structure of "timestamp-image data-status identifier" (status identifier includes "unprocessed", "processing", and "transmitted"), and stores the data continuously in the order of acquisition time. An index table is established to support fast retrieval by time or motion status.

[0111] Meanwhile, the cache management adopts a strategy that combines cyclic overwriting and priority protection: when the cache capacity reaches 90%, the earliest "transmitted" or "processed" frame is overwritten; images marked as "key frames" (frames with sudden changes in particle motion state) are given priority protection to avoid being overwritten.

[0112] In addition, the image caching module 24 communicates with the particle motion-morphology coupling reconstruction unit 3 through the PCIe 2.0 interface at a transmission rate of 5Gbps. It has a built-in FIFO buffer to buffer the difference in read and write speeds. The CRC16 check algorithm is used to ensure transmission integrity, and retransmission is triggered if the check fails.

[0113] The particle motion-morphology coupling reconstruction unit 3 uses an embedded processor to integrate the acceleration data output by the IMU to obtain the relative velocity of the particles, and combines it with multi-frame images to reconstruct the particle motion path; it determines the motion stability by the velocity change rate of the particles in consecutive frames; it establishes a motion-morphology coupling mapping model, using the fuzzy particle shape, relative velocity, motion direction, and motion stability determination results as inputs to generate realistic shape features; for the motion posture of non-spherical particles, it performs directional rotation compensation based on the motion direction; it runs a drift-morphology coupling correction algorithm, which synchronously eliminates motion drift and shape distortion through bidirectional feedback adjustment of motion data and shape data, and outputs realistic three-dimensional shape data of the particles.

[0114] Specifically, the particle motion-morphology coupling reconstruction unit 3 adopts a high-performance embedded processor. Through sequential communication and linkage of the motion parameter calculation module 31, motion state determination module 32, morphology mapping modeling module 33, attitude compensation module 34, coupling correction module 35, and morphology data output module 36, it realizes deep coupling between particle motion data and morphology data. The core improvement lies in establishing a two-way feedback correction mechanism for motion-morphology, an adaptive compensation strategy for non-spherical particle attitude, and a coupling mapping model with dynamic weight adaptation, which simultaneously eliminates motion drift and morphology distortion, and accurately outputs the three-dimensional real morphology data of particles.

[0115] The particle motion-morphology coupling reconstruction unit 3 includes a motion parameter calculation module 31, a motion state determination module 32, a morphology mapping modeling module 33, an attitude compensation module 34, a coupling correction module 35, and a morphology data output module 36, which are connected in sequence via communication.

[0116] In this embodiment, the motion parameter calculation module 31 integrates the acceleration data output by the IMU to obtain the relative velocity of the particles, and reconstructs the particle motion path by combining multiple frames of particle images. The process of integrating the acceleration data to obtain the relative velocity of the particles and reconstructing the motion path in the motion parameter calculation module 31 includes the following steps:

[0117] S31.1 Obtain the continuous acceleration data output by the IMU and record the acceleration value at each moment;

[0118] Specifically, the system receives continuous acceleration data output from particle motion state sensing unit 1. This data has undergone Kalman filtering for noise reduction and zero-bias calibration, and the data format is "timestamp-three-axis acceleration value" (timestamp accuracy 1μs, acceleration value unit m / s²). The data is cached in the embedded processor's on-chip RAM according to the timestamp sequence, with a cache depth of 2000 frames to ensure the continuity of subsequent integration operations. At the same time, the cached data is time-series checked, and data frames with timestamp intervals exceeding ±0.1ms are removed to avoid timing misalignment affecting calculation accuracy.

[0119] It is understandable that if the timestamp interval deviation of 3 or more consecutive frames exceeds ±0.1ms, the data verification is deemed to have failed. At this time, the following operations are performed: discard the current abnormal frame data; send a resampling command to the particle motion state sensing unit 1 to trigger the particle motion-morphology coupling reconstruction unit 3 to re-acquire 1 frame of data; if the data is still abnormal after resampling, record the abnormal log and switch to the backup IMU channel (this embodiment is configured with 2 redundant IMUs).

[0120] S31.2. The relative velocity of the particles is obtained by calculating the acceleration data using the trapezoidal integral method. ;

[0121] Specifically, based on cached continuous acceleration data, the trapezoidal integral method is used to calculate the relative velocity of particles, with an integration time step of [missing information]. The formula is consistent with the acceleration data acquisition frequency (default 1ms):

[0122] ;

[0123] in: , , The first Frame particles in The relative velocity of the shaft, The first frame relative velocity of the shaft, , , For the first Frame triaxial acceleration values, , , For the first Frame triaxial acceleration values, The integration time step is 1 ms; the initial velocity is... (Applicable to all three axes).

[0124] After the calculation is completed, the relative velocities of the three axes are combined into a velocity vector. This is the relative velocity output of the particle in the current frame.

[0125] S31.3 Extract contour feature points from multi-frame particle images, and obtain the pixel displacement of particles between adjacent frames through an inter-frame feature point matching algorithm;

[0126] Specifically, the operations for contour feature point extraction and inter-frame matching are as follows:

[0127] Image preprocessing: First, perform grayscale conversion on the multi-frame particle images (using a weighted average method). Then, a 5×5 Gaussian filter kernel is used for noise reduction to suppress the interference of image noise on feature point extraction;

[0128] Contour feature point extraction: The Canny edge detection algorithm is used to extract particle contours. The high threshold is set to 200 and the low threshold is set to 100 to retain edge points with clear contours. Edge points are filtered by the density of 8 neighboring points (points with a density of less than 3 per pixel are considered outliers). After removing outliers, the effective contour feature point set is obtained.

[0129] Inter-frame feature point matching: The SIFT feature matching algorithm is used to match the contour feature points of two adjacent frames. The Euclidean distance between the feature points is calculated (with a distance threshold of 50). Matching pairs with an Euclidean distance less than the threshold are selected to obtain the pixel displacement of the particles between adjacent frames. (i.e., the pixel coordinate difference of the matching pair).

[0130] S31.4. Combining the calibration relationship between the image and the actual space, the pixel displacement is converted into actual displacement to help reconstruct the continuous motion path of the particles, making... It is consistent with the temporal changes of the motion path.

[0131] Specifically, the pixel displacement conversion and particle motion path restoration operations are as follows:

[0132] Calibration relationship establishment: Image calibration is performed beforehand using a standard calibration board with an accuracy of 0.01 mm. Multiple calibration images are acquired by placing the calibration board at different positions in the imaging area. The pixel equivalent is determined by the ratio of the known size of the calibration board to the pixel size in the image. (Unit: mm / pixel);

[0133] Actual displacement transformation: shifting pixels in adjacent frames Convert to actual displacement The formula is: ;

[0134] Motion path reconstruction: based on actual displacement in consecutive frames The continuous motion path of the particles is obtained through linear fitting. ;

[0135] Timing calibration: Comparison of motion paths With relative velocity The temporal variation trend of the two is analyzed, and the correlation coefficient is calculated. If the correlation coefficient is lower than 0.95 (this threshold is a common high correlation standard in the field of dynamic data association; a correlation coefficient above 0.95 can ensure strong consistency between the trends of motion path and relative velocity, and is a conventional threshold to ensure the reliability of coupling motion parameters and morphological data), the integral coefficient is fine-tuned (adjustment range ±5%, which is within the mild optimization range of parameter calibration, correcting temporal deviations without excessively changing the rationality of the original acceleration data, conforming to the conventional operation of parameter fine-tuning in engineering practice), until the correlation coefficient is ≥0.95, ensuring... Consistent with the temporal changes in the motion path.

[0136] In this embodiment, the motion state determination module 32 determines motion stability based on the velocity change rate of particles in consecutive frames; the motion state determination module 32 determines motion stability through the following steps:

[0137] S32.1 Obtain the relative particle velocities of two consecutive frames output by the motion parameter calculation module 31. Record the time interval between the acquisition of two frames;

[0138] Specifically, the steps to obtain the relative velocity and acquisition time interval between two consecutive frames are as follows:

[0139] Obtain the relative velocity vectors of particles from two consecutive frames containing three-axis velocity components from the motion parameter calculation module 31. Simultaneously extract the timestamps of two frames of data , (Accuracy 1 μs); Calculation time interval ,like If the deviation from the preset acquisition interval (1ms) exceeds ±0.1ms (this threshold is based on a 10% tolerance setting of the acquisition interval, which ensures the acquisition accuracy of the time interval without excessively discarding valid data due to overly stringent standards, and is a common tolerance ratio in time series data acquisition timing verification), the set of data is discarded and the next set of continuous frame data is acquired again to ensure the accuracy of the time interval.

[0140] S32.2, based on two frames The rate of change of velocity was calculated using the time interval. ;

[0141] Specifically, the quantitative index of the rate of change of speed is calculated. The specific steps are as follows:

[0142] First, calculate the magnitude of the relative velocity between the two frames. The formula is as follows:

[0143] ;

[0144] ;

[0145] The rate of change of velocity is then calculated using the following formula. :

[0146]

[0147] in: A quantitative indicator for the rate of change of velocity (unit: m / s³). The relative velocity magnitude of the previous frame. The relative velocity magnitude for the next frame. The time interval between two frame acquisitions.

[0148] S32.3, according to Determine the motion state within the numerical range: when When it is in the low range, it is judged as a change in motion; when When the motion is in the high range, it is determined to be stable, and the determination result is synchronously transmitted to the morphology mapping modeling module 33.

[0149] Specifically, the operations for determining the particle's motion state and transmitting the results are as follows:

[0150] Threshold determination: Based on the normal motion characteristics of cloud particles in an airborne environment, a preset threshold is established. (Based on the airborne equipment environmental requirements of the "Normal Category Aircraft Airworthiness Regulations" (CCAR-23-R4), the cloud particle acceleration change rate in civil airborne scenarios is generally ≤0.3m / s³; combined with the 0.05mg / √Hz acceleration noise characteristics of EveryChinaIMU 6-1A, a 0.2m / s³ error margin is reserved, which can cover more than 99% of civil airborne cloud particle scenarios and ensure reliable status judgment.) This threshold can be dynamically adjusted according to the actual application scenario (such as different flight altitudes and airflow conditions).

[0151] State determination: when When the motion is variable (the particle's motion state fluctuates, and its shape is easily distorted due to sudden changes in motion); when When the motion is stable (the particle motion is stable and the texture details can be accurately extracted);

[0152] Result transmission: Follow the sequence "Data frame header - Decision result - " The data is organized in the format of "value-timestamp-check mark". The check mark is generated by XORing the preceding data to ensure transmission integrity and is synchronously transmitted to the morphological mapping modeling module 33.

[0153] In this embodiment, the morphology mapping modeling module 33 establishes a motion-morphology coupled mapping model, using the fuzzy particle shape, relative velocity, motion direction, and motion stability determination results as inputs to generate realistic morphological features; the morphology mapping modeling module 33 includes a feature extraction submodule, a weight configuration submodule, and a model calculation submodule that are sequentially connected in communication, wherein:

[0154] The feature extraction submodule extracts features from fuzzy particle morphology. Separate and extract edge features and texture features;

[0155] Specifically, fuzzy particle morphology The method of obtaining it is:

[0156] The raw particle image acquired by particle motion state sensing unit 1 is received, and the following preprocessing operations are performed on the raw image in sequence:

[0157] Median filtering (3×3 window size) is used to remove image noise;

[0158] Particle regions are extracted using threshold segmentation (adaptive Otsu thresholding) to eliminate background interference;

[0159] Normalize the particle region to grayscale (map to the 0-255 range), and the resulting particle region image is the blurred particle shape. .

[0160] Specifically, from the perspective of fuzzy particle morphology The specific steps for extracting edge and texture features are as follows:

[0161] Edge feature extraction: The Sobel operator is used to detect edges, with horizontal convolution kernels. Vertical convolution kernel Calculate the gradient value for each pixel. Pixels with gradient values ​​greater than 50 are identified as edge points, and these are integrated to obtain the edge feature vector. (Including edge point coordinates and gradient intensity information);

[0162] Texture feature extraction: Gray-level co-occurrence matrix (GLCM) is used to extract texture, with a window size of 5×5 and pixel distance set. ,angle Calculate the four quantization parameters of the matrix at each angle: contrast, energy, entropy, and correlation. Take the average of the four angle parameters to obtain the texture feature vector. ;

[0163] Feature normalization: Perform the following formula on... and Normalization of parameters in each dimension:

[0164] ;

[0165] in These are the original parameters. For the minimum value of the parameter, The maximum value of the parameter;

[0166] The normalized parameter range is [0,1], ensuring consistent dimensionality in the fusion operation.

[0167] The weight configuration submodule outputs the values ​​from the motion state determination module 32. Adjust the fusion weight of morphological features and motion parameters, increase the weight ratio of texture features when the motion is stable, and increase the weight ratio of edge features when the motion is variable speed.

[0168] Specifically, according to The specific steps for adjusting the fusion weights are as follows:

[0169] Set the sum of fusion weights to satisfy ;in For edge feature weights, For texture feature weights, The total weight of the motion parameters;

[0170] When the motion is stable ( ): Particle shape has no obvious motion blur, configuration This weighting is based on the general machine vision morphology fusion strategy that "in low motion blur scenes, the recognizability and information content of texture features are higher than those of edge features," which is in line with conventional practices in the field of image feature extraction.

[0171] When the speed changes during motion ( Particles are prone to motion blur, configuration This configuration emphasizes the contribution of edge features. It is because motion blur reduces the effectiveness of texture features, while edge contours are more morphologically distinctive. This is a typical weight setting method for feature fusion in high motion blur scenarios in machine vision.

[0172] Motion parameter breakdown: medium relative velocity weight Direction of movement weight This weight allocation is based on the motion-morphology coupling relationship that "relative velocity has a greater impact on the degree of morphological ambiguity than motion direction", which is consistent with the conventional logic of ranking the contribution of motion parameters in particle dynamic morphology reconstruction. Depend on Direction cosine representation: .

[0173] The model computation submodule uses extracted morphological features and particle relative velocities. , direction of movement and Using this as input, perform fusion operations to generate realistic morphological features. .

[0174] Specifically, performing fusion operations generates real-world morphological features. The specific steps are as follows:

[0175] With normalized and As input, the calculation is performed using a linear fusion model, and the formula is:

[0176] ;

[0177] in: The true morphological feature vector (including particle outline coordinates, texture quantization parameters, morphological size, etc.);

[0178] By first Take the modulus value to unify the dimensions, and then combine it with... The motion parameter fusion values ​​are obtained by weighted summation of the edge features, texture features, and motion parameter fusion values ​​respectively; finally, the edge features, texture features, and motion parameter fusion values ​​are linearly superimposed according to the configured weights to output the true morphological features. .

[0179] In this embodiment, the attitude compensation module 34 performs directional rotation compensation based on the motion direction for the motion attitude of non-spherical particles; the attitude compensation module 34 includes a particle type identification submodule, a shape axis extraction submodule, and a rotation correction submodule that are sequentially connected in communication, wherein:

[0180] The particle type recognition submodule is based on real morphological features The differences in the contour features distinguish between two types of non-spherical particles: ice crystals and graupel.

[0181] Specifically, the steps to distinguish between ice crystals and graupel are as follows: Define two core contour quantification metrics, based on... Contour information calculation:

[0182] Aspect Ratio Determine the length of the longest axis of the profile using the minimum bounding rectangle algorithm. with the shortest axis length , ;

[0183] Circularity Calculate the area of ​​the outline With perimeter , (The closer the roundness is to 1, the closer the particle is to a sphere).

[0184] Judgment rules: (Based on observations of typical aspect ratios of ice crystals in atmospheric physics, ice crystals are mostly irregular in shape, such as columnar or hexagonal, and their aspect ratio is usually not less than 2.5.) (Based on the morphological characteristics of ice crystals, their outlines are highly irregular, and their roundness is usually below 0.6, which is a common statistical range in the field of atmospheric observation) when they are identified as ice crystals; (Based on the morphological characteristics of graupel, graupel consists of semi-melted ice particles with a relatively compact shape. Its length-to-width ratio is mostly in the range of 1.2-2.5, which is typical for meteorological observation.) (Based on the morphological characteristics of graupel, its outline is more compact than that of ice crystals, and its roundness is usually between 0.6 and 0.8, which is a conventional statistical result in the field of atmospheric physics.) When it is determined to be graupel.

[0185] The morphology axis extraction submodule targets different types of non-spherical particles, from The core morphological axis is extracted as the calibration benchmark.

[0186] Specifically, the specific steps for extracting the core morphological axis of non-spherical particles are as follows:

[0187] Ice crystals: Principal component analysis (PCA) algorithm was used to analyze the ice crystals. The contour coordinates are used to calculate the covariance matrix of the cloud computing, and the eigenvalues ​​and eigenvectors of the matrix are solved. The eigenvector corresponding to the largest eigenvalue is taken as the core morphological axis. ;

[0188] Scattershot: First, calculate the centroid of the contour using the contour coordinates. (coordinate (Obtained by averaging the coordinates of all contour points), then fit the minimum circumsphere of the contour, taking the center of the sphere. (coordinate ) and center of gravity Connecting vectors As the core form axis .

[0189] The rotation correction submodule performs directional rotation correction on the core shape axis based on the particle's motion direction, thereby improving the adaptability of the particle's shape and motion posture.

[0190] Specifically, the specific operation for performing directional rotation correction on the core shape axis is as follows:

[0191] Angle calculation: Calculate the core shape axis through vector dot product. and direction of motion The included angle The formula is:

[0192] ;

[0193] in The magnitude of the core morphological axis vector. ( (for unit vectors)

[0194] Rotation matrix construction: around a perpendicular to Plane construction rotation matrix ,when Perform rotational correction at the time ( (Considered as posture adaptation, no correction required)

[0195] Oriented rotation: All contour coordinate points according to Rotation transformation, making the core shape axis (Core Form Axis) The longest principal axis of the particle shape is used to perform PCA algorithm. Principal component analysis is performed on the contour coordinate points, and the direction of the first principal component is taken as... ) and direction of motion Parallel alignment, complete pose compensation, and output the compensated morphological features. .

[0196] It should be added that the rotation matrix is ​​constructed based on the open-source matrix computation library Eigen 3.4 or later; the rotation angle... ( For SIFT feature points perpendicular to (Inter-frame displacement in the plane), rotation axis for The perpendicular vector projected onto the XY plane ( hour, (Used after normalization).

[0197] In this embodiment, the coupling correction module 35 runs a drift-morphology coupling correction algorithm to synchronously eliminate motion drift and morphological distortion through bidirectional feedback adjustment of motion data and morphological data. The bidirectional feedback adjustment process in the coupling correction module 35 includes the following steps:

[0198] S35.1, The relative velocity of the particles output by the motion parameter calculation module 31. Based on this, the true morphological characteristics were corrected. The distorted parts were used to obtain the morphological features after preliminary correction. ;

[0199] Specifically, correcting the true form characteristics The specific operation of the distortion part is as follows: using the output of motion parameter calculation module 31 Based on this, (Right now To perform distortion correction, the formula is:

[0200] ;

[0201] in: These are the morphological features after preliminary correction. These are the morphological features after attitude compensation. Speed ​​correction factor (ice crystals) ,sleet (Based on the reasonable setting of the morphological characteristics of the two particles). The relative velocity magnitude, for and The difference vector;

[0202] Correction logic: Through Quantify the degree of distortion, combined with The morphological differences reflected, according to The distortion is compensated in reverse to restore the true form.

[0203] It should be added that, in order to accurately assess the cumulative error of particle motion parameter calculation and ensure the accuracy of morphological coupling reconstruction, motion drift in coupling correction module 35 is addressed. The quantization calculation adopts the deviation measurement method between "inertial integral solution speed" and "image visual reconstruction speed", as follows:

[0204] motion drift The core quantization formula is the L2 norm (modulus) difference of the three-dimensional velocity vector, expressed as follows:

[0205] ;

[0206] in: Modulo operations on vectors are used to convert three-dimensional velocity vectors into scalar form, intuitively reflecting the degree of drift. The physical unit is m / s, and the larger the value, the more significant the motion drift.

[0207] The particle's three-dimensional velocity vector, output by the motion parameter calculation module, is obtained from the IMU acceleration signal via trapezoidal integration, and its magnitude is calculated using the corrected triaxial velocity components.

[0208] ;

[0209] In the formula For the first The relative velocities (initial velocities) of the frame particles along the X, Y, and Z axes. This directly corresponds to the calculation result of the trapezoidal integral formula, ensuring the consistency between the sign and the value;

[0210] This represents the actual motion velocity vector obtained based on cloud particle image feature matching:

[0211] Feature point extraction and matching: The SIFT algorithm is used to extract stable feature points of the same particle in two adjacent frames (using its scale and rotation invariance to ensure robustness). Initial matching pairs are selected by Euclidean distance, and then erroneous matches are eliminated by RANSAC algorithm to ensure matching accuracy.

[0212] Pixel displacement conversion: Combining the core parameters of the imager (focal length f=8mm, object distance L=50m), the pixel coordinate difference of the feature points is calculated. Converted to actual spatial displacement The conversion formula is: ;

[0213] 3D velocity synthesis: combining frame intervals The two-dimensional velocity components are calculated from the actual displacement, and then the Z-axis velocity components are completed using attitude data output from the IMU. Finally, a three-dimensional velocity vector is synthesized and its magnitude is calculated. .

[0214] Furthermore, motion drift The quantization results directly guide the execution of the coupling correction strategy, and the specific rules are as follows:

[0215] when When the motion drift is determined to be within an acceptable range, it is directly adopted. Perform morphological coupling to reduce redundant calculations;

[0216] when At that time, weighted fusion correction is initiated, using the formula... Optimize the velocity vector to balance real-time performance and accuracy in the solution;

[0217] when When this occurs, the IMU zero bias recalibration process is triggered to clear accumulated errors and recalculate parameters to ensure the long-term stability of the system.

[0218] S35.2, with With fuzzy particle morphology Based on the characteristic differences, the integral benchmark of the motion parameter solution module 31 is optimized in reverse;

[0219] Specifically, the steps for optimizing the motion parameter calculation integral benchmark are as follows:

[0220] by and Based on the characteristic differences, the calibration benchmark for acceleration data is adjusted using the following formula:

[0221] ;

[0222] in: For the optimized first Frame acceleration data, For the original number Frame acceleration data, This is the feedback adjustment coefficient (default 0.1, which can be dynamically adjusted according to the actual application scenario). for and Euclidean distance (quantifying the absolute degree of morphological distortion).

[0223] Optimization logic: The more severe the shape distortion, the greater the error in the motion parameters obtained through integration. Adjust the acceleration data to optimize the benchmark for subsequent integral calculations and reduce motion drift.

[0224] S35.3 Repeat steps S35.1 to S35.2 until the amplitude of motion drift and shape distortion changes narrows synchronously, and output the final three-dimensional real shape data of the particles.

[0225] Specifically, the steps for iterating repeatedly until convergence and then outputting the result are as follows:

[0226] Iterative execution: Feedback is sent to motion parameter calculation module 31 for recalculation. And through motion state determination, shape mapping modeling, and posture compensation, a new... ;

[0227] Convergence criterion: Calculate the change in motion drift between two consecutive iterations. and morphological distortion change ,in Quantified by the deviation between relative velocity and actual displacement (here) Motion drift defined in coupling correction module 35 (Using the same parameter); preset convergence threshold. (Referring to the cloud particle morphology observation standards of the International Satellite Cloud Climate Program (ISCCP), the minimum resolution of the airborne particle imager is 0.02 pixels / μm; the imaging equipment used in this embodiment has a pixel accuracy of 0.01μm.) Setting it to 0.01 ensures that the morphological reconstruction results match the imaging accuracy, avoiding wasted computing power due to excessive iteration. and Stop iterating when the time comes;

[0228] Output results: After iterative convergence, the output includes the particle's three-dimensional true morphological data, which includes the three-dimensional contour coordinates in the inertial coordinate system, core morphological axis parameters, morphological size, and texture quantization index.

[0229] It should be added that if the number of iterations exceeds 20, or Still greater than If the iteration fails, it is determined that the iteration has not converged. In this case, the following operations are performed: adjust the dynamic weights. The adjustment step size is increased to twice the original step size; the morphological fusion feature vector is recalculated. and motion drift Then perform the iteration again; if convergence is still not achieved after adjusting the step size, fix the weights to the default values ​​for stable motion scenarios. Output the current iteration result and record the convergence exception log.

[0230] In this embodiment, the morphology data output module 36 outputs the three-dimensional real morphology data of the particles.

[0231] Specifically, the steps for outputting the true 3D shape data of particles are as follows:

[0232] Data processing: Data is processed according to the structure of "collection timestamp - particle type (spherical / ice crystal / graupel) - 3D contour coordinate set - morphological parameter set - check code". The 3D contour coordinate set is the set of all contour points. Coordinates (unit: mm), the morphological parameter set includes quantitative indicators such as core morphological axis length, aspect ratio, roundness, and posture compensation angle;

[0233] Data format: stored in binary format, coordinate data is represented by 32-bit floating-point numbers, morphological parameters are represented by 16-bit integers or 32-bit floating-point numbers depending on the precision requirements, and the check code is calculated based on all preceding data fields using the CRC32 algorithm;

[0234] Data output: Data is transmitted to the morphological data integration output unit 4 via the LVDS high-speed serial interface. The transmission rate is set to 1Mbps (matching the front-end acquisition frequency) to ensure real-time performance. At the same time, a data ready signal is output to inform the subsequent units that data can be received, ensuring transmission coordination.

[0235] The morphological data integration and output unit 4 stores the reconstructed particle morphological data locally and adapts to the remote transmission protocol to output the morphological analysis results to the outside world.

[0236] In this embodiment, the morphological data integration and output unit 4 includes a data verification module 41, a local storage module 42, a transmission protocol adaptation module 43, and an output control module 44, which are connected in sequence via communication.

[0237] The data verification module 41 receives the three-dimensional real shape data of the particles, verifies the integrity and format of the data, ensures the integrity of the data through the CRC32 verification algorithm, removes abnormal data and triggers retransmission;

[0238] Specifically, the operation for verifying the three-dimensional true morphological data of particles is as follows: Receive binary data from particle motion-morphology coupling reconstruction unit 3, and verify the validity of field length, data type, and particle type identifier according to the structure of "acquisition timestamp-particle type-three-dimensional contour coordinate set-morphological parameter set-check code", discarding data with abnormal format; for compliant data, recalculate the check code using the CRC32 algorithm and compare it with the data's built-in check code to ensure data integrity; for abnormal data, send a retransmission request containing a timestamp to particle motion-morphology coupling reconstruction unit 3, with a maximum of 3 retransmissions. If the data is still abnormal, log it and discard it to avoid consuming system resources.

[0239] The local storage module 42 uses industrial-grade storage media to classify and store the verified data according to "collection time-particle type", establishes an index for easy retrieval, and cyclically overwrites and prioritizes the retention of the latest data when the storage is full;

[0240] Specifically, the operation of storing the verified data is as follows:

[0241] Industrial-grade eMMC storage media is used, and data is stored in a two-level directory categorized as "collection date (YYYYMMDD) - particle type". The data file is named "timestamp_particle type_ID.bin". An index table containing file path, collection time, particle type, and morphological parameters is also established, supporting retrieval by time, type, and parameter range with a response time ≤100ms. When the storage capacity reaches 90%, the oldest non-critical data is overwritten in a "first-in, first-out" manner, critical data is backed up to a dedicated protected area, and the latest data is retained first.

[0242] The transmission protocol adaptation module 43 dynamically adapts the communication protocol according to the transmission scenario. When transmitting to airborne internal equipment, it adapts to the CAN bus protocol, and when transmitting to ground stations remotely, it adapts to the airborne Ethernet protocol. It encapsulates the morphological data into protocol frames and adds an identity field.

[0243] Specifically, the steps for adapting to the transmission protocol and encapsulating data are as follows:

[0244] The transmission scenario is determined by system configuration or external commands. If it is an internal airborne transmission, the CAN bus (500kbps baud rate, using standard data frame format) is adapted. If it is a ground-based remote transmission, the airborne Ethernet TCP / IP protocol stack (100Mbps rate, using UDP protocol) is adapted. In the corresponding scenario, the CAN frame is encapsulated as "identity identifier - fragment number - fragment data - check bit", and the Ethernet frame adds an identity identifier field to the header and is encapsulated as fragments according to MTU, including fragment number and total number.

[0245] The output control module 44 coordinates the priorities of local storage and remote transmission. When there is a real-time transmission requirement, it prioritizes ensuring data output bandwidth. When only offline analysis is required, it controls the data to only be stored locally and outputs data processing status indicators synchronously.

[0246] Specifically, the operations for coordinating storage and transmission are as follows:

[0247] Priority is determined by the received instructions. When a real-time transmission instruction is received from the ground station, it is determined to be of high priority, and when there is no real-time requirement, it is determined to be of low priority. In high-priority scenarios, transmission bandwidth is guaranteed first, while in low-priority scenarios, only local storage is performed. At the same time, an electrical signal status identifier (high level for normal, low level for abnormal) is output to the system, and a status field is added to the index table. When data is abnormal, it automatically switches to local storage and updates the status.

[0248] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0249] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A particle morphology analysis system based on a high-resolution airborne cloud particle imager, characterized in that, include: The particle motion state sensing unit (1) adopts a high-precision IMU. The high-precision IMU and the airborne cloud particle imager are fixed on the mounting carrier. The original data of acceleration and angular velocity of the airborne cloud particle imager moving with the mounting carrier are collected in real time. After Kalman filtering noise reduction and quaternion method attitude angle calculation, the acceleration and angular velocity parameters of the particles in the airborne environment are analyzed and the real-time motion state of the particles is output. The real-time motion state of the particles is a fusion parameter set of the absolute acceleration, absolute velocity, motion direction and carrier attitude angle of the particles in the inertial coordinate system. Multi-frame imaging trigger acquisition unit (2) adopts micro-array shutter CMOS and FPGA, and the FPGA synchronously triggers CMOS to expose the region independently, acquires and buffers multiple frames of particle images; The particle motion-morphology coupling reconstruction unit (3) uses an embedded processor to integrate the particle acceleration data relative to the mounting carrier output by the particle motion state perception unit (1) using the trapezoidal integral method to obtain the relative velocity of the particle relative to the mounting carrier. It then combines multiple frames of images to reconstruct the particle motion path. The motion stability is determined by the velocity change rate of the particle in consecutive frames. A motion-morphology coupling mapping model is established, using the blurred particle shape obtained from the preprocessed original particle image, the relative velocity of the particle relative to the mounting carrier, the motion direction of the particle relative to the mounting carrier, and the motion stability determination result as inputs to generate real morphological features. The real morphological features are generated by extracting edge and texture features from the blurred particle shape and normalizing them, and then combining them with the particle motion parameters through dynamic weight linear fusion. The feature vector that reflects the actual shape of the particle includes particle contour coordinates, texture quantization parameters, and core information about shape size. For non-spherical particles, the core shape axis is extracted and rotated and corrected according to the particle's motion direction relative to the mounting carrier to align the core shape axis with the motion direction. A drift-shape coupling correction algorithm is run to synchronously eliminate motion drift and shape distortion through bidirectional feedback adjustment of motion data and shape data, and outputs three-dimensional true shape data of the particle. The motion data is the relative velocity of the particle relative to the mounting carrier, the shape data is the particle shape feature vector and contour coordinate information corresponding to the fuzzy particle shape and the true shape features, and the true shape data is the particle shape data after iterative convergence, which includes the particle's three-dimensional contour coordinates in the inertial coordinate system, core shape axis parameters, shape size, and texture quantization index. The morphological data integration and output unit (4) stores the reconstructed particle morphological data locally and adapts to the remote transmission protocol to output the morphological analysis results to the outside world.

2. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 1, characterized in that, The particle motion state sensing unit (1) includes a data acquisition module (11), a data preprocessing module (12), a motion state calculation module (13), and a data output module (14), wherein: The data acquisition module (11) establishes a communication connection with the high-precision IMU to acquire the original acceleration and angular velocity data of the airborne cloud particle imager as it moves with the mounting carrier; The data preprocessing module (12) performs outlier removal and noise suppression on the collected raw data, and uses the Kalman filter algorithm for noise suppression; The motion state calculation module (13) integrates the acceleration data of the pre-processed airborne cloud particle imager moving with the installation carrier to obtain the relevant displacement data of the installation carrier. Combined with the pre-processed triaxial angular velocity data of the airborne cloud particle imager moving with the installation carrier, the attitude angle of the installation carrier is calculated by the quaternion method. Based on the relative motion relationship between the installation carrier motion and the particle motion, the particle motion parameters are analyzed. The particle motion parameters are the absolute acceleration, absolute velocity and motion direction of the particle in the inertial coordinate system. The real-time motion state of the particle is generated by fusion. The data output module (14) transmits the real-time motion state of the particles to the particle motion-morphology coupling reconstruction unit (3).

3. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 2, characterized in that, The multi-frame imaging trigger acquisition unit (2) includes a trigger driving module (21), a zone exposure control module (22), an image acquisition module (23), and an image buffer module (24), wherein: The trigger drive module (21) generates a synchronous trigger signal based on the FPGA; The partitioned exposure control module (22) receives the synchronous trigger signal and controls the microarray shutter CMOS to perform independent exposure of different regions; The image acquisition module (23) acquires particle images by performing a micro-array shutter CMOS with regional exposure; The image caching module (24) caches the acquired multi-frame particle images.

4. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 3, characterized in that, The particle motion-morphology coupling reconstruction unit (3) includes a motion parameter calculation module (31), a motion state determination module (32), a morphology mapping modeling module (33), a posture compensation module (34), a coupling correction module (35), and a morphology data output module (36), which are connected in sequence. The motion parameter calculation module (31) integrates the acceleration data output by the IMU to obtain the relative velocity of the particles, and combines it with multiple frames of particle images to reconstruct the particle motion path; The motion state determination module (32) determines motion stability based on the velocity change rate of particles in consecutive frames; The morphology mapping modeling module (33) establishes a motion-morphology coupling mapping model, taking the fuzzy particle morphology, relative velocity, motion direction and motion stability determination results as inputs to generate real morphological features; The attitude compensation module (34) performs directional rotation compensation based on the direction of motion for the motion attitude of non-spherical particles. The coupling correction module (35) runs a drift-morphology coupling correction algorithm to eliminate motion drift and morphological distortion through bidirectional feedback adjustment of motion data and morphological data. The morphology data output module (36) outputs the three-dimensional real morphology data of the particles.

5. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 4, characterized in that, In the motion parameter calculation module (31), the process of integrating the acceleration data to obtain the relative velocity of the particles and reconstructing the motion path includes the following steps: S31.1 Obtain the continuous acceleration data of the particle relative to the mounting carrier output by the particle motion state sensing unit (1), and record the acceleration value corresponding to the acceleration data at each moment; S31.

2. The acceleration data is calculated using the trapezoidal integral method to obtain the relative velocity of the cloud particles with respect to the mounting carrier. ; S31.3 Extract contour feature points from multi-frame particle images, and obtain the pixel displacement of particles between adjacent frames through an inter-frame feature point matching algorithm; S31.

4. Combining the ratio of image pixel size to actual spatial physical size established in advance through standard calibration plate, the pixel displacement is converted into actual displacement according to this ratio, which helps to reconstruct the continuous motion path of particles, so that... It is consistent with the temporal changes of the motion path.

6. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 5, characterized in that, The motion state determination module (32) includes the following steps in determining motion stability: S32.1 Obtain the relative particle velocities of two consecutive frames output by the motion parameter calculation module (31). Record the time interval between the acquisition of two frames; S32.2, based on two frames The rate of change of velocity was calculated using the time interval. ; S32.3, according to Determining the motion state by the numerical range 7. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 6, characterized in that, The morphological mapping modeling module (33) includes a feature extraction submodule, a weight configuration submodule, and a model computation submodule that are connected in sequence, wherein: The feature extraction submodule extracts from fuzzy particle morphology Separate and extract edge features and texture features; The weight configuration submodule is based on the output of the motion state determination module (32). Adjust the fusion weight of morphological features and motion parameters, increase the weight ratio of texture features when the motion is stable, and increase the weight ratio of edge features when the motion is variable speed. The model computation submodule is based on the fuzzy particle form. Extracted cloud particle edge features and texture features, particle relative velocity , direction of movement and Using this as input, perform fusion operations to generate realistic morphological features. .

8. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 7, characterized in that, The attitude compensation module (34) includes a particle type identification submodule, a shape axis extraction submodule, and a rotation correction submodule that are connected in sequence, wherein: The particle type recognition submodule is based on real morphological features. The differences in the contour features distinguish between two types of non-spherical particles: ice crystals and graupel. The morphological axis extraction submodule targets different types of non-spherical particles, from... The core morphological axis is extracted as the calibration benchmark. The rotation correction submodule performs directional rotation correction on the core shape axis based on the particle's motion direction, thereby improving the adaptability of the particle's shape and motion posture.

9. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 8, characterized in that, The process of performing bidirectional feedback adjustment in the coupling correction module (35) includes the following steps: S35.1, Particle relative velocity output by motion parameter calculation module (31) Based on this, the true morphological characteristics were corrected. The distorted parts were used to obtain the morphological features after preliminary correction. ; S35.2, with With fuzzy particle morphology Based on the characteristic differences, the integral benchmark of the motion parameter solution module (31) is optimized in reverse; S35.3 Repeat steps S35.1 to S35.2 until the amplitude of motion drift and shape distortion changes narrows synchronously, and output the final three-dimensional real shape data of the particles.

10. The particle morphology analysis system based on a high-resolution airborne cloud particle imager according to claim 9, characterized in that, The morphological data integration and output unit (4) includes a data verification module (41), a local storage module (42), a transmission protocol adaptation module (43), and an output control module (44) that are connected in sequence, wherein: The data verification module (41) receives the three-dimensional real shape data of the particles, verifies the integrity and format standardization of the data, ensures the integrity of the data through the CRC32 verification algorithm, removes abnormal data and triggers retransmission; The local storage module (42) uses industrial-grade storage media to classify and store the verified data according to "collection time-particle type", establish an index for easy retrieval, and cyclically overwrite and prioritize the retention of the latest data when the storage is full; The transmission protocol adaptation module (43) dynamically adapts the communication protocol according to the transmission scenario. When transmitting to airborne internal equipment, it adapts to the CAN bus protocol. When transmitting to ground station remotely, it adapts to the airborne Ethernet protocol. It encapsulates the morphological data into protocol frames and adds an identity field. The output control module (44) coordinates the priorities of local storage and remote transmission. When there is a real-time transmission requirement, it prioritizes ensuring the data output bandwidth. When only offline analysis is required, it controls the data to only perform local storage and outputs the data processing status indicator synchronously.

Images