A pollen real-time online monitoring and identifying device

Abstract

The application belongs to the technical field of microfluidics and spectral analysis, and discloses a pollen real-time online monitoring and identifying device. The device comprises: a sample inlet unit for continuously sucking in an air sample containing pollen particles; a sheathless gas acoustic flow focusing unit for generating a standing wave field in a microfluidic chip flow channel through an ultrasonic transducer, and for three-dimensionally focusing the pollen particles to the center axis of the flow channel under the condition of sheathless gas and sheathless liquid by acoustic radiation force to form a single-column ordered arrangement; a spectral detection unit for laser exciting and collecting fluorescence or Raman spectral signals of the pollen particles passing through one by one after focusing; and a signal processing and identifying unit internally provided with a pollen standard spectral feature library, for extracting spectral features and comparing them with standard spectral fingerprints, and for outputting pollen species identification results and counting statistical information. The application realizes real-time, consumable-free and high-precision automatic monitoring and identification of pollen through integrated integration of sheathless gas acoustic flow focusing and multi-dimensional spectral online identification.

Description

Technical Field

[0001] This invention relates to the fields of microfluidics and spectral analysis technology, and in particular to a real-time online pollen monitoring and identification device. Background Technology

[0002] Real-time online monitoring and accurate identification of pollen have significant application value in fields such as agricultural pollination tracking, crop breeding purity identification, urban allergenic pollen early warning, and ecological environment surveys. In recent years, significant progress has been made in automated detection technologies for airborne bioaerosol particles, resulting in various technical solutions such as flow cytometry detection systems based on hydrodynamic sheath flow, automated monitoring devices based on high-speed imaging and image recognition, and allergenic aerosol monitors that differentiate pollen using scattered light and fluorescence properties. These solutions have, to some extent, replaced traditional manual microscopic examination, promoting the automation of pollen monitoring.

[0003] However, existing technologies generally face a fundamental contradiction: high-precision identification and low-maintenance portability are difficult to achieve simultaneously. On the one hand, while detection systems based on sheath flow fluid dynamics can achieve single-row ordered arrangement of particles and precise optical detection, they rely on sheath gas or sheath fluid systems for fluid dynamic focusing. This results in bulky and complex devices that require continuous consumption of sheath fluid or high-pressure gas, leading to extremely high maintenance costs. Furthermore, there is a risk of pipeline icing in low-temperature outdoor environments, making them unsuitable for long-term unattended deployment in the field. On the other hand, while simple devices based on image recognition or scattered light detection have relatively compact structures, the lack of an effective physical focusing mechanism results in random particle distribution along the channel depth, often leading to off-target imaging or uneven light spot excitation. This severely restricts the signal-to-noise ratio and identification accuracy, making it particularly difficult to distinguish between morphologically similar pollen species. Therefore, there is an urgent need for a fully automated real-time monitoring solution that can completely eliminate the reliance on consumables in sheath flow systems and precisely lock particles onto the optical detection focal plane using non-contact physical means.

[0004] Therefore, this invention proposes a real-time online pollen monitoring and identification device. Summary of the Invention

[0005] This invention provides a real-time online pollen monitoring and identification device. By integrating sheathless gas acoustic flow control three-dimensional focusing and multi-dimensional spectral online identification, it completely eliminates the dependence on consumables in sheath flow systems and precisely locks pollen particles onto the optical detection focal plane, achieving high-precision, consumable-free, all-weather real-time online automatic identification and counting of pollen species with similar morphology.

[0006] This invention provides a real-time online pollen monitoring and identification device, comprising: The sample introduction unit is used to continuously inhale ambient air samples containing pollen particles to achieve real-time online sampling. The sheathless gas-acoustic fluid focusing unit includes a microfluidic chip and an ultrasonic transducer. The ultrasonic transducer is coupled to the microfluidic chip to generate an ultrasonic standing wave field in the flow channel of the microfluidic chip. The ultrasonic transducer uses acoustic radiation force to focus the randomly distributed pollen particles in the flow channel to the central axis of the flow channel in three dimensions without sheath gas or sheath fluid, forcing the pollen particles to form a single-row orderly arrangement in the optical detection area. The spectral detection unit is located in the downstream detection area of ​​the sheathless gas-acoustic fluid control focusing unit. It is used to laser excite the single-row pollen grains that pass through the detection window one by one after focusing, and to collect the fluorescence spectral signal or Raman spectral signal of the pollen grains. The signal processing and recognition unit has a built-in standard pollen spectral feature library. It is used to receive the spectral signals collected by the spectral detection unit, extract spectral features and compare them with the standard spectral fingerprints in the standard pollen spectral feature library, and output the pollen species identification results and counting statistics.

[0007] Furthermore, the sheathless gas acoustic flow control focusing unit includes the following structural and control features: The ultrasonic transducer is a piezoelectric ceramic transducer with an operating frequency that can be tuned within a first preset range to adapt to microfluidic chips with different flow channel widths, so that the ultrasonic standing wave field forms a single pressure node in the flow channel cross section. The microfluidic chip uses a flat rectangular microchannel, the width of which is matched with the operating frequency of the ultrasonic transducer to form a half-wavelength standing wave field. The material is optical glass or medical-grade polydimethylsiloxane to ensure that the ultrasonic standing wave field forms a single pressure node within the cross-section of the channel.

[0008] Furthermore, the sheathless air-acoustic flow control focusing unit is equipped with a power adaptive adjustment module, the working logic of which is as follows: The aerodynamic particle size distribution of pollen particles entering the flow channel is measured in real time using a front-mounted optical scattering particle size sensor. Based on the measured current mainstream particle size range, the optimal resonant frequency is automatically selected within the second preset frequency band, and the driving voltage amplitude is dynamically adjusted within the corresponding power range. When the field microenvironment is dominated by pollen grains with smaller diameters, the resonant frequency is automatically increased and the sound field intensity is enhanced to maintain focusing accuracy; when the pollen grains are dominated by larger diameters, the resonant frequency is automatically decreased and the power is appropriately reduced to reduce shear damage to the grains in the flow channel. The aforementioned closed-loop regulation ensures that pollen grains of different size ranges can be stably focused on the central axis of the flow channel, and the focusing deviation is always controlled within the preset threshold.

[0009] Furthermore, the sheathless gas acoustic flow control focusing unit is equipped with a high-concentration anti-overlap mechanism. The working logic of the high-concentration anti-overlap mechanism is as follows: When the photodetector of the spectral detection unit detects that the pulse width or frequency of the trigger signal within a unit time exceeds the preset overlap threshold, it is determined that the current pollen grain concentration has caused multiple grains to enter the detection window at the same time. The ultrasonic transducer's driving frequency is automatically switched from the current fundamental frequency mode to the frequency harmonic mode, thereby forming multiple pressure nodes in the flow channel cross section. In the normal monitoring mode under non-overlapping conditions, the ultrasonic transducer always operates in the fundamental frequency mode that forms a single pressure node in the flow channel cross section. Multiple pressure nodes disperse the pollen grain flow, which was originally concentrated on a single central axis, into multiple parallel flow channels, reducing the grain density passing through a single detection window and thus reducing the grain overlap rate.

[0010] Furthermore, the sample introduction unit includes a miniature brushless air pump, a dust pre-filter membrane, and a flow stabilizer; The miniature brushless air pump continuously draws in ambient air at a preset flow rate, and the operating noise does not exceed a preset decibel threshold. A dust pre-filter membrane is placed upstream of the sample inlet to intercept large particles larger than the pollen particle size range to prevent clogging of the microfluidic chip channels; The flow stabilizer is used to control airflow fluctuations within a preset range, providing stable airflow conditions for the sheathless air-acoustic flow control focusing unit and preventing flow velocity fluctuations from causing instability of the standing wave field and focusing deviation.

[0011] Furthermore, the spectral detection unit includes a laser source, a collimating and focusing lens group, a photodetector, and a spectral acquisition module; The excitation wavelength of the laser source is selected in the ultraviolet to near-infrared range according to the detection mode and the spectral response characteristics of the target pollen species. It supports fluorescence spectroscopy detection mode and Raman spectroscopy detection mode. The two modes can be switched according to the characteristic spectral discrimination of the pollen species to be tested. The laser light source spot size is matched with the particle distribution area after focusing, ensuring that all pollen particles passing in a single row fall into the excitation range of the light spot. The spectral acquisition module covers a third preset range of acquisition bands, with spectral resolution not exceeding a preset threshold and single-particle spectral acquisition time not exceeding a preset millisecond threshold, in order to meet the needs of high-throughput real-time online detection.

[0012] Furthermore, the pollen standard spectral feature library adopts the following standardized construction and updating method: Under controlled experimental conditions, repeated spectral acquisitions were performed on pure pollen samples of known species identified by plant taxonomy. The number of repeated acquisitions for a single pollen species was not less than the minimum number of samples required for the preset statistical reliability. Each set of spectral data is labeled with the gold standard true value, and the labeling information includes pollen species, collection time, excitation wavelength and detection mode; Multidimensional feature parameters were extracted from the labeled spectral data and organized into a database. The extracted feature parameters include: the characteristic peak positions, relative intensities, and full width at half maximum (FWHM) of fluorescence spectra, as well as the characteristic shifts and peak intensity ratios of Raman spectra. The data in the database is indexed at multiple levels according to pollen type and collection time, supporting millisecond-level fast retrieval; The pollen standard spectral feature library stores pollen species ranging from allergenic pollen to crop pollen. Allergenic pollen includes white poplar, willow, wormwood, ragweed, cypress and sycamore, while crop pollen includes wheat, corn, rice, rapeseed, cotton and soybean. The pollen standard spectral feature library supports remote incremental updates. The standard spectral fingerprints of newly added pollen species can be remotely pushed and loaded into the library without interrupting the device's online monitoring tasks.

[0013] Furthermore, the signal processing and recognition unit employs a two-layer fusion recognition strategy for spectral matching and comparison: The first layer uses a spectral angle mapping algorithm for coarse classification. It calculates the angle between the spectral vector of the pollen to be tested and the standard spectral vector in the standard spectral feature library of pollen in high-dimensional space one by one. Based on the size of the angle, it quickly filters out the candidate species set and excludes species that are obviously mismatched. The second layer uses Euclidean distance, support vector machine, or convolutional neural network algorithms to perform fine matching within the candidate species set, calculates the fine similarity between the spectrum of the pollen to be tested and the standard spectrum of each candidate species, and outputs the final pollen species identification result.

[0014] Furthermore, the signal processing and recognition unit incorporates a pre-trained deep learning recognition model, which has the following architecture and training characteristics: A one-dimensional convolutional neural network architecture is adopted to adapt to the one-dimensional sequence characteristics of pollen spectral data, which has a lower number of parameters and faster inference speed compared to two-dimensional convolutional neural networks. The input is a normalized preprocessed pollen spectral vector, with the vector dimension corresponding to the wavelength resolution of the acquisition band; The network structure includes multiple one-dimensional convolutional layers for automatically extracting hierarchical waveform features and feature peak distribution patterns from spectral curves, as well as pooling layers for progressively reducing feature dimensions. The last layer is the Softmax classification layer, which outputs the probability distribution of pollen species belonging to each category, and takes the category corresponding to the highest probability value as the recognition result. During training, a multi-class cross-entropy loss function is used, the Adam optimizer is used for parameter optimization, and an early stopping method is employed to prevent overfitting.

[0015] Furthermore, the device also includes an output and communication unit and an automatic calibration unit; The output and communication unit includes a display screen, an audible and visual alarm, and a multi-mode wireless communication module, which is used to display pollen type, count, and concentration information in real time. When the pollen concentration exceeds the preset warning threshold, an audible and visual alarm is triggered, and the warning information and monitoring data are remotely pushed to the user terminal through the multi-mode wireless communication module. The automatic calibration unit is used to perform the power-on self-test process, which includes the following steps in sequence: detection of air pump operation status, calibration of sample injection flow rate stability, automatic frequency sweep to find the optimal resonant frequency of the ultrasonic transducer under the current environmental conditions, automatic adjustment of focusing power to make the particle focusing deviation reach the preset standard, detection of optical path connectivity and spot position, wavelength calibration and intensity calibration of the spectral system, and integrity verification and loading of the pollen standard spectral feature library.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: This invention overcomes the fundamental deficiency of existing pollen monitoring technologies, which, while achieving high-precision identification, cannot escape dependence on sheath gas / sheath fluid systems. It innovatively integrates sheathless gas-free acoustic flow control three-dimensional focusing technology with multi-dimensional spectral online identification technology into a single device, replacing the complex and cumbersome fluid dynamics focusing system in traditional flow cytometry. Utilizing the acoustic radiation force generated by the ultrasonic standing wave field, randomly distributed pollen particles in the air are precisely focused onto the central axis of the flow channel and arranged in a single, orderly column without the need for any sheath gas or sheath fluid. This ensures that each pollen grain is excited by laser at the optimal position on the optical detection focal plane, eliminating spectral signal fluctuations caused by particle positional jitter at the source, and significantly improving the signal-to-noise ratio and consistency. Simultaneously, by collecting the fluorescence or Raman spectral fingerprints of single pollen grains and performing a two-layer fusion identification comparison with a standardized pollen standard spectral feature library, it can accurately distinguish between morphologically similar pollen species, fundamentally solving the misjudgment problem caused by insufficient single morphological identification dimensions. The device completely eliminates sheath fluid pipelines, high-pressure gas cylinders, and waste liquid treatment units, achieving zero-consumable operation and eliminating the risk of pipeline icing in low-temperature outdoor environments. It provides a truly long-term deployable real-time online pollen monitoring solution for unattended scenarios such as farmland, weather stations, and urban communities.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a core architecture diagram of the pollen real-time online monitoring and identification device in this embodiment of the invention; Figure 2 This is a flowchart illustrating the power adaptive adjustment and high-concentration anti-overlap mechanism in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] refer to Figure 1 and Figure 2 This invention provides an embodiment of a real-time online pollen monitoring and identification device, comprising: The sample introduction unit is used to continuously inhale ambient air samples containing pollen particles to achieve real-time online sampling. The sheathless gas-acoustic fluid focusing unit includes a microfluidic chip and an ultrasonic transducer. The ultrasonic transducer is coupled to the microfluidic chip to generate an ultrasonic standing wave field in the flow channel of the microfluidic chip. The ultrasonic transducer uses acoustic radiation force to focus the randomly distributed pollen particles in the flow channel to the central axis of the flow channel in three dimensions without sheath gas or sheath fluid, forcing the pollen particles to form a single-row orderly arrangement in the optical detection area. The spectral detection unit is located in the downstream detection area of ​​the sheathless gas-acoustic fluid control focusing unit. It is used to laser excite the single-row pollen grains that pass through the detection window one by one after focusing, and to collect the fluorescence spectral signal or Raman spectral signal of the pollen grains. The signal processing and recognition unit has a built-in standard pollen spectral feature library. It is used to receive the spectral signals collected by the spectral detection unit, extract spectral features and compare them with the standard spectral fingerprints in the standard pollen spectral feature library, and output the pollen species identification results and counting statistics.

[0022] In this embodiment, the sample introduction unit continuously draws in ambient air samples containing pollen particles to achieve real-time online sampling. Specifically, it includes the following structure and operating process: a miniature brushless air pump is used as the pumping power source to continuously draw in ambient air at a preset flow rate, with operating noise not exceeding a preset decibel threshold; a dust pre-filter membrane is installed upstream of the sample inlet to intercept large particles larger than the pollen particle size range to prevent clogging of the microfluidic chip's flow channels; after the intake airflow is processed by a flow stabilizer, the flow rate fluctuation is controlled within a preset range, providing stable intake conditions for the downstream sheathless gas-acoustic flow control focusing unit and preventing flow rate fluctuations from causing standing wave field instability and focusing deviation. The sample introduction flow rate can be adjusted from 1 mL / min to 20 mL / min according to the actual pollen concentration and detection requirements to adapt to different pollen emission intensities in different seasons and regions.

[0023] In this embodiment, the sheathless air-acoustic fluid control focusing unit includes a microfluidic chip and an ultrasonic transducer, with the ultrasonic transducer coupled to the microfluidic chip. The microfluidic chip employs a flat rectangular microchannel, made of optical glass (such as borosilicate glass or quartz glass) or medical-grade polydimethylsiloxane. The channel width is matched to the operating frequency of the ultrasonic transducer to form a half-wavelength standing wave field, ensuring that the ultrasonic standing wave field forms a single pressure node within the channel cross-section. The ultrasonic transducer is a piezoelectric ceramic transducer, and its operating frequency is configured to match the channel width of the microfluidic chip to form a half-wavelength standing wave field and a single pressure node within the channel cross-section. For pollen particles of different size ranges, microfluidic chips with different channel widths can be selected, and the operating frequency of the ultrasonic transducer is tuned to the resonant frequency required for the corresponding channel width, thereby adapting to the focusing requirements of various pollen particle sizes within the frequency range of 1MHz to 5MHz. Frequency tuning is achieved by adjusting the output frequency of the signal generator within a preset frequency range, and the impedance change of the ultrasonic transducer is monitored in real time during the tuning process to lock the optimal resonant point. When a piezoelectric ceramic transducer is subjected to an AC signal of a specific frequency and voltage, it generates high-frequency mechanical vibration, forming an ultrasonic standing wave field within the microfluidic chip's flow channel. Based on the principle of acoustic radiation force, pollen grains suspended in the air are subjected to acoustic radiation force directed towards the pressure node. The magnitude of this force is proportional to the grain volume, sound field energy density, and acoustic contrast factor. Because the acoustic contrast factor of the pollen grains is positive, the grains are rapidly pushed towards the pressure node on the central axis of the flow channel, achieving three-dimensional focusing without the need for sheath gas or sheath fluid, ultimately forming a single, orderly row in the optical detection area. The focusing deviation is controlled within a preset threshold, ensuring that each pollen grain passes through the detection window at the optimal position on the optical detection focal plane.

[0024] In this embodiment, the injection flow rate and the cross-sectional area of ​​the microfluidic chip channel are co-designed to ensure that the Reynolds number of the airflow within the channel is below 2000, resulting in a stable laminar flow state. Under laminar flow conditions, the radial pulsation velocity of the airflow is extremely small, and the acoustic radiation force generated by the ultrasonic standing wave field can stably overcome air viscous resistance, capturing and confining pollen particles at the pressure node on the central axis of the channel, ensuring the reliability and stability of three-dimensional focusing.

[0025] In this embodiment, the spectral detection unit is located downstream of the sheathless gas-acoustic flow-controlled focusing unit and includes a laser source, a collimating and focusing lens group, a photodetector, and a spectral acquisition module. The excitation wavelength of the laser source is selected in the ultraviolet to near-infrared range according to the detection mode and the spectral response characteristics of the target pollen species. It supports both fluorescence spectral detection and Raman spectral detection modes, which can be switched according to the characteristic spectral discrimination of the pollen species being tested. The spot size of the laser source matches the distribution area of ​​the focused particles, ensuring that all pollen particles passing through in a single row fall within the excitation range of the spot. The spectral acquisition module covers a wavelength range of 200nm to 1100nm, with a spectral resolution not exceeding a preset threshold and a single-particle spectral acquisition time not exceeding a preset millisecond threshold, to meet the requirements of high-throughput real-time online detection. When the focused single-row pollen particles pass through the optical detection window one by one, the photodiode detects a change in light intensity exceeding a preset signal threshold and immediately triggers the spectrometer shutter to acquire the spectral signal.

[0026] In this embodiment, the signal processing and recognition unit incorporates a standard pollen spectral feature library. The library is constructed as follows: under controlled experimental conditions, pure pollen samples of known species identified by plant taxonomy are subjected to repeated spectral acquisitions multiple times. The number of repeated acquisitions for a single pollen species is no less than the minimum sample size required for a preset statistical confidence level. Each set of spectral data is labeled with the gold standard true value, and the labeling information includes pollen species, acquisition time, excitation wavelength, and detection mode. Multidimensional feature parameters are extracted from the labeled spectral data and stored in the library. The extracted feature parameters include the characteristic peak position, relative intensity, and half-width at half-maximum (WHM) of the fluorescence spectrum, as well as the characteristic shift and peak intensity ratio of the Raman spectrum. The stored data is indexed at multiple levels according to pollen species and acquisition time, supporting millisecond-level fast retrieval. The pollen species stored in the standard pollen spectral feature library cover allergenic pollen and crop pollen. Allergenic pollen includes Populus tomentosa, Salix matsudana, Artemisia, Rapeseed, Cypress, and Sycamore. Crop pollen includes wheat, corn, rice, rapeseed, cotton, and soybean. After the signal processing and recognition unit receives the spectral signal collected by the spectral detection unit, it first performs normalization preprocessing on the spectral data to eliminate the intensity difference caused by the fluctuation of the light source power, then extracts the spectral features and compares them with the standard spectral fingerprint in the pollen standard spectral feature library, and finally outputs the pollen species identification results and counting statistics.

[0027] Furthermore, the sheathless gas acoustic flow control focusing unit includes the following structural and control features: The ultrasonic transducer is a piezoelectric ceramic transducer with an operating frequency that can be tuned within a first preset range to adapt to microfluidic chips with different flow channel widths, so that the ultrasonic standing wave field forms a single pressure node in the flow channel cross section. The microfluidic chip uses a flat rectangular microchannel, the width of which is matched with the operating frequency of the ultrasonic transducer to form a half-wavelength standing wave field. The material is optical glass or medical-grade polydimethylsiloxane to ensure that the ultrasonic standing wave field forms a single pressure node within the cross-section of the channel.

[0028] In this embodiment, the sheathless air-acoustic fluid control focusing unit includes the following structure and control features. The ultrasonic transducer is a piezoelectric ceramic transducer, made of PZT-4 or PZT-8 type lead zirconate titanate piezoelectric ceramic. Its operating frequency is continuously adjustable within the range of 1MHz to 5MHz to adapt to the standing wave field conditions required by pollen particles of different sizes. The piezoelectric ceramic transducer is bonded to the flow channel sidewall of the microfluidic chip using epoxy conductive adhesive. When the signal generator outputs an AC sinusoidal signal with a specific frequency and peak-to-peak value, it is amplified by a power amplifier and applied to the piezoelectric ceramic transducer. The transducer generates high-frequency mechanical vibration and couples into the flow channel to form ultrasonic waves. The ultrasonic waves reflect back and forth between the opposite walls of the flow channel, superimposing to form an ultrasonic standing wave field. The microfluidic chip uses a flat rectangular microchannel, the width of which matches the operating frequency of the ultrasonic transducer to form a half-wavelength standing wave field. The microchannel is made of optical glass or medical-grade polydimethylsiloxane. When selecting optical glass, borosilicate glass or quartz glass can be used. Quartz glass has high transmittance in the 200nm to 1100nm wavelength range, which is beneficial for the transmission of fluorescence excitation light in the ultraviolet band and the acquisition of broadband spectral signals. When medical-grade polydimethylsiloxane is selected, its good light transmittance and biocompatibility can meet the requirements for pollen spectral detection in the visible to near-infrared band. Since the channel width is designed to be an integer multiple of half the wavelength of ultrasound in air at the current operating frequency, the ultrasonic standing wave field forms only a single pressure node in the channel cross-section. This pressure node is located exactly at the central axis of the channel. Pollen particles are pushed towards and stably constrained at this node under the action of acoustic radiation force, thus ensuring that the ultrasonic standing wave field forms a single pressure node in the channel cross-section. All particles, regardless of where they enter the channel from the inlet, will eventually converge at the same focusing position on the central axis of the channel.

[0029] Furthermore, the sheathless air-acoustic flow control focusing unit is equipped with a power adaptive adjustment module, the working logic of which is as follows: The aerodynamic particle size distribution of pollen particles entering the flow channel is measured in real time using a front-mounted optical scattering particle size sensor. Based on the measured current mainstream particle size range, the optimal resonant frequency is automatically selected within the second preset frequency band, and the driving voltage amplitude is dynamically adjusted within the corresponding power range. When the field microenvironment is dominated by pollen grains with smaller diameters, the resonant frequency is automatically increased and the sound field intensity is enhanced to maintain focusing accuracy; when the pollen grains are dominated by larger diameters, the resonant frequency is automatically decreased and the power is appropriately reduced to reduce shear damage to the grains in the flow channel. The aforementioned closed-loop regulation ensures that pollen grains of different sizes can be stably focused on the central axis of the flow channel, and the focusing deviation is always controlled within the preset threshold.

[0030] In this embodiment, the sheathless air-acoustic fluid control focusing unit is equipped with a power adaptive adjustment module. The working logic of the power adaptive adjustment module is as follows: a front-mounted optical scattering particle size sensor is integrated on the flow channel section between the microfluidic chip inlet and the acoustic field action zone. This sensor uses the laser scattering principle to measure the aerodynamic particle size distribution of pollen particles entering the flow channel in real time, and feeds back the measurement results to the embedded processor in the form of a particle size distribution histogram. Based on the measured current mainstream particle size range, the embedded processor automatically selects the optimal resonant frequency from the factory-preset frequency points and dynamically adjusts the driving voltage amplitude within the corresponding power range. When replacing the microfluidic chip with one of different flow channel widths to adapt to different particle size ranges, the power adaptive adjustment module re-executes automatic frequency sweeping to find the optimal resonant frequency corresponding to the current chip. Specifically, the control strategy is as follows: when the field microenvironment is dominated by smaller pollen particles, the resonant frequency is automatically increased and the acoustic field intensity is enhanced to maintain focusing accuracy; when the field microenvironment is dominated by larger pollen particles, the resonant frequency is automatically decreased and the power is appropriately reduced to minimize shear damage to the particles within the flow channel. The aforementioned closed-loop regulation ensures that pollen grains of different sizes can be stably focused on the central axis of the flow channel, and the focusing deviation is always controlled within the preset threshold.

[0031] Furthermore, the sheathless gas acoustic flow control focusing unit is equipped with a high-concentration anti-overlap mechanism. The working logic of the high-concentration anti-overlap mechanism is as follows: When the photodetector of the spectral detection unit detects that the pulse width or frequency of the trigger signal within a unit time exceeds the preset overlap threshold, it is determined that the current pollen grain concentration has caused multiple grains to enter the detection window at the same time. The ultrasonic transducer's driving frequency is automatically switched from the current fundamental frequency mode to the frequency harmonic mode, thereby forming multiple pressure nodes in the flow channel cross section. In the normal monitoring mode under non-overlapping conditions, the ultrasonic transducer always operates in the fundamental frequency mode that forms a single pressure node in the flow channel cross section. Multiple pressure nodes disperse the pollen grain flow, which was originally concentrated on a single central axis, into multiple parallel flow channels, reducing the grain density passing through a single detection window and thus reducing the grain overlap rate.

[0032] In this embodiment, the sheathless gas-acoustic flow control focusing unit is equipped with a high-concentration anti-overlap mechanism. The working logic of this high-concentration anti-overlap mechanism is as follows: when the photodetector of the spectral detection unit detects that the pulse width or frequency of the trigger signal within a unit time exceeds a preset overlap threshold, it determines that the current pollen grain concentration has caused multiple grains to simultaneously enter the detection window. At this time, the embedded processor automatically switches the driving frequency of the ultrasonic transducer from the current fundamental frequency mode to a frequency doubling mode, for example, from 2.0MHz to 4.0MHz, forming two pressure nodes within the same flow channel cross-section instead of the original single pressure node. Multiple pressure nodes disperse the pollen grain flow originally concentrated on a single central axis to two parallel flow channel positions, each flow channel position carrying a portion of the grain flow. The grain density passing through a single detection window is thus reduced, thereby reducing the grain overlap rate. When the pollen concentration returns to a normal level, the system automatically switches the driving frequency back to the fundamental frequency mode, restoring the focusing state of a single pressure node.

[0033] Furthermore, the sample introduction unit includes a miniature brushless air pump, a dust pre-filter membrane, and a flow stabilizer; The miniature brushless air pump continuously draws in ambient air at a preset flow rate, and the operating noise does not exceed a preset decibel threshold. A dust pre-filter membrane is placed upstream of the sample inlet to intercept large particles larger than the pollen particle size range to prevent clogging of the microfluidic chip channels; The flow stabilizer is used to control airflow fluctuations within a preset range, providing stable airflow conditions for the sheathless air-acoustic flow control focusing unit and preventing flow velocity fluctuations from causing instability of the standing wave field and focusing deviation.

[0034] In this embodiment, the sample introduction unit includes a miniature brushless air pump, a dust pre-filter membrane, and a flow stabilizer. The miniature brushless air pump is a diaphragm pump driven by a brushless DC motor, continuously drawing in ambient air at a preset flow rate. Its operating noise does not exceed a preset decibel threshold, and it requires no carbon brush replacement during long-term continuous operation, making it suitable for unattended outdoor scenarios. The dust pre-filter membrane is located upstream of the sample inlet, using a stainless steel filter screen with a 200-mesh pore size. It intercepts large particles larger than pollen grains, such as dust, willow catkins, and insect remains, preventing blockage of the microfluidic chip's flow channels. The flow stabilizer is located in the air path between the dust pre-filter membrane and the microfluidic chip's sample introduction interface. It employs a high-precision regulating valve and a buffer chamber structure to control airflow fluctuations within a preset range, providing stable airflow conditions for the sheathless air-acoustic flow control focusing unit and preventing flow velocity fluctuations from causing standing wave field instability and focusing deviation.

[0035] Furthermore, the spectral detection unit includes a laser source, a collimating and focusing lens group, a photodetector, and a spectral acquisition module; The excitation wavelength of the laser source is selected in the ultraviolet to near-infrared range according to the detection mode and the spectral response characteristics of the target pollen species. It supports fluorescence spectroscopy detection mode and Raman spectroscopy detection mode. The two modes can be switched according to the characteristic spectral discrimination of the pollen species to be tested. The laser light source spot size is matched with the particle distribution area after focusing, ensuring that all pollen particles passing in a single row fall into the excitation range of the light spot. The spectral acquisition module covers a third preset range of acquisition bands, with spectral resolution not exceeding a preset threshold and single-particle spectral acquisition time not exceeding a preset millisecond threshold, in order to meet the needs of high-throughput real-time online detection.

[0036] In this embodiment, the spectral detection unit includes a laser source, a collimating and focusing lens group, a photodetector, and a spectral acquisition module. The excitation wavelength of the laser source is selected in the ultraviolet to near-infrared range according to the detection mode and the spectral response characteristics of the target pollen species. When using the fluorescence spectral detection mode, a 405nm semiconductor laser can be selected to detect endogenous fluorescent substances such as proteins and chlorophyll in pollen grains using characteristic excitation peaks. When using the Raman spectral detection mode, a 532nm or 785nm laser can be selected to reduce fluorescence background interference and obtain a high signal-to-noise ratio Raman fingerprint signal. The two detection modes can be switched according to the characteristic spectral distinguishability of the pollen species to be tested. For pollen species of the same genus that are similar in morphology but different in chemical composition, the Raman spectral detection mode is preferred to obtain richer molecular vibrational fingerprint information. For pollen species of different genera with significant fluorescence differences, the fluorescence spectral detection mode is selected to obtain a faster detection speed. The laser spot size is adjusted by a collimating and focusing lens group to match the distribution area of ​​the focused pollen grains, ensuring that all pollen grains passing through in a single row fall within the excitation range of the spot. This avoids missed detection due to an excessively small spot size or increased background noise due to an excessively large spot size. The spectral acquisition module covers a wavelength range of 200nm to 1100nm, with a spectral resolution not exceeding 1.5nm and a single-particle spectral acquisition time not exceeding 10ms, to meet the requirements of high-throughput real-time online detection. As each focused pollen grain passes through the optical detection window, the photodetector detects the change in light intensity caused by the grains blocking or scattering the laser beam. When the light intensity change exceeds a preset signal threshold, the spectral acquisition module is immediately triggered to acquire the spectral signal, achieving precise synchronous trigger detection.

[0037] Furthermore, the pollen standard spectral feature library adopts the following standardized construction and updating method: Under controlled experimental conditions, repeated spectral acquisitions were performed on pure pollen samples of known species identified by plant taxonomy. The number of repeated acquisitions for a single pollen species was not less than the minimum number of samples required for the preset statistical reliability. Each set of spectral data is labeled with the gold standard true value, and the labeling information includes pollen species, collection time, excitation wavelength and detection mode; Multidimensional feature parameters were extracted from the labeled spectral data and organized into a database. The extracted feature parameters include: the characteristic peak positions, relative intensities, and full width at half maximum (FWHM) of fluorescence spectra, as well as the characteristic shifts and peak intensity ratios of Raman spectra. The data in the database is indexed at multiple levels according to pollen type and collection time, supporting millisecond-level fast retrieval; The pollen standard spectral feature library stores pollen species ranging from allergenic pollen to crop pollen. Allergenic pollen includes white poplar, willow, wormwood, ragweed, cypress and sycamore, while crop pollen includes wheat, corn, rice, rapeseed, cotton and soybean. The pollen standard spectral feature library supports remote incremental updates. The standard spectral fingerprints of newly added pollen species can be remotely pushed and loaded into the library without interrupting the device's online monitoring tasks.

[0038] In this embodiment, the pollen standard spectral feature library adopts the following standardized construction and updating method. Under controlled experimental conditions, repeated spectral acquisitions are performed on pure pollen samples of known species identified by plant taxonomy. A calibrated fluorescence spectrometer or Raman spectrometer is used to continuously acquire different single grains of the same pollen species under constant temperature, humidity, and excitation power conditions. The number of repeated acquisitions for a single pollen species is no less than the minimum sample size required for the preset statistical reliability. Each set of spectral data is labeled with the gold standard true value. The labeling information includes pollen species, acquisition time, excitation wavelength, and detection mode. The gold standard true value is confirmed by plant taxonomy experts through both microscopic morphological identification and molecular biological identification to ensure the accuracy and traceability of the labeling information. Multidimensional feature parameters are extracted from the labeled spectral data and organized into the library. The extracted feature parameters include the characteristic peak positions, relative intensities, and full width at half maximum (FWHM) of the characteristic peaks in the fluorescence spectrum, as well as the characteristic shifts and peak intensity ratios of different characteristic peaks in the Raman spectrum. The data in the library is indexed at multiple levels according to pollen species and acquisition time. The first-level index is for pollen species, and the second-level index is for acquisition time, supporting millisecond-level fast retrieval. The pollen standard spectral feature library stores pollen species covering two main categories: allergenic pollen and crop pollen. Allergenic pollen includes white poplar, willow, artemisia, ragweed, cypress, and sycamore, while crop pollen includes wheat, corn, rice, rapeseed, cotton, and soybean. The pollen standard spectral feature library supports remote incremental updates. The standard spectral fingerprints of newly added pollen species can be remotely pushed from the cloud server to the device's local storage via a wireless communication module. This allows for the loading and index reconstruction of new data while existing monitoring tasks are running normally, and the update process does not interrupt the device's online monitoring function.

[0039] Furthermore, the signal processing and recognition unit employs a two-layer fusion recognition strategy for spectral matching and comparison: The first layer uses a spectral angle mapping algorithm for coarse classification. It calculates the angle between the spectral vector of the pollen to be tested and the standard spectral vector in the standard spectral feature library of pollen in high-dimensional space one by one. Based on the size of the angle, it quickly filters out the candidate species set and excludes species that are obviously mismatched. The second layer uses Euclidean distance, support vector machine, or convolutional neural network algorithms to perform fine matching within the candidate species set, calculates the fine similarity between the spectrum of the pollen to be tested and the standard spectrum of each candidate species, and outputs the final pollen species identification result.

[0040] In this embodiment, the signal processing and recognition unit employs a two-layer fusion recognition strategy for spectral matching and comparison. The first layer uses a spectral angle mapping algorithm for coarse classification. The algorithm calculates the angle between the normalized preprocessed pollen spectral vector and the standard spectral vectors of each known species stored in the pollen standard spectral feature library in high-dimensional space. The spectral angle mapping algorithm treats two spectral curves as two vectors in high-dimensional space, measuring the similarity of spectral morphology by calculating the angle between the two vectors. The smaller the angle, the closer the waveforms of the two spectral curves are, and the greater the likelihood that they belong to the same species. When the angle is less than a preset coarse classification threshold, the pollen species corresponding to that standard spectrum is included in the candidate species set; when the angle is greater than the preset coarse classification threshold, the species is directly excluded and does not proceed to the subsequent fine matching stage. Through this rapid screening, most obviously mismatched species in the pollen standard spectral feature library can be excluded, retaining only a small number of candidate species for the second layer of processing. The second layer performs fine matching within the candidate species set, employing one or a combination of Euclidean distance, support vector machine (SVM), or convolutional neural network (CNN) algorithms to calculate the fine similarity between the spectrum of the pollen to be tested and the standard spectra of each candidate species. The Euclidean distance algorithm directly calculates the sum of the absolute deviations of the reflectance values ​​at each wavelength position between the two spectral vectors; a smaller distance indicates higher similarity. The SVM algorithm classifies candidate species using a pre-trained classification hyperplane. The CNN algorithm automatically extracts hierarchical waveform features and characteristic peak distribution patterns from the spectral curves using a pre-trained one-dimensional CNN model, with the probability distribution output by the Softmax classification layer. These three fine matching algorithms can be flexibly selected or combined based on the number of candidate species, computational resources, and accuracy requirements, ultimately outputting the pollen species identification result. The core advantage of the dual-layer fusion identification strategy lies in limiting the computationally intensive fine matching algorithm to a small range after coarse classification, balancing identification speed and accuracy.

[0041] Furthermore, the signal processing and recognition unit incorporates a pre-trained deep learning recognition model, which has the following architecture and training characteristics: A one-dimensional convolutional neural network architecture is adopted to adapt to the one-dimensional sequence characteristics of pollen spectral data, which has a lower number of parameters and faster inference speed compared to two-dimensional convolutional neural networks. The input is a normalized preprocessed pollen spectral vector, with the vector dimension corresponding to the wavelength resolution of the acquisition band; The network structure includes multiple one-dimensional convolutional layers for automatically extracting hierarchical waveform features and feature peak distribution patterns from spectral curves, as well as pooling layers for progressively reducing feature dimensions. The last layer is the Softmax classification layer, which outputs the probability distribution of pollen species belonging to each category, and takes the category corresponding to the highest probability value as the recognition result. During training, a multi-class cross-entropy loss function is used, the Adam optimizer is used for parameter optimization, and an early stopping method is employed to prevent overfitting.

[0042] In this embodiment, the signal processing and recognition unit incorporates a pre-trained deep learning recognition model. The deep learning recognition model employs a one-dimensional convolutional neural network architecture to adapt to the one-dimensional sequence characteristics of pollen spectral data, offering lower parameter counts and faster inference speeds compared to two-dimensional convolutional neural networks. The model's input is a normalized preprocessed pollen spectral vector, with the vector dimension corresponding to the wavelength resolution of the acquisition band; each wavelength position corresponds to an element in the vector. The network structure includes multiple one-dimensional convolutional layers for automatically extracting hierarchical waveform features and feature peak distribution patterns from the spectral curve. Each one-dimensional convolutional layer uses a set of learnable one-dimensional convolutional kernels to perform sliding convolution operations along the wavelength direction of the spectral vector, extracting local waveform patterns. Pooling layers follow the convolutional layers to progressively reduce the feature dimension. The pooling layers use max pooling to take the maximum value within a local window as the output, reducing the parameter count and computational cost of subsequent layers. The final layer is a Softmax classification layer. The Softmax classification layer maps the original scores of each category output from the previous fully connected layer to probability values ​​for each category using an exponential function. The sum of all category probabilities is 1, and the category corresponding to the highest probability value is taken as the recognition result. During training, a multi-class cross-entropy loss function is used. This loss function measures the difference between the probability distribution predicted by the model and the true labeled class distribution; the smaller the difference, the more accurate the model prediction. The Adam optimizer is used for parameter optimization. The Adam optimizer adaptively adjusts the learning rate for each parameter by calculating the first and second moment estimates of the gradient. Early stopping is employed to prevent overfitting. Early stopping continuously monitors the model's loss function value on the validation set during training. Training is terminated early when the validation set loss function value no longer decreases for several consecutive training epochs, preventing the model from overfitting on the training set and thus reducing its generalization ability. After model training is complete, the network weight parameters and structure configuration are fixed into an inference mode and deployed in the embedded processor of the signal processing and recognition unit. In inference mode, backpropagation and parameter updates are no longer performed; only forward propagation is executed. The latency of a single spectral inference operation is less than a preset millisecond threshold to meet the requirements of real-time online detection.

[0043] To achieve the aforementioned real-time inference performance, the deep learning recognition model employs a lightweight design. Specific lightweight strategies include: using a one-dimensional convolutional neural network architecture, leveraging the characteristic of its one-dimensional convolutional kernels performing sliding convolution operations along the spectral wavelength direction, resulting in a significantly lower number of model parameters compared to convolutional neural networks processing two-dimensional images; and controlling the total number of model parameters within a preset range by adjusting the number of convolutional layers and the number of kernels per layer to adapt to the storage capacity and computing power of embedded processors. When the model is deployed to an embedded processor, industrial-grade inference optimization tools are further used to perform graph optimization and operator fusion on the network structure, and by quantizing network weights and activation values ​​into a low-precision format, the computational overhead and memory usage during forward propagation are significantly reduced. Combined with these optimizations, the deep learning recognition model can achieve real-time recognition performance with a single spectral inference latency of less than 10 milliseconds on resource-constrained embedded processors.

[0044] In one specific embodiment, the one-dimensional convolutional neural network contains three one-dimensional convolutional layers with 32, 64 and 128 convolutional kernels respectively, and the kernel size is 3. Each convolutional layer is followed by a max pooling layer with a window size of 2. Through the above pooling operation, the feature dimension is gradually halved after each pooling layer. Finally, the extracted feature vector is flattened and fed into the Softmax classification layer to output the probability distribution of each pollen species.

[0045] Furthermore, the device also includes an output and communication unit and an automatic calibration unit; The output and communication unit includes a display screen, an audible and visual alarm, and a multi-mode wireless communication module, which is used to display pollen type, count, and concentration information in real time. When the pollen concentration exceeds the preset warning threshold, an audible and visual alarm is triggered, and the warning information and monitoring data are remotely pushed to the user terminal through the multi-mode wireless communication module. The automatic calibration unit is used to perform the power-on self-test process, which includes the following steps in sequence: detection of air pump operation status, calibration of sample injection flow rate stability, automatic frequency sweep to find the optimal resonant frequency of the ultrasonic transducer under the current environmental conditions, automatic adjustment of focusing power to make the particle focusing deviation reach the preset standard, detection of optical path connectivity and spot position, wavelength calibration and intensity calibration of the spectral system, and integrity verification and loading of the pollen standard spectral feature library.

[0046] In this embodiment, the device further includes an output and communication unit and an automatic calibration unit. The output and communication unit includes a display screen, an audible and visual alarm, and a multi-standard wireless communication module. The display screen refreshes in real time, showing the currently identified pollen species name, the corresponding particle count for each pollen species, and the pollen concentration value per unit volume of air. When the pollen concentration exceeds a preset warning threshold, the audible and visual alarm emits a buzzer and flashes a light, while the multi-standard wireless communication module remotely pushes warning data containing pollen species, concentration, exceedance level, monitoring timestamp, and device geographical location information to the user terminal via 4G, narrowband IoT, wireless LAN, or long-range radio communication. The user terminal can be a smartphone, computer, or cloud monitoring platform. The automatic calibration unit performs the power-on self-test process. Each time the device is powered on or receives a remote calibration command, the automatic calibration unit sequentially executes the following self-test steps: First, it checks the air pump's operating status, verifying that the current waveform and speed of the miniature brushless air pump are within the normal range. Second, it calibrates the sample injection flow rate stability, continuously monitoring the airflow through a built-in flow sensor and adjusting it accordingly to keep flow rate fluctuations within a preset range. Third, it automatically sweeps the frequency to find the optimal resonant frequency of the ultrasonic transducer under the current environmental conditions. The signal generator sweeps the frequency within a preset frequency range with preset step values, while simultaneously monitoring the impedance changes of the ultrasonic transducer. The frequency corresponding to the minimum impedance value is selected as the current optimal resonant frequency; this step compensates for sound velocity drift caused by changes in ambient temperature. Fourth, it automatically adjusts the focus. The power is adjusted by driving the ultrasonic transducer at the optimal resonant frequency determined in the previous step, gradually adjusting the driving voltage amplitude. Simultaneously, the focusing position of the standard-sized fluorescent microspheres is observed using a microscopic imaging system. Once the focusing deviation reaches the preset standard, the driving voltage is locked. The fifth step involves detecting the optical path connectivity and spot position, confirming that the laser source output is normal and the spot center coincides with the flow channel's central axis. The sixth step involves wavelength and intensity calibration of the spectral system, using a mercury-argon standard light source to calibrate the wavelength of the spectral acquisition module and a standard fluorescent plate to calibrate the response sensitivity of each wavelength. The seventh step involves verifying and loading the integrity of the pollen standard spectral feature library, ensuring the data integrity of the locally stored pollen standard spectral feature library is not damaged and that the version is synchronized with the cloud. After successful verification, the feature library is loaded into memory for subsequent identification and comparison. After all self-test steps pass, the device automatically enters real-time online monitoring mode. If any step fails, the device displays the corresponding fault code on the screen and pauses monitoring until the fault is resolved.

[0047] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A real-time online pollen monitoring and identification device, characterized in that, include: The sample introduction unit is used to continuously draw in ambient air samples containing pollen particles, enabling real-time online sampling. The sheathless gas-acoustic fluid focusing unit includes a microfluidic chip and an ultrasonic transducer. The ultrasonic transducer is coupled to the microfluidic chip to generate an ultrasonic standing wave field within the flow channel of the microfluidic chip. The operating frequency is configured to match the width of the flow channel of the microfluidic chip to form a half-wavelength standing wave field and a single pressure node within the flow channel cross-section. The unit uses acoustic radiation force to three-dimensionally focus randomly distributed pollen particles within the flow channel to the central axis of the flow channel without the need for sheath gas or sheath fluid, forcing the pollen particles to form a single-row orderly arrangement in the optical detection area. The spectral detection unit is located in the downstream detection area of ​​the sheathless gas-acoustic fluid control focusing unit. It is used to laser excite the single-row pollen grains that pass through the detection window one by one after focusing, and to collect the fluorescence spectral signal or Raman spectral signal of the pollen grains. The signal processing and recognition unit has a built-in standard pollen spectral feature library. It is used to receive the spectral signals collected by the spectral detection unit, extract spectral features and compare them with the standard spectral fingerprints in the standard pollen spectral feature library, and output the pollen species identification results and counting statistics.

2. The pollen real-time online monitoring and identification device according to claim 1, characterized in that, The sheathless gas acoustic flow control focusing unit includes the following structural and control features: The ultrasonic transducer is a piezoelectric ceramic transducer with an operating frequency that can be tuned within a first preset range to adapt to microfluidic chips with different flow channel widths, so that the ultrasonic standing wave field forms a single pressure node in the flow channel cross section. The microfluidic chip uses a flat rectangular microchannel, the width of which is matched with the operating frequency of the ultrasonic transducer to form a half-wavelength standing wave field. The material is optical glass or medical-grade polydimethylsiloxane to ensure that the ultrasonic standing wave field forms a single pressure node within the cross-section of the channel.

3. The pollen real-time online monitoring and identification device according to claim 2, characterized in that, The sheathless air-acoustic flow control focusing unit is equipped with a power adaptive adjustment module. The working logic of the power adaptive adjustment module is as follows: The aerodynamic particle size distribution of pollen particles entering the flow channel is measured in real time using a front-mounted optical scattering particle size sensor. Based on the measured current mainstream particle size range, the optimal resonant frequency is automatically selected within the second preset frequency band, and the driving voltage amplitude is dynamically adjusted within the corresponding power range. When the field microenvironment is dominated by pollen grains with smaller diameters, the resonant frequency is automatically increased and the sound field intensity is enhanced to maintain focusing accuracy; when the pollen grains are dominated by larger diameters, the resonant frequency is automatically decreased and the power is appropriately reduced to reduce shear damage to the grains in the flow channel. The aforementioned closed-loop regulation ensures that pollen grains of different size ranges can be stably focused on the central axis of the flow channel, and the focusing deviation is always controlled within the preset threshold.

4. The pollen real-time online monitoring and identification device according to claim 2, characterized in that, The sheathless gas acoustic flow control focusing unit is equipped with a high-concentration anti-overlap mechanism. The working logic of the high-concentration anti-overlap mechanism is as follows: When the photodetector of the spectral detection unit detects that the pulse width or frequency of the trigger signal within a unit time exceeds the preset overlap threshold, it is determined that the current pollen grain concentration has caused multiple grains to enter the detection window at the same time. The ultrasonic transducer's driving frequency is automatically switched from the current fundamental frequency mode to the frequency harmonic mode, thereby forming multiple pressure nodes in the flow channel cross section. In the normal monitoring mode under non-overlapping conditions, the ultrasonic transducer always operates in the fundamental frequency mode that forms a single pressure node in the flow channel cross section. Multiple pressure nodes disperse the pollen grain flow, which was originally concentrated on a single central axis, into multiple parallel flow channels, reducing the grain density passing through a single detection window and thus reducing the grain overlap rate.

5. The pollen real-time online monitoring and identification device according to claim 1, characterized in that, The sample introduction unit includes a miniature brushless air pump, a dust pre-filter membrane, and a flow stabilizer; The miniature brushless air pump continuously draws in ambient air at a preset flow rate, and the operating noise does not exceed a preset decibel threshold. A dust pre-filter membrane is placed upstream of the sample inlet to intercept large particles larger than the pollen particle size range to prevent clogging of the microfluidic chip channels; The flow stabilizer is used to control airflow fluctuations within a preset range, providing stable airflow conditions for the sheathless air-acoustic flow control focusing unit and preventing flow velocity fluctuations from causing instability of the standing wave field and focusing deviation.

6. The pollen real-time online monitoring and identification device according to claim 1, characterized in that, The spectral detection unit includes a laser source, a collimating and focusing lens group, a photodetector, and a spectral acquisition module; The excitation wavelength of the laser source is selected in the ultraviolet to near-infrared range according to the detection mode and the spectral response characteristics of the target pollen species. It supports fluorescence spectroscopy detection mode and Raman spectroscopy detection mode. The two modes can be switched according to the characteristic spectral discrimination of the pollen species to be tested. The laser light source spot size is matched with the particle distribution area after focusing, ensuring that all pollen particles passing in a single row fall into the excitation range of the light spot. The spectral acquisition module covers a third preset range of acquisition bands, with spectral resolution not exceeding a preset threshold and single-particle spectral acquisition time not exceeding a preset millisecond threshold, in order to meet the needs of high-throughput real-time online detection.

7. The pollen real-time online monitoring and identification device according to claim 1, characterized in that, The pollen standard spectral feature library adopts the following standardized construction and updating method: Under controlled experimental conditions, repeated spectral acquisitions were performed on pure pollen samples of known species identified by plant taxonomy. The number of repeated acquisitions for a single pollen species was not less than the minimum number of samples required for the preset statistical reliability. Each set of spectral data is labeled with the gold standard true value, and the labeling information includes pollen species, collection time, excitation wavelength and detection mode; Multidimensional feature parameters were extracted from the labeled spectral data and organized into a database. The extracted feature parameters include: the characteristic peak positions, relative intensities, and full width at half maximum (FWHM) of fluorescence spectra, as well as the characteristic shifts and peak intensity ratios of Raman spectra. The data in the database is indexed at multiple levels according to pollen type and collection time, supporting millisecond-level fast retrieval; The pollen standard spectral feature library stores pollen species ranging from allergenic pollen to crop pollen. Allergenic pollen includes white poplar, willow, wormwood, ragweed, cypress and sycamore, while crop pollen includes wheat, corn, rice, rapeseed, cotton and soybean. The pollen standard spectral feature library supports remote incremental updates. The standard spectral fingerprints of newly added pollen species can be remotely pushed and loaded into the library without interrupting the device's online monitoring tasks.

8. The pollen real-time online monitoring and identification device according to claim 1, characterized in that, The signal processing and recognition unit employs a dual-layer fusion recognition strategy for spectral matching and comparison. The first layer uses a spectral angle mapping algorithm for coarse classification. It calculates the angle between the spectral vector of the pollen to be tested and the standard spectral vector in the standard spectral feature library of pollen in high-dimensional space one by one. Based on the size of the angle, it quickly filters out the candidate species set and excludes species that are obviously mismatched. The second layer uses Euclidean distance, support vector machine, or convolutional neural network algorithms to perform fine matching within the candidate species set, calculates the fine similarity between the spectrum of the pollen to be tested and the standard spectrum of each candidate species, and outputs the final pollen species identification result.

9. The pollen real-time online monitoring and identification device according to claim 1, characterized in that, The signal processing and recognition unit has a built-in pre-trained deep learning recognition model, which has the following architecture and training characteristics: A one-dimensional convolutional neural network architecture is adopted to adapt to the one-dimensional sequence characteristics of pollen spectral data, which has a lower number of parameters and faster inference speed compared to two-dimensional convolutional neural networks. The input is a normalized preprocessed pollen spectral vector, with the vector dimension corresponding to the wavelength resolution of the acquisition band; The network structure includes multiple one-dimensional convolutional layers for automatically extracting hierarchical waveform features and feature peak distribution patterns from spectral curves, as well as pooling layers for progressively reducing feature dimensions. The last layer is the Softmax classification layer, which outputs the probability distribution of pollen species belonging to each category, and takes the category corresponding to the highest probability value as the recognition result. During training, a multi-class cross-entropy loss function is used, the Adam optimizer is used for parameter optimization, and an early stopping method is employed to prevent overfitting.

10. The pollen real-time online monitoring and identification device according to claim 1, characterized in that, The device also includes an output and communication unit and an automatic calibration unit; The output and communication unit includes a display screen, an audible and visual alarm, and a multi-mode wireless communication module, which is used to display pollen type, count, and concentration information in real time. When the pollen concentration exceeds the preset warning threshold, an audible and visual alarm is triggered, and the warning information and monitoring data are remotely pushed to the user terminal through the multi-mode wireless communication module. The automatic calibration unit is used to perform the power-on self-test process, which includes the following steps in sequence: detection of air pump operation status, calibration of sample injection flow rate stability, automatic frequency sweep to find the optimal resonant frequency of the ultrasonic transducer under the current environmental conditions, automatic adjustment of focusing power to make the particle focusing deviation reach the preset standard, detection of optical path connectivity and spot position, wavelength calibration and intensity calibration of the spectral system, and integrity verification and loading of the pollen standard spectral feature library.

Images