Biological aerosol automatic identification system and method based on ultraviolet fluorescence imaging

Abstract

The invention relates to the technical field of meteorological ecological observation, biological safety detection, environment monitoring and the like, and discloses a biological aerosol automatic identification system and method based on ultraviolet fluorescence imaging, and the system comprises an environment control module, an aerosol sampling module, an adhesive tape transmission module, an ultraviolet microscopic imaging module and a control and data processing module. The environment control module is used for constructing a constant-temperature closed optical darkroom; the aerosol sampling module captures biological particles in the air on the surface of a carrier adhesive tape by utilizing an acceleration nozzle and an inertial impact principle; the adhesive tape transmission module realizes automatic transfer of a sample from a sampling position to a detection position through stepping control; the ultraviolet microscopic imaging module adopts a coaxial incident light path structure and is matched with a dichroic spectroscope and an optical filter to separate exciting light and fluorescence signals. According to the invention, full-process automatic in-situ monitoring of the biological aerosol is realized, the problem of background interference is effectively solved, and the real-time performance of detection and the recognition accuracy are improved.

Description

Technical Field

[0001] This invention relates to the technical fields of meteorological and ecological observation, biosafety detection and environmental monitoring, specifically to an automatic bioaerosol identification system and method based on ultraviolet fluorescence imaging. Background Technology

[0002] Bioaerosols refer to particulate matter such as bacteria, fungal spores, pollen, viruses, and biological debris suspended in the air. They are not only important pathogens causing respiratory diseases and allergic reactions, but also potential carriers of bioterrorism. Therefore, real-time monitoring of the concentration and types of bioaerosols in ambient air is of great significance for public health security and environmental quality early warning.

[0003] Currently, traditional methods for detecting bioaerosols mainly rely on Anderson impactors or liquid impactor samplers for sample collection, followed by transfer to laboratories for microbial culture and counting or artificial staining and microscopic examination. However, this offline detection mode has significant time and space lag, often taking hours or even days from sampling to obtaining results. Furthermore, the process is cumbersome and labor-intensive, making it difficult to meet the urgent needs for real-time and continuous data in public health emergencies or dynamic environmental monitoring.

[0004] With the development of optoelectronic technology, real-time detection technology based on the fluorescence characteristics of biomolecules (such as NADH and riboflavin) under ultraviolet light excitation has gradually become a research hotspot. However, existing online fluorescence monitoring devices still have significant limitations in optical imaging and data processing. In terms of optical structure, existing devices mostly adopt side illumination or simple direct light paths, resulting in uneven distribution of excitation light on the carrier surface and easy shadowing effects between particles. At the same time, the carrier or filter substrate used to capture particles usually has a certain autofluorescence background. In addition, the lack of fine spectral separation methods in the optical path design makes stray light and background noise seriously interfere with the weak biofluorescence signal, resulting in a low imaging signal-to-noise ratio and difficulty in capturing clear microscopic features.

[0005] In qualitative and quantitative analysis, existing technologies largely rely on fluorescence intensity thresholds to distinguish between biological and non-biological particles. However, in complex real-world atmospheric environments, many non-biological particles (such as polycyclic aromatic hydrocarbons, paper dust, and textile fibers) also exhibit fluorescence characteristics, making false positives highly likely based solely on fluorescence intensity. Furthermore, due to a lack of effective extraction and in-depth analysis capabilities for detailed features such as particle microstructure and texture, existing systems struggle to accurately distinguish between different biological categories with similar fluorescence intensities (e.g., pollen and fungal spores), resulting in insufficient accuracy and precision in monitoring data, hindering accurate qualitative classification and quantitative inversion.

[0006] Therefore, there is an urgent need to develop an automatic bioaerosol identification system that integrates fully automated sampling and transmission, high signal-to-noise ratio precision imaging, and multi-dimensional feature depth recognition. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an automatic bioaerosol identification system and method based on ultraviolet fluorescence imaging. This solves the problems of poor real-time performance caused by the physical separation of sampling and detection in existing bioaerosol monitoring equipment, as well as low accuracy of qualitative and quantitative analysis due to large background fluorescence interference and difficulty in identifying microscopic morphology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides an automatic bioaerosol identification system based on ultraviolet fluorescence imaging.

[0009] This system integrates gas, optical, and electrical circuits through a modular design. In terms of physical architecture, the system is equipped with an environmental control module that defines the system's physical boundaries, creating a double-layered protective structure through an outer enclosure and an internal sealed optical darkroom. The sealed optical darkroom provides a light-interference-free detection environment, and together with an integrated temperature control unit, ensures the focal length stability of the optical components and the viability of the biological samples.

[0010] In the sample capture stage, the system includes an aerosol sampling module. This module utilizes aerodynamic principles, driving airflow through an acceleration nozzle via a flow control unit. The geometry of the acceleration nozzle is configured to accelerate the airflow to a preset velocity. By leveraging the mass difference between bioaerosol particles and gas molecules, inertia causes the particles to deviate from the streamline in areas where the airflow sharply changes direction, thus depositing them on the surface of the medium located below the nozzle.

[0011] To achieve spatiotemporal transfer of samples, the system is equipped with a tape transport module. The carrier tape serves as the collection medium, passing through the sampling and detection positions. Drive rollers, in conjunction with a stepper motor, convert continuous time-series sampling points into spatially discrete test points and precisely transport them to the center of the optical imaging field of view.

[0012] In the detection and imaging stage, the system employs an ultraviolet microscopy imaging module. This module constructs the microscopic imaging optical path, with the microscope objective configured to focus on the surface of the carrier tape. An ultraviolet excitation source provides excitation energy in a specific wavelength band, inducing the biological particles attached to the carrier tape surface to produce a Stokes-shifted fluorescence signal. An image sensor is then used to capture this weak fluorescence microscopic image.

[0013] The control and data processing module, as the logical core of the system, coordinates the timing of each execution unit through electrical connections and undertakes data calculation tasks. This module converts optical images into digital signals and executes specific image processing algorithms to separate the target from the background, thereby completing the qualitative and quantitative analysis of particles.

[0014] In specific implementations, the present invention further optimizes the system performance through the following technical means: Regarding the optical system construction, the ultraviolet microscopy imaging module is configured with a coaxial incident optical path structure to address the shadowing issue that may occur with side illumination. Specifically, a dichroic beam splitter is placed at the intersection of the optical path, and its spectral characteristics are configured to reflect short-wavelength ultraviolet excitation light and transmit long-wavelength visible fluorescence. The excitation light, after being reflected by the dichroic beam splitter, is incident perpendicularly onto the surface of the carrier tape through the microscope objective; conversely, the fluorescence signal excited by the particles returns along the original optical path and is transmitted through the dichroic beam splitter. A long-pass filter is further incorporated into the optical path to physically block out the reflected residual excitation light, ensuring that the image sensor receives only the biological fluorescence signal.

[0015] Regarding the stability of sampling and imaging, the tape transport module is equipped with a rigid support plate at the detection position directly below the microscope objective. This plate restricts the displacement of the tape in the Z-axis direction, ensuring that the particles on the surface of the tape are always within the depth of field of the microscope objective, thus avoiding image defocusing caused by the deformation of the flexible tape.

[0016] Regarding the data processing logic, the control and data processing module implements a background suppression strategy. Considering that the substrate of the adhesive tape may have weak autofluorescence, this module is configured to acquire images of blank areas as a background reference, and subtract background pixel values ​​from the measured fluorescence micrographs through differential operations, thereby highlighting the fluorescence characteristics of the particles.

[0017] Furthermore, the recognition algorithm employs a two-stage cascaded architecture. The first stage utilizes image segmentation techniques to extract regions of interest (ROIs) for candidate particles based on connected component features; the second stage utilizes a pre-built convolutional neural network classification model to perform feature encoding and classification only on the extracted ROIs. This processing logic reduces the computational overhead of full-image deep learning inference while ensuring recognition accuracy. The system ultimately outputs standardized concentration data based on sampling throughput, sampling duration, number of recognitions, and acquisition efficiency coefficient.

[0018] A second aspect of the present invention provides an automatic identification method for bioaerosols based on ultraviolet fluorescence imaging.

[0019] Based on the aforementioned system architecture, this method automates physical processes through timing control, and includes the following steps: During the sampling phase, a constant sampling airflow is established, and the inertial force generated by the accelerating nozzle is used to separate the bioaerosols in the air and deposit them onto the surface of the stationary carrier tape. This process is continued for a preset duration to enrich the sample.

[0020] During the transmission phase, step control logic is executed to drive the carrier tape to move along the transmission path by precise physical steps, translating the enriched sampling area to the center of the field of view of the microscopic imaging optical path.

[0021] During the imaging stage, an ultraviolet excitation source is activated, and the beam is perpendicularly irradiated onto the sample through a coaxial optical path, inducing the biological particles to produce characteristic fluorescence; simultaneously, an image sensor is triggered to acquire a microscopic image containing fluorescence intensity and morphological information.

[0022] In the processing stage, the original image is first denoised and subjected to background subtraction to eliminate interference from non-target signals. Then, the particle coordinates are located and the regions of interest are extracted through connected component analysis. Finally, these regions are input into a deep learning model for classification, and the aerosol concentration in the environment is calculated by combining the physical parameters of the sampling process.

[0023] This invention provides an automatic identification system and method for bioaerosols based on ultraviolet fluorescence imaging. It has the following beneficial effects: 1. This invention establishes an automated link from aerodynamic capture to optical imaging by physically integrating an aerosol sampling module, a tape transport module, and an ultraviolet microscopy imaging module. Compared with traditional offline laboratory culture or manual staining microscopy methods, this system uses the principle of inertial impaction to directly fix particles onto the carrier tape and automatically transports them to the detection position by a stepper motor, eliminating the time lag of sample transfer and manual processing, thereby enabling the output of continuous monitoring data of bioaerosol concentration with high temporal resolution.

[0024] 2. This invention employs a coaxial incident optical path structure, combined with the spectral selectivity of a dichroic beam splitter and a long-pass filter, to achieve physical separation of excitation light and emitted fluorescence. This ensures that the light is incident perpendicularly on the tape surface, avoiding particle occlusion or shadow effects that may occur with side illumination. Furthermore, the filter forces the cutoff of residual ultraviolet light reflected back, allowing the image sensor to receive only the biofluorescence signal. This reduces background noise caused by stray light from the tape substrate and the environment, providing high-contrast raw data for subsequent image recognition.

[0025] 3. This invention employs a strategy combining background subtraction preprocessing and convolutional neural network (CNN) classification in the data processing stage. Static background interference from the adhesive tape is eliminated through differential operations, and a deep learning model is used to jointly analyze the texture and morphological features of the extracted regions of interest. The system can effectively distinguish between non-biological dust with similar fluorescence intensity and target biological aerosols (such as pollen and spores). This dual discrimination mechanism of fluorescence and morphology solves the problem of false alarms easily caused by single fluorescence intensity detection methods, improving the accuracy of counting and concentration inversion of specific categories of biological particles. Attached Figure Description

[0026] Figure 1 This is a system overall framework diagram of the present invention; Figure 2 This is a block diagram of the electrical control connection of the system according to the present invention; Figure 3 This is a flowchart illustrating the automated operation sequence of the system according to the present invention. Figure 4 This is a block diagram of the image recognition algorithm and data processing logic of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] See attached document Figure 1 The present invention provides an automatic identification system for bioaerosols based on ultraviolet fluorescence imaging, which may include an environmental control module, an aerosol sampling module, a tape transmission module, an ultraviolet microscopy imaging module, and a control and data processing module.

[0029] The environmental control module includes an outer enclosure and a sealed optical darkroom located inside the outer enclosure. The outer enclosure is the outermost protective structure of the system, and its walls are equipped with a temperature regulation unit, which may include, for example, a thermoelectric cooler or a heater. This unit, through a temperature sensor and a control and data processing module, forms a closed-loop control system to maintain the temperature within the sealed optical darkroom within a preset stable range. The sealed optical darkroom is constructed of opaque material, and its inner wall surface is coated with a matte material, such as matte black paint, to absorb stray light entering the darkroom, providing a high signal-to-noise ratio operating environment for the ultraviolet microscopy imaging module.

[0030] The aerosol sampling module includes an inlet air duct, an accelerating nozzle, a flow control unit, and a suction pump. One end of the inlet air duct is connected to the outside atmosphere, and the other end is connected to the accelerating nozzle. The accelerating nozzle is a tubular structure whose diameter gradually narrows along the airflow direction and extends into the interior of a sealed optical dark chamber. The flow control unit, such as a mass flow controller, is located in the inlet air duct or suction line and is electrically connected to the control and data processing module to precisely adjust the operating power of the suction pump to maintain a constant sampling flow rate.

[0031] To further ensure the optical sealing of the sealed optical darkroom, a labyrinthine light trap structure is installed at the connection between the air intake channel and the outer casing, as well as at the entrance of the accelerating nozzle into the sealed optical darkroom. This labyrinthine light trap structure consists of multiple layers of staggered light-shielding baffles, the surfaces of which are also coated with an matting material. This structure is configured to allow airflow to enter the darkroom through a tortuous path, while simultaneously forcing external incident light to undergo multiple reflections and absorptions between the baffles. This increases the attenuation rate of ambient light before it reaches the sampling area to a preset threshold, preventing ambient light from interfering with fluorescence detection.

[0032] The aerosol sampling module is configured to operate based on the principle of inertial impaction. When the suction pump is activated, air containing bioaerosols is drawn into the inlet airflow channel at a set volumetric flow rate. The airflow velocity increases significantly as it passes through the acceleration nozzle, forming a high-speed jet at the nozzle exit. Whether particles can separate from the airflow and impact the substrate depends on their Stokes number. Specifically, this manifests as: ; in, Particle density; The aerodynamic diameter of the particle; This is the Cunningham slip correction factor; The average airflow velocity at the nozzle exit; Aerodynamic viscosity; The nozzle outlet diameter is [value].

[0033] The tape transport module includes a feed roller, a take-up roller, a carrier tape, a drive roller, a stepper motor, and a rigid support plate. The feed roller and take-up roller are located on opposite sides of a sealed optical darkroom for storing and retrieving the carrier tape. The carrier tape is made of a low-self-fluorescence material, with one surface coated with an optical-grade adhesive.

[0034] The transport path of the carrier tape is designed to pass directly below the accelerating nozzle and through the center of the field of view of the ultraviolet microscopy imaging module. The surface of the drive roller is in close contact with the non-adhesive surface of the carrier tape, and this drive roller is connected to a stepper motor via a transmission mechanism. The control and data processing module controls the rotation angle of the stepper motor by sending pulse signals, thereby enabling the stepping movement of the carrier tape between the sampling and detection positions.

[0035] Directly below the ultraviolet microscopy imaging module, a rigid support plate is positioned. The top surface of this rigid support plate contacts the back of the carrier tape, providing a flat and fixed support plane for the flexible carrier tape at the detection location. The height of this support plane is precisely calibrated to ensure that the adhesive surface of the carrier tape is within the depth of field of the ultraviolet microscopy imaging module, thereby preventing image defocusing due to carrier tape deformation.

[0036] In a preferred embodiment, the rigid support plate is not directly fixed to the base of the sealed optical darkroom, but is installed via a Z-axis fine-tuning mechanism. This Z-axis fine-tuning mechanism, which can employ a precision micrometer screw gauge or a piezoelectric ceramic actuator, is used to perform micrometer-level vertical adjustment of the rigid support plate's height during system factory calibration or maintenance. This fine-tuning mechanism compensates for thickness tolerances between different batches of the carrier tape and mechanical assembly errors, ensuring that the upper surface of the carrier tape is always precisely positioned on the optimal focal plane of the microscope objective.

[0037] See attached document Figure 1 and attached Figure 2 Based on the aforementioned system architecture, the ultraviolet microscopy imaging module is specifically configured to adopt a coaxial incident optical architecture, which includes an ultraviolet excitation source, a collimating lens group, a dichroic beam splitter, a microscope objective, a long-pass filter, an imaging lens, and an image sensor.

[0038] The ultraviolet excitation source, serving as the energy provider for fluorescence excitation, is configured to emit narrowband ultraviolet light with a central wavelength. In this specific configuration, the central wavelength is selected from 280nm±10nm or 385nm±25nm, which corresponds to the absorption peak of biofluorophores commonly found in bioaerosols, such as nicotinamide adenine dinucleotide (NADH) or riboflavin. The divergent beam emitted by the ultraviolet excitation source is shaped by a collimating lens group and converted into a parallel beam with uniform intensity distribution, which is then projected onto the surface of the dichroic beam splitter.

[0039] A dichroic spectrometer is positioned at the intersection of the optical axes of the ultraviolet excitation source, the microscope objective, and the image sensor, typically at an angle relative to the incident optical axis. This dichroic spectrometer possesses specific spectral transmittance and reflectance characteristics, and its cutoff wavelength is designed to lie between the excitation wavelength and the fluorescence emission peak wavelength of the target bioaerosol.

[0040] Specifically, for light with wavelengths shorter than the cutoff wavelength, the dichroic beam splitter exhibits high reflectivity; for light with wavelengths longer than the cutoff wavelength, it exhibits high transmittance. Therefore, the ultraviolet excitation light from the collimating lens group is reflected by the dichroic beam splitter, deflected by 90 degrees, and then incident perpendicularly along the optical axis of the microscope objective.

[0041] In this optical path, the microscope objective functions as both a condenser and an objective. It focuses the reflected ultraviolet excitation light onto the surface of the adhesive tape at the detection location, forming a high-energy-density excitation spot. After absorbing ultraviolet photon energy, the bioparticles attached to the adhesive tape surface undergo Stokes shift, emitting a fluorescence signal with a wavelength equal to the peak fluorescence emission wavelength, where the peak emission wavelength is greater than the center wavelength. Since fluorescence radiates in all directions, the microscope objective collects the fluorescence signal within its numerical aperture (NA) range and converts it into a parallel beam that returns along the original optical path.

[0042] The returning beam contains the fluorescence signal generated by the target particles and the residual excitation light reflected or scattered by the surface of the carrier tape. This mixed beam reaches the dichroic beam splitter again. Based on the aforementioned spectral characteristics, the shorter-wavelength residual excitation light is reflected back to the light source, while the longer-wavelength fluorescence signal is transmitted through the dichroic beam splitter and enters the subsequent imaging process.

[0043] A long-pass filter is positioned between the dichroic beam splitter and the image sensor as a second-stage spectral filtering element. Its cutoff frequency is configured to block all wavelengths below the cutoff wavelength, further filtering out residual ultraviolet noise that may pass through the dichroic beam splitter, ensuring that only the fluorescence signal can reach the imaging surface. The filtered fluorescence signal is focused by the imaging lens and finally imaged on the photosensitive surface of the image sensor.

[0044] Under this optical path configuration, the fluorescence light flux received by the image sensor Depending on the collection efficiency of the optical system and the fluorescence properties of the particles, the quantitative relationship can be expressed as: ; in, The proportionality constant of the system; The collecting solid angle of the microscope objective is determined by the numerical aperture; The intensity distribution of excitation light on the sample surface; The absorption cross-section spectrum of biological particles; The fluorescence quantum yield spectrum; This refers to the total spectral transmission efficiency of the imaging link (including filters, lenses, and sensors).

[0045] Image sensors, such as high-sensitivity CMOS or scientific-grade CCDs, convert received photon signals into digital grayscale images and transmit them to the control and data processing module. This optical design maximizes the intensity difference between the target signal and background noise at the hardware level through physical spectral separation.

[0046] See attached document Figure 3 The identification method provided by this invention mainly uses a control and data processing module as the main control unit to coordinate the execution of periodic sampling, transmission, and detection action sequences by various hardware modules.

[0047] After system initialization, the sampling phase begins. The control and data processing module sends a start command to the aerosol sampling module. Specifically, the control and data processing module establishes the sampling airflow by adjusting the drive voltage or duty cycle (PWM signal) output to the suction pump. The flow control unit monitors the instantaneous flow rate in the intake channel in real time and transmits the feedback signal to the control and data processing module, forming a closed-loop control to maintain a constant sampling volume flow rate. During this phase, the stepper motor is locked, keeping the drive roller stationary, thereby ensuring that the position of the carrier tape relative to the accelerating nozzle is fixed. The airflow containing bioaerosols is accelerated by the accelerating nozzle and then impacts the sampling area of ​​the carrier tape vertically. The particles are captured based on the aforementioned inertial impaction principle. This process continues for a preset sampling duration to ensure that a sufficient number of test particles are enriched on the surface of the carrier tape.

[0048] After the sampling period ends, the control and data processing module shuts down the vacuum pump or adjusts the flow rate to zero, and the system enters the transmission phase. The control and data processing module sends a predetermined number of drive pulses to the stepper motor. The stepper motor drives the drive roller to rotate through mechanical transmission, and relies on friction to pull the conveyor belt along the transmission direction.

[0049] The distance traveled is precisely set to the physical distance between the central axis of the accelerating nozzle and the optical axis of the ultraviolet microscopy module in the direction of the conveyor belt. The required number of pulses is calculated based on the stepper motor's step angle, reduction ratio, and the radius of the drive roller. ; in, This is the inherent step angle of the stepper motor; This is the mechanical reduction ratio between the motor shaft and the drive roller; The effective radius of the driving roller.

[0050] Through the above control, the area of ​​the carrier tape that has just been sampled is precisely delivered to the area directly below the microscope objective, i.e., the detection position.

[0051] The imaging stage then begins. The control and data processing module first sends a command to activate the ultraviolet excitation source. To avoid photobleaching and reduce heat accumulation, the ultraviolet excitation source typically operates in stroboscopic mode, illuminating only during the exposure window. Once the light source brightness is stable, the control and data processing module sends a trigger signal to the image sensor to control the opening and closing of the electronic shutter. The image sensor accumulates photon charges within the preset exposure time, converting the fluorescence signals generated by the particles attached to the surface of the carrier tape into a digitized fluorescence microscopic image.

[0052] After image acquisition, the control and data processing module shuts down the ultraviolet excitation source and enters the data processing stage. In this stage, the processor within the control and data processing module performs a series of algorithmic operations on the raw image data. First, image preprocessing is performed by performing a difference operation between a pre-stored blank tape background image and the currently acquired fluorescence microscopy image to remove background noise. Then, a connected component analysis algorithm is used to scan the image, identifying all independent regions with brightness above a threshold, marking them as candidate particles, and extracting the corresponding regions of interest (ROIs).

[0053] After the extracted regions of interest (ROIs) are normalized, they are input into a pre-defined convolutional neural network classification model. This model outputs the particle category (e.g., pollen, spores, or abiotic fluorescent dust) corresponding to each ROI. Finally, the control and data processing module counts the number of particles in various bioaerosols and, combined with the current sampling flow rate, sampling duration, and pre-calibrated system sampling efficiency, calculates the concentration indices of various biological components in the current ambient air. After the calculation is complete, the system determines the end of the current cycle and automatically resets to the sampling phase to begin the next monitoring cycle.

[0054] See attached document Figure 4 The control and data processing module is configured to execute specific algorithmic logic to convert the raw digital images acquired by the image sensor into quantitative bioaerosol concentration data. This processing flow mainly includes four stages: image preprocessing, region of interest extraction, deep learning classification, and concentration inversion.

[0055] During the image preprocessing stage, the control and data processing module aims to eliminate static background noise caused by the substrate material of the adhesive tape. The system pre-stores one or more frames of blank adhesive tape images taken in a particle-free state as reference background images. After acquiring the raw fluorescence micrograph containing particles within the detection cycle, the processor performs pixel-level differential operations.

[0056] Specifically, the gray value of the corresponding pixel in the reference background image is subtracted from the gray value of each pixel in the original image, and the result is restricted to a non-negative range to obtain the net signal image after background subtraction. This step effectively removes fixed-mode noise and autofluorescent background caused by reflections from the adhesive tape or even the optical system.

[0057] In the Region of Interest (ROI) extraction stage, the control and data processing module performs binarization segmentation on the net signal image. An adaptive thresholding algorithm (such as the Otsu's algorithm) is typically used to calculate the optimal global or local segmentation threshold, converting the grayscale image into a binary mask image. To eliminate isolated noise points caused by thermal noise from the image sensor, a morphological opening operation (erosion followed by dilation) is performed on the binary mask image to smooth boundaries and break up fine adhesions. Subsequently, a connected component labeling algorithm is used to traverse the binary image, identifying all independent pixel connected regions. For each connected region, its bounding rectangle is calculated, and a sub-image containing a single particle, i.e., the ROI, is cropped from the net signal image based on the rectangle's coordinates.

[0058] In the deep learning classification stage, a pre-trained convolutional neural network classification model is pre-installed within the control and data processing module. This model includes an input layer, multiple convolutional layers, pooling layers, and fully connected layers. First, the extracted irregularly sized regions of interest are uniformly scaled to the standard size required by the model's input layer (e.g., 64×64 pixels).

[0059] Before being deployed to the control and data processing module, the convolutional neural network classification model needs to undergo supervised training on a high-performance computing platform. The training dataset contains a large number of pre-labeled fluorescence microscopic images of various biological aerosols and non-biological particles. The training process uses the cross-entropy loss function as the objective function to measure the difference between the predicted probability distribution and the true label distribution. The gradient is calculated using the backpropagation algorithm, and the weight and bias parameters in the network are iteratively updated using the adaptive moment estimation optimizer until the model's classification accuracy on the validation set converges to a preset standard. After training, the model parameters are fixed and ported to the embedded memory of the control and data processing module.

[0060] Subsequently, the standardized image matrix is ​​input into a convolutional neural network. The network extracts texture, edge, and high-dimensional morphological features of the image layer by layer, and finally generates a probability vector in the output layer. Each element of this vector corresponds to the confidence level of a preset particle category (such as pollen, fungal spores, bacterial aggregates, or abiotic fluorescent dust). The control and data processing module selects the category with the highest probability as the final classification type of the particle and updates the counter for the corresponding category.

[0061] In the concentration inversion stage, the control and data processing module calculates the mass concentration or number concentration of that type of bioaerosol in the current ambient air based on the statistically determined number of particles of a specific category, combined with the sampled physical parameters. For the first... Bio-aerosols, their concentration in the air Calculated based on the following formula: ; in, To identify the first in a single detection cycle The cumulative amount of bio-aerosols; The sampling volume flow rate maintained by the aerosol sampling module, typically expressed in liters per minute (L / min). The duration of a single sampling period, in minutes (min); The overall capture efficiency coefficient of the system is a dimensionless parameter that combines the physical impact efficiency of the accelerating nozzle with the recall rate of the image recognition algorithm. It is usually pre-calibrated through standard aerosol experiments.

[0062] Through the above algorithm logic, the system completes the transformation from microscopic image information to macroscopic environmental monitoring indicators, and stores the final concentration data in a local database or transmits it to a remote monitoring terminal through a communication interface.

Claims

1. An automatic bioaerosol identification system based on ultraviolet fluorescence imaging, characterized in that, include: An environmental control module includes an outer casing and a sealed optical darkroom disposed inside the outer casing; The aerosol sampling module includes an inlet air passage, an accelerating nozzle, and a flow control unit, wherein the accelerating nozzle extends into the sealed optical dark chamber; The tape transport module includes a carrier tape, a drive roller, and a stepper motor disposed in the sealed optical dark chamber. The carrier tape is located directly below the acceleration nozzle, and the stepper motor is used to drive the carrier tape to move stepwise between the sampling position and the detection position. The ultraviolet microscopy imaging module includes an ultraviolet excitation source, a microscope objective, and an image sensor. The microscope objective is aimed at the area where the carrier tape is located at the detection position, and is used to acquire fluorescence microscopic images of particles attached to the surface of the carrier tape. The control and data processing module is electrically connected to the flow control unit, the stepper motor, the ultraviolet excitation source and the image sensor, respectively, and is used to control the timing of sampling and imaging, and to identify and count the fluorescence micrographs.

2. The automatic bioaerosol identification system based on ultraviolet fluorescence imaging according to claim 1, characterized in that, The outer casing is equipped with a temperature regulation unit to maintain the internal operating temperature of the system within a preset range. The inner wall of the sealed optical darkroom is coated with an matting material to shield external ambient light and stray light from inside the system.

3. The automatic bioaerosol identification system based on ultraviolet fluorescence imaging according to claim 1, characterized in that, The aerosol sampling module operates based on the principle of inertial impaction. A preset distance is maintained between the outlet end face of the accelerating nozzle and the surface of the carrier tape, so that the bioaerosol particles in the airflow detach from the streamline and adhere to the surface of the carrier tape when the airflow changes direction.

4. The automatic bioaerosol identification system based on ultraviolet fluorescence imaging according to claim 1, characterized in that, The tape transport module also includes a rigid support plate, which is disposed on the back of the tape carrier and located directly below the microscope objective, to keep the flatness of the tape carrier at the detection position within the depth of field of the microscope objective.

5. The automatic bioaerosol identification system based on ultraviolet fluorescence imaging according to claim 1, characterized in that, The ultraviolet microscopy imaging module adopts a coaxial incident optical path structure and also includes a dichroic beam splitter and a long-pass filter. The dichroic beam splitter is positioned at the intersection of the ultraviolet excitation source, the microscope objective, and the optical axis of the image sensor. The dichroic beam splitter is used to reflect the excitation light emitted by the ultraviolet excitation source to the microscope objective and to allow the fluorescence signal generated by the particles to be transmitted. The long-pass filter is disposed between the dichroic beam splitter and the image sensor to filter out residual excitation light entering the image sensor.

6. The automatic bioaerosol identification system based on ultraviolet fluorescence imaging according to claim 5, characterized in that, The center wavelength of the ultraviolet excitation source is selected from 280nm±10nm or 385nm±25nm; The cutoff wavelength of the dichroic spectroscope is between the center wavelength and the peak fluorescence emission wavelength of the target bioaerosol.

7. The automatic bioaerosol identification system based on ultraviolet fluorescence imaging according to claim 1, characterized in that, The steps for the control and data processing module to identify and count the fluorescence micrographs include: Background subtraction processing is performed on the fluorescence micrograph: A background image of a blank carrier tape is acquired, and the acquired fluorescence micrograph is compared with the background image to eliminate the autofluorescence interference of the carrier tape substrate.

8. The automatic bioaerosol identification system based on ultraviolet fluorescence imaging according to claim 1, characterized in that, The control and data processing module is also used to extract the region of interest of candidate particles in the fluorescence microscopy image using an image segmentation algorithm, and input the region of interest into a preset convolutional neural network classification model, which outputs the category probability of each particle and determines the particle type.

9. The automatic bioaerosol identification system based on ultraviolet fluorescence imaging according to claim 1, characterized in that, The control and data processing module is also used to calculate the concentration of bioaerosols based on the set parameters; The parameters include: sampling flow rate per unit time, sampling duration of a single sampling cycle, number of identified category particles, and the sampler's capture efficiency coefficient.

10. An automatic identification method for bioaerosols based on ultraviolet fluorescence imaging, characterized in that, The automatic bioaerosol identification system based on ultraviolet fluorescence imaging, as described in any one of claims 1-9, comprises the following steps: The flow control unit is activated, causing air containing bioaerosols to impact a stationary carrier tape through an accelerating nozzle for a preset sampling time. The stepper motor is controlled to drive the carrier tape to move a preset step distance, transporting the sampled area to the detection position; Turn on the ultraviolet excitation light source and use a microscope objective to excite the particles attached to the surface of the carrier tape to produce fluorescence; Acquire fluorescence microscopic images using an image sensor; The fluorescence microscopy images were preprocessed by denoising and background subtraction, and the regions of interest of the particles were extracted by connected component analysis. Deep learning algorithms are used to classify and identify regions of interest, count the number of various types of bioaerosols, and calculate their concentrations.

Images