A pollen monitoring method based on fourier-wavelet transform fusion

Abstract

The application discloses a pollen monitoring method based on Fourier-wavelet transform fusion, and belongs to the technical field of environmental monitoring, and specifically comprises the following steps: S1, pollen sampling is carried out by using a sampler, and a hologram of the pollen sample is obtained; S2, numerical reconstruction is carried out on the obtained hologram, Fourier transform and frequency domain transfer function are used to simulate the propagation of light, and a complex amplitude distribution on a reconstruction plane is obtained; S3, two-dimensional wavelet decomposition is carried out on the reconstructed image, and wavelet coefficient variance is calculated; S4, multi-scale depth fusion reconstruction is carried out, the maximum wavelet coefficient variance is selected by traversing different reconstruction depths, and an optimal reconstructed image is obtained through fusion; S5, post-processing is carried out on the reconstructed image; and S6, particles are detected on the processed image, morphological features are extracted, and pollen particles are classified, counted and visualized based on the features. The application significantly improves the accuracy of reproduction and identification of pollen particles in the lensless digital holographic technology.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to a pollen monitoring method based on Fourier-wavelet transform fusion. Background Technology

[0002] Pollen monitoring plays a vital role in ecological research, agricultural production, and public health. Effective pollen monitoring data can serve as a bioindicator of environmental change, providing crucial data support for relevant scientific research; it can also optimize urban green space planning, guide public travel, and help allergy sufferers take preventative measures in advance to reduce health risks.

[0003] Traditional pollen monitoring mainly relies on manual collection and microscopic analysis. However, the sampling and analysis processes of this method are easily affected by human factors, resulting in significant uncertainty in the monitoring results.

[0004] Digital holography uses photoelectric sensors to record the interference pattern between the object's light wave and a reference light wave, and then reconstructs the object's complex amplitude using numerical reconstruction algorithms, enabling three-dimensional measurement of information such as particle outline, particle size, and position. Lensless imaging technology based on digital holography eliminates the need for a lens-focused optical path, offering advantages such as simple structure and low cost, making it suitable for rapid monitoring of atmospheric particulate matter.

[0005] The reconstruction of holographic images is affected by factors such as sample complexity, speckle noise, sensor resolution, reconstruction algorithm, and computational efficiency. In the actual atmosphere, pollen grain size and morphology are highly diverse, and the complex patterns on the pollen surface can easily cause speckle interference. Traditional reconstruction algorithms are difficult to separate effective information. These reasons make it difficult for lensless digital holography to achieve accurate automatic identification and classification of pollen species. In addition, existing pollen databases are mostly based on optical microscopes or electron microscopes, which have problems with incompatibility with holographic imaging features.

[0006] Furthermore, in pollen monitoring and identification applications, while pollen detection methods based on Fourier-wavelet transform fusion significantly improve detection and identification accuracy, they still face a specific technical challenge in practical applications. During pollen sampling, other non-pollen particles, such as dust and tiny insect fragments, may be introduced. These interfering substances may resemble some pollen particles in morphology and size. Existing methods primarily rely on morphological features for classification, making it difficult to effectively distinguish these non-pollen interfering substances, potentially leading to misjudgments and false positives. This problem affects the accuracy of pollen monitoring, especially under conditions of poor air quality or complex sampling environments. Summary of the Invention

[0007] To address the shortcomings of existing lensless digital holographic technology in pollen monitoring and identification, this invention provides a pollen monitoring method based on Fourier-wavelet transform fusion, which can significantly improve the accuracy of pollen particle reproduction and identification in this technology.

[0008] To solve the above-mentioned technical problems, the basic idea of ​​the technical solution adopted by the present invention is as follows:

[0009] A pollen monitoring method based on Fourier-wavelet transform fusion includes the following steps:

[0010] S1, Pollen is sampled using a sampler to obtain a hologram of the pollen sample;

[0011] S2, numerical reconstruction of the acquired hologram is performed, and the propagation of light is simulated using Fourier transform and frequency domain transfer function to obtain the complex amplitude distribution on the reconstruction plane;

[0012] S3, perform two-dimensional wavelet decomposition on the reconstructed image and calculate the variance of the wavelet coefficients;

[0013] S4. Perform multi-scale deep fusion reconstruction. By traversing different reconstruction depths, select the wavelet coefficients with the largest variance and fuse them to obtain the optimal reconstructed image.

[0014] S5 performs post-processing on the reconstructed image, including operations such as grayscale conversion, noise reduction, and binarization.

[0015] S6 detects pollen grains on the processed image, extracts morphological features, and classifies, counts, and visualizes pollen grains based on these features.

[0016] Preferably, step S7 is included: using carbon-based conductive adhesive to fix the glass slide onto the electron microscope sample stage, acquiring a secondary electron image of the pollen sample for comparison and verification of holographic image reconstruction, and adjusting the parameters in the pollen hologram reconstruction process according to the comparison and verification results.

[0017] Preferably, in step S1: the sampler includes an air inlet duct with a conical diffusion structure, a detachable glass slide holder, and an adjustable speed fan.

[0018] Preferably, step S5 specifically includes: post-processing the reconstructed image, including converting the floating-point reconstructed image into an 8-bit grayscale image, using Gaussian blur for noise reduction, using an adaptive thresholding method for binarization, and using opening and closing operations for noise removal and hole filling.

[0019] Preferably, step S6 specifically includes: using the cv2.findContours() function of the OpenCV library to find the particle contours in the binary image, and extracting the morphological features of the pollen particles, including area, perimeter, roundness, equivalent diameter, and major and minor axis lengths.

[0020] To address the issue that existing methods primarily rely on morphological features for classification, which makes it difficult to effectively distinguish these non-pollen interferences and may lead to misjudgments or false positives, this invention proposes a pollen monitoring method based on Fourier-wavelet transform fusion. This approach introduces a multi-band narrowband light source and a spectral sensor to capture the unique spectral characteristics of pollen grains, building upon existing morphological analysis.

[0021] To solve the above-mentioned technical problems, the basic idea of ​​the technical solution adopted by the present invention is as follows:

[0022] A pollen monitoring method based on Fourier-wavelet transform fusion, which, in addition to the above method, further includes:

[0023] The sampler integrates multiple narrowband LED light sources, covering the visible to near-infrared spectral range;

[0024] Image acquisition is performed using a multispectral camera;

[0025] For each detected particle, its reflection intensity at different wavelengths is extracted to form a spectral feature vector;

[0026] Develop a new data fusion algorithm that combines morphological features with spectral features, uses principal component analysis to reduce feature dimensionality, and then applies support vector machine for classification;

[0027] Collect and analyze the spectral characteristics of known pollen species to establish a reference database;

[0028] Implement an adaptive learning algorithm that continuously updates and optimizes the classification model based on newly collected data.

[0029] Compared with the prior art, the beneficial technical effects of the present invention include at least one of the following:

[0030] 1. Existing digital holography records the phase delay information of light waves, which is determined by the physical thickness and average refractive index of the particles. For pollen particles with complex structures, it is difficult to accurately reconstruct their geometric morphology, and numerical reconstruction distortion is unavoidable. This invention uses Fourier transform for particle morphology reconstruction, combined with wavelet transform to filter non-stationary signals, integrating morphological information at different depths, capturing more refined details of particle changes, and improving the adaptability of the data reconstruction process to pollen. Post-processing is also performed to address issues such as the potentially insufficient refractive index difference between pollen and the surrounding medium under specific conditions, resulting in weak phase difference changes in the hologram, and common pollen adhesion and porosity problems, significantly improving the accuracy of numerical reconstruction.

[0031] 2. By combining morphological and spectral features, this invention can more accurately distinguish between pollen and non-pollen grains. In addition, the adaptive learning mechanism enables the system to continuously optimize and adapt to the challenges brought by different environments and seasons. As data accumulates, the system's recognition accuracy will gradually improve, and it will also be able to identify new pollen types or abnormal samples. This optimization scheme can not only effectively address the problem of misjudging non-pollen interferences, but also improve the accuracy of pollen species identification, providing more reliable data support for applications such as air quality monitoring and allergen early warning. Attached Figure Description

[0032] Figure 1 This is a flowchart of the holographic image numerical reconstruction and feature extraction process of the present invention;

[0033] Figure 2 This is a schematic diagram of a sampler in one embodiment of the present invention;

[0034] Figure 3 This is a diagram illustrating the effect of particle feature recognition and inversion achieved by the present invention.

[0035] Figure 4 This is a comparison image of the numerically reconstructed image and the electron microscope image of the present invention. Detailed Implementation

[0036] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0037] Please refer to the following for comprehensive information. Figures 1 to 4 A pollen monitoring method based on Fourier-wavelet transform fusion includes the following steps:

[0038] S1, Pollen is sampled using a sampler to obtain a hologram of the pollen sample;

[0039] S2, numerical reconstruction of the acquired hologram is performed, and the propagation of light is simulated using Fourier transform and frequency domain transfer function to obtain the complex amplitude distribution on the reconstruction plane;

[0040] S3, perform two-dimensional wavelet decomposition on the reconstructed image and calculate the variance of the wavelet coefficients;

[0041] S4. Perform multi-scale deep fusion reconstruction. By traversing different reconstruction depths, select the wavelet coefficients with the largest variance and fuse them to obtain the optimal reconstructed image.

[0042] S5 performs post-processing on the reconstructed image, including operations such as grayscale conversion, noise reduction, and binarization.

[0043] S6 detects pollen grains on the processed image, extracts morphological features, and classifies, counts, and visualizes pollen grains based on these features.

[0044] The core of this invention lies in combining Fourier transform and wavelet transform to achieve accurate reconstruction and identification of pollen grains. First, Fourier transform is used to reconstruct the grain morphology. Fourier transform converts spatial domain information to the frequency domain, and by simulating light propagation through the frequency domain transfer function, three-dimensional reconstruction is achieved. This step can initially restore the morphology of pollen grains. Second, wavelet transform is introduced to filter non-stationary signals. Wavelet transform has multi-resolution characteristics, capable of capturing local features of signals at different scales. By performing two-dimensional wavelet decomposition on the reconstructed image, different frequency components can be identified, effectively suppressing speckle noise and environmental interference. Then, multi-scale depth fusion reconstruction integrates morphological information from different depths. Since pollen is typically distributed in different depth planes, by selecting the coefficient with the largest variance in each frequency band, the clearest details at each depth can be fused into a single image, more accurately capturing the detailed changes in the particles. Finally, image post-processing and morphological feature extraction further improve the accuracy of pollen grain identification. Post-processing steps, such as Gaussian blur denoising and adaptive threshold binarization, help address issues like insignificant refractive index between pollen and the surrounding medium, adhesion, and porosity. Morphological feature extraction provides a reliable basis for subsequent classification and statistics. This method significantly improves the adaptability of the data reconstruction process to pollen, overcomes the limitations of traditional methods in handling pollen with complex structures, and thus improves the accuracy of pollen detection and identification.

[0045] The following is a description of the specific implementation process of this method:

[0046] 1. Sampling Stage: Pollen sampling is performed using a specially designed sampler. The sampler has a conical air inlet 1 with a conical diffusion structure at the top, forming a unidirectional laminar flow field, and a built-in adjustable-speed fan to accelerate pollen sampling. A detachable glass slide holder 2 is installed in the sampler chamber for collecting pollen samples.

[0047] 2. Holographic image acquisition: After sampling stops, the glass slide containing the pollen sample is used for holographic imaging detection to acquire the hologram of the pollen sample.

[0048] 3. Numerical Reconstruction:

[0049] (1) Define the spatial frequency domain. Construct a frequency domain coordinate system to provide a computational network for subsequent calculations:

[0050]

[0051]

[0052] In the formula, Represents spatial frequency coordinates. This indicates the number of pixels horizontally and vertically in the image. Indicates pixel size.

[0053] (2) Establish the transfer function. Simulate the physical process of light waves propagating from the holographic plane to the reconstruction plane, and achieve three-dimensional reconstruction through frequency domain phase modulation. It can be expressed by the following formula:

[0054]

[0055] In the formula, This represents the frequency domain transfer function based on the Fourier transform. Indicates wave number, Wavelength; Indicates the transmission distance. The changes indicate that object layers of different depths can be reconstructed; and Represents spatial frequency components.

[0056] (3) Holographic reconstruction calculation. In the frequency domain, the transfer function is used to simulate the propagation of light. The original hologram is Fourier transformed and then multiplied by... Then, an inverse Fourier transform is performed to obtain the complex amplitude distribution on the reconstruction plane:

[0057]

[0058] In the formula, Represents the original holographic image. Two-dimensional Fourier transform, This represents the two-dimensional inverse Fourier transform. This represents the reconstructed amplitude image.

[0059] 4. Wavelet decomposition:

[0060] (1) For each reconstructed image Perform two-dimensional wavelet decomposition:

[0061]

[0062] In the formula, This represents the two-dimensional discrete wavelet transform function. These represent different wavelet coefficients, namely low-frequency components and vertical, horizontal, and diagonal high-frequency components.

[0063] (2) Calculate the variance of wavelet coefficients:

[0064]

[0065] In the formula, Represents the variance of the coefficients. Represents wavelet coefficients, Represents the mean of a specific coefficient. This indicates the total number of coefficients.

[0066] 5. Multi-scale depth fusion: In multi-layer reconstruction, different reconstruction depths are traversed. We perform wavelet decomposition on the image at each depth. Then, we select the wavelet coefficients with the largest variance and obtain the optimal reconstructed image through multi-depth fusion. This can be expressed by the formula: for a specific wavelet coefficient... ,if,

[0067]

[0068]

[0069] This fusion method is based on the principle that reconstructed images at different depths focus on particles at different locations. When the reconstruction depth matches the actual location of the particle, the image of that particle is the clearest, and the variance of its corresponding wavelet coefficients (especially the high-frequency components) is the largest. In the reconstruction of pollen particles, low-frequency components represent the main structural and contour information of the particles; a larger variance indicates better contrast and clearer structure. High-frequency components contain detailed information such as the edges and textures of the particles; a larger variance indicates richer details contained in that high-frequency component. In summary, by selecting the coefficient with the largest variance in each frequency band, we can fuse the clearest details at each depth into a single image.

[0070] 6. Image Post-processing: Post-processing is performed on the reconstructed image to facilitate morphological feature analysis. Specific steps include converting the floating-point reconstructed image to an 8-bit grayscale image; using Gaussian blur to reduce background noise; using an adaptive thresholding method to binarize the image for contour detection; and using opening and closing operations for noise removal and hole filling, respectively.

[0071] 7. Particle detection and feature extraction: The cv2.findContours() function of the OpenCV library is used to detect particle contours and extract morphological features such as area, perimeter, roundness, equivalent diameter, and major and minor axis lengths.

[0072] 8. Classification and Statistics: Based on the extracted features, pollen grains are classified, statistically analyzed, and visualized.

[0073] Through this series of steps, the present invention can effectively reconstruct and identify complex pollen grains, overcoming the limitations of traditional methods in processing pollen samples with complex structures and uneven distribution, and significantly improving the accuracy of pollen detection and identification.

[0074] Figure 3The diagram shows the results of particle feature recognition and inversion, demonstrating the reconstruction results of particle holograms, particle locations, and result confidence levels. This proves that the proposed method can not only reproduce the morphological features of collected atmospheric particulate matter, but also accurately separate morphologically similar but different pollen species, such as juniper and willow, and effectively split adherent particles. Among the extracted morphological parameters, roundness can be directly used for classification model training.

[0075] In one embodiment of the present invention, step S7 is included: using carbon-based conductive adhesive to fix a glass slide onto the electron microscope sample stage, acquiring a secondary electron image of the pollen sample for comparison and verification of holographic image reconstruction, and adjusting the parameters in the pollen holographic reconstruction process according to the comparison and verification results.

[0076] This embodiment provides an independent verification method to evaluate the accuracy of holographic image reconstruction results. High-resolution images obtained using an electron microscope can serve as a reference standard, helping to improve the accuracy of the holographic reconstruction algorithm. Furthermore, through comparative verification, the reconstruction algorithm and parameter settings in the core technical solution can be continuously optimized and improved. The parameters adjusted in the pollen hologram reconstruction process mainly include reconstruction depth and wavelet function coefficients, which determine the details of image fusion. Additionally, the noise reduction intensity and adaptive threshold window in the post-processing stage determine the accuracy of particle contour detection and morphological feature extraction.

[0077] Figure 4 The paper presents reconstructed images and electron microscope (EM) comparisons of pollen from juniper, willow, and apricot trees. The images show consistency between the reconstructed and EM images in key morphological features. For example, juniper pollen appears circular, willow pollen elliptical, and apricot pollen triangular. Simultaneously, the EM image provides a high-resolution morphological image of the pollen, which can be compared with the reconstructed image. The results demonstrate a high degree of agreement between the pollen grain size data calibrated by the EM image and the measured values ​​in the reconstructed image, proving that the reconstruction technique of this invention can achieve micrometer-level geometric accuracy. This dual consistency in morphology and size verifies the effectiveness of the Fourier-wavelet fusion algorithm in the three-dimensional reconstruction of complex biological particles, overcoming the bottleneck of traditional holographic techniques in restoring pollen microstructure.

[0078] In one embodiment of the present invention, in step S1: the sampler includes an air inlet duct with a conical diffusion structure, a detachable glass slide holder, and an adjustable speed fan.

[0079] In this embodiment, the sampler includes the following structures: 1. A conical air inlet duct 1: located at the top of the sampler, its conical shape facilitates the formation of a unidirectional laminar flow field, improving sampling efficiency and uniformity; 2. A detachable slide holder 2: installed in the sampler chamber, used to fix quartz slides, facilitating rapid sample replacement and processing, and improving work efficiency; 3. An adjustable-speed fan mounting bracket 3: used to install an adjustable-speed fan, which can adjust the sampling speed according to different environmental conditions, increasing sampling flexibility and adaptability. This design makes the sampling process more efficient and controllable, providing high-quality samples for subsequent holographic image acquisition.

[0080] In one embodiment of the present invention, step S5 specifically includes: post-processing the reconstructed image, including converting the floating-point reconstructed image into an 8-bit grayscale image, using Gaussian blur for noise reduction, using an adaptive thresholding method for binarization, and using opening and closing operations for noise removal and hole filling.

[0081] In this embodiment, according to Figure 4 Taking the juniper pollen mentioned in the text as an example: 1. The reconstructed floating-point juniper pollen image is converted into an 8-bit grayscale image, with grayscale values ​​ranging from 0 to 255; 2. Gaussian blur is applied to the grayscale image using a 3x3 or 5x5 Gaussian kernel to reduce background noise; 3. Binarization is performed using an adaptive thresholding method, dividing the image into black and white levels to highlight the pollen grain outline; 4. Opening operations are applied to remove fine noise, and then closing operations are used to fill the small pores on the pollen surface. These post-processing steps make the outline and surface features of the juniper pollen clearer, laying the foundation for subsequent morphological feature extraction and classification.

[0082] In one embodiment of the present invention, step S6 specifically includes: using the cv2.findContours() function of the OpenCV library to find the particle contours in the binary image, and extracting the morphological features of the pollen particles, including area, perimeter, roundness, equivalent diameter, and major and minor axis lengths.

[0083] In this embodiment, according to Figure 4 Taking willow pollen as an example: 1. Use the cv2.findContours() function to detect pollen grain contours on the post-processed binary image; 2. Calculate the area (in pixels²) and perimeter (in pixels) of each detected contour; 3. Calculate the roundness: 4π * area / perimeter², willow pollen is usually circular, with a roundness close to 1; 4. Calculate the equivalent diameter: 5. Use the cv2.fitEllipse() function to fit an ellipse, obtain the lengths of the major and minor axes, and describe the shape characteristics of the pollen. Through these steps, we can obtain detailed morphological characteristics of willow pollen, which can be used to distinguish willow pollen from other types of pollen, such as juniper or apricot pollen.

[0084] In one embodiment of the present invention, a pollen monitoring method based on Fourier-wavelet transform fusion further includes:

[0085] The sampler integrates multiple narrowband LED light sources, covering the visible to near-infrared spectral range;

[0086] Image acquisition is performed using a multispectral camera;

[0087] For each detected particle, its reflection intensity at different wavelengths is extracted to form a spectral feature vector;

[0088] Develop a new data fusion algorithm that combines morphological features with spectral features, uses principal component analysis to reduce feature dimensionality, and then applies support vector machine for classification;

[0089] Collect and analyze the spectral characteristics of known pollen species to establish a reference database;

[0090] Implement an adaptive learning algorithm that continuously updates and optimizes the classification model based on newly collected data.

[0091] In this embodiment, the narrowband LED light source refers to a light source that emits a specific wavelength range. Specifically, it can be implemented using an LED array with a center wavelength spacing of 50-100nm, covering a spectral range of 400-1000nm, to stimulate differences in the reflectance characteristics of pollen particles at different wavelengths. The multispectral camera refers to a sensor capable of simultaneously acquiring images across multiple wavelength bands. Specifically, it can be implemented using an imaging system combining a beam splitter prism and multiple CCDs, to acquire morphological and spectral response information of the particles. The spectral feature vector refers to multidimensional data composed of the reflectance intensity of particles at multiple wavelengths. Specifically, it can be formed by normalizing the gray values ​​of each channel to create a vector, used to characterize the differences in spectral reflectance characteristics of different substances. The data fusion algorithm refers to a machine learning model that combines morphological and spectral features. Specifically, it can use a feature-level fusion method to concatenate the two types of features into a hybrid feature vector, used to improve the discriminative ability of the classifier. Principal component analysis refers to a linear dimensionality reduction method, specifically implemented by calculating the eigenvectors of the covariance matrix, used to remove redundant information and retain key classification features. A reference database refers to a sample library storing known pollen spectral characteristics. Specifically, a relational database can be used to store spectral data of different species under standard testing conditions, providing training samples for the classification model. An adaptive learning algorithm is a machine learning method that dynamically updates model parameters. Specifically, it can be implemented using incremental support vector machines to continuously optimize the classification model based on newly collected data.

[0092] Specifically, during the sampling phase, morphological images and multispectral reflectance data of the particles are simultaneously acquired using a multispectral imaging system. The contour features of each detected particle in the visible light band and its spectral response in the near-infrared band are extracted separately, forming morphological parameters including area, perimeter, and equivalent diameter, as well as reflectance intensity vectors for 5-8 feature bands. These two types of features are merged using a data fusion algorithm, and principal component analysis compresses the mixed feature dimension from 20-30 dimensions to 5-8 dimensions, eliminating correlations between features. During the training phase, a support vector machine classification model is built using spectral-morphological feature combinations of known pollen from a reference database. During the deployment phase, an adaptive learning algorithm feeds the classification results of newly detected particles back into the model, dynamically adjusting the classification hyperplane parameters.

[0093] Compared to existing technologies, traditional methods rely solely on single morphological features for classification, failing to distinguish interfering substances with significantly different spectral characteristics. This approach introduces multispectral imaging technology, utilizing the differences in reflectance between pollen and dust in the near-infrared band, combined with morphological features to construct hybrid classification features. Existing static classification models cannot adapt to environmental changes; this approach employs an incremental learning mechanism, enabling the classifier to continuously optimize based on new samples, thus addressing the problem of insufficient generalization ability of traditional methods in complex sampling environments.

[0094] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A pollen monitoring method based on Fourier-wavelet transform fusion, characterized in that, Includes the following steps: S1, Pollen is sampled using a sampler to obtain a hologram of the pollen sample; S2, numerical reconstruction of the acquired hologram is performed, and the propagation of light is simulated using Fourier transform and frequency domain transfer function to obtain the complex amplitude distribution on the reconstruction plane; S3, perform two-dimensional wavelet decomposition on the reconstructed image and calculate the variance of the wavelet coefficients; S4. Perform multi-scale deep fusion reconstruction. By traversing different reconstruction depths, select the wavelet coefficients with the largest variance and fuse them to obtain the optimal reconstructed image. S5 performs post-processing on the reconstructed image, including operations such as grayscale conversion, noise reduction, and binarization. S6 detects particles on the processed image, extracts morphological features, and classifies, counts, and visualizes pollen particles based on these features; in: The sampler integrates multiple narrowband LED light sources, covering the visible to near-infrared spectral range; Image acquisition is performed using a multispectral camera; For each detected particle, its reflection intensity at different wavelengths is extracted to form a spectral feature vector; Develop a new data fusion algorithm that combines morphological features with spectral features, uses principal component analysis to reduce feature dimensionality, and then applies support vector machine for classification; Collect and analyze the spectral characteristics of known pollen species to establish a reference database; Implement an adaptive learning algorithm that continuously updates and optimizes the classification model based on newly collected data.

2. The pollen monitoring method based on Fourier-wavelet transform fusion according to claim 1, characterized in that, Step S7 includes: using carbon-based conductive adhesive to fix the glass slide onto the electron microscope sample stage, acquiring a secondary electron image of the pollen sample for comparison and verification with the holographic image reconstruction, and adjusting the parameters in the pollen holographic reconstruction process based on the comparison and verification results.

3. The pollen monitoring method based on Fourier-wavelet transform fusion according to claim 1, characterized in that, In step S1: the sampler includes an air inlet with a conical diffusion structure, a detachable glass slide holder, and an adjustable speed fan.

4. The pollen monitoring method based on Fourier-wavelet transform fusion according to claim 1, characterized in that, Step S5 specifically includes: post-processing the reconstructed image, including converting the floating-point reconstructed image to an 8-bit grayscale image, using Gaussian blur for noise reduction, using an adaptive thresholding method for binarization, and using opening and closing operations for noise removal and hole filling.

5. The pollen monitoring method based on Fourier-wavelet transform fusion according to claim 1, characterized in that, Step S6 specifically includes: using the cv2.findContours() function of the OpenCV library to find the particle contours in the binary image and extracting the morphological features of the pollen particles, including area, perimeter, roundness, equivalent diameter, and major and minor axis lengths.

Images