CN111896430B

CN111896430B - Pollen monitoring method and device

Info

Publication number: CN111896430B
Application number: CN202010011480.7A
Authority: CN
Inventors: 戎恺
Original assignee: Shanghai Kaiqing Intelligent Technology Co ltd
Current assignee: Shanghai Kaiqing Intelligent Technology Co ltd
Priority date: 2020-01-06
Filing date: 2020-01-06
Publication date: 2024-11-19
Anticipated expiration: 2040-01-06
Also published as: CN111896430A

Abstract

The invention relates to the technical field of environmental monitoring, in particular to a pollen monitoring method and a pollen monitoring device; including air sampling; separating particles in the air according to the size; fixing the sampled particles on a glass slide; imaging by a particle sampling microscope; analyzing images; storing the processed slide; the monitoring efficiency is greatly improved, the labor intensity is reduced, the monitoring accuracy is improved, and in addition, the sample can be observed repeatedly by fixing pollen particles.

Description

Pollen monitoring method and device

Technical Field

The invention relates to the technical field of environmental monitoring, in particular to a pollen monitoring method and a pollen monitoring device.

Background

The traditional collection method mainly adopts a gravity sedimentation method to collect pollen, and then adopts a manual microscopic examination method to classify and count pollen. The artificial microscopic examination method is that an experienced observer carefully observes pollen samples which are 400 times amplified under a biological optical microscope line by line, manually distinguishes the types of pollen particles, and manually counts the number of different pollens and the total number of pollens. The manual microscopy method mainly has the following defects:

1) The labor capacity of observers is too great, so that physical fatigue is easily caused, and the monitoring accuracy is reduced;

2) The observation technical requirements on the monitor are high, long-term continuous technical training is needed, and the microscopic morphological characteristics of different pollen particles can be accurately mastered, so that pollen can be accurately identified, and a new observer needs to be retrained once the observer is mobilized. The manpower and material resources are relatively large;

3) The monitoring result is influenced by human factors, and the observation results obtained by different observers have larger difference;

4) In the pollen transmission peak period, the manual distinguishing time is prolonged, and the eye fatigue of an observer is easily caused, so that the pollen counting working efficiency is influenced.

In addition, the quality guarantee period of the sample by the microscopic examination method is short, so that the sample cannot be stored, and the sample cannot be recounted again due to the loss of data. In addition, the sample cannot be counted for multiple times so as to prevent counting errors and other problems. Because of the above defects in the technology, the current manual microscopic examination method has low monitoring and analysis efficiency, and the monitoring accuracy is difficult to ensure. In addition, because the pollen scattering peak period data is manually monitored and does not have online capability, the pollen scattering peak period data is released at least 24 hours away from the peak period, and no method for instant forecasting exists, so that people are reminded of early prevention.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides a pollen monitoring method and a monitoring device which can be used for the monitoring method.

The technical scheme of the invention is as follows:

A pollen monitoring method comprising the steps of:

1) Sampling air;

2) Separating particles in the air according to the size;

3) Fixing the sampled particles on a glass slide;

4) Imaging by a particle sampling microscope;

5) Image analysis (including classification and counting);

6) Storing the treated slide (containing the analyzed particles);

7) And carrying out the next round of sampling, and transmitting the analysis result to the server according to the set time interval.

Further, in the step 1, a virtual impactor (also called an impact type sampler) is used to collect surrounding air and impact the air on a glass slide, gelatin and glycerin are coated on the glass slide, and pollen particles in the air are adhered by the glycerin, so that sampling is realized.

Further, in the step 2, the air is collected at a speed of 60m ³/h in the air around the virtual impactor, the virtual impactor takes the central air flow (about 6m ³/h), and the virtual impactor firstly separates particles with different sizes in the air and then pushes the particles with the particle size according with the pollen to the glass slide.

Further, in the step 3, the gelatin is melted by heating, pollen particles adhered by glycerol are trapped in the gelatin, and then the pollen is permanently fixed in the gelatin by cooling, so that the subsequent process is facilitated.

Further, in the step 4, the sample on the slide glass is scanned and photographed by using an electron microscope, so that an image stack is obtained and stored in a hard disk. Specifically:

The slide glass and the electron microscope can rotate at a certain angle to facilitate shooting of samples at a plurality of angles, the electron microscope shoots the samples at 120 positions on each slide glass, and the shooting results are spliced again into an image stack which facilitates subsequent analysis and processing according to the separation of target particles by utilizing image preprocessing (mainly comprising image transformation, image enhancement, edge detection and segmentation methods) and stored in a hard disk.

And further, in the step 5, the image stack obtained in the step 4 is exported from a hard disk for image analysis, and pollen classification identification and counting are further carried out on the image stack through big data and a machine learning algorithm. Mainly comprises the following procedures:

1) And (5) extracting characteristics. Feature extraction is carried out on an image stack of target particles (mainly comprising pollen outline features, pollen structure features and pollen texture features), if one feature extraction fails, the particles are judged not to be pollen, and if all the extraction is successful, the extracted parameter values form a feature vector set of each particle object (used for automatic classification in the next step).

2) Automatic (identification) classification. The system uses a multi-stage classifier (comprising three sub-classifiers: an outer contour recognition module, a membership-based internal structure recognition module and a texture recognition module) to classify pollen particles of a target (determine plant species to which pollen corresponds).

3) Counting. Repeating the steps to process all the shooting results of the electron microscope in a period of time to obtain the quantity of pollen of different plant types.

Pollen monitoring devices, he includes rotating device, rotatory sample platform, slide glass and heating module, rotating device installs on a base, rotating device's pivot is vertical upwards to be set up, the pivot top is connected with rotatory sample platform, be provided with a plurality of recesses on the rotatory sample platform, the recess is circular and its bottom is provided with the through-hole, the slide glass has been placed to detachably in the recess, be provided with the gelatin layer on the slide glass, be provided with the glycerol layer on the gelatin layer, heating module is still installed on the base, heating module's heating end corresponds rotatory sample platform's recess setting for the slide glass can be heated by heating module's heating end when the recess rotates to heating module's top. When the device is used, pollen is impacted on the glass slide through the impact type sampler and other devices, at the moment, the pollen is stuck by the glycerol layer, and then the rotary sample table is rotated, so that the glass slide in the groove is positioned above the heating module and heated. The pollen is sunk into the melted gelatin layer after heating, and the fixation of the pollen can be realized after cooling. And then the rotary sample stage is rotated, and the pollen sample is observed by using an electron microscope.

In some embodiments, the rotating means employs an indexing disc. The angle of each rotation can be precisely controlled.

In some embodiments, the material of the rotary sample stage is one of cast iron plate, stainless steel plate or high temperature resistant resin plate. Has enough hardness and is not afraid of high temperature.

In some embodiments, the heating module is a far infrared heater with its heating direction facing vertically upwards. As is well known, the far infrared heater has limited heating effect on the air medium, so most heat can directly reach the glass slide, the heating effect is improved, the heat loss is reduced, and the air around the device is not heated greatly.

The beneficial effects of the invention are as follows: the monitoring efficiency is greatly improved, the labor intensity is reduced, the monitoring accuracy is improved, and in addition, the sample can be observed repeatedly by fixing pollen particles.

Drawings

FIG. 1 is a schematic diagram of the present invention;

FIGS. 2-3 are schematic diagrams of step 4;

fig. 4 is a schematic diagram of a multi-stage classifier.

Detailed Description

The following is a further description of embodiments of the invention, taken in conjunction with the accompanying drawings:

example 1

As shown in fig. 1-3, gas is collected from ambient air at a rate of 60m ³/h, a central gas stream (about 6m ³/h) is taken into a virtual impactor for later sampling analysis, the virtual impactor first separates particles of different sizes in the air, then pushes particles of a size consistent with the pollen particle size toward a slide (covered with gelatin and glycerol on the slide), the virtual impactor is operated continuously at set intervals until one sampling period is completed (the sampling period can be set, for example, 3 hours), the rotary sample stage is rotated to a heating module, the slide is heated to 90 ℃, and the viscous surface on the slide is melted at high temperature so that the particles are trapped in a medium, so that the positions of the particles are fixed and adjusted, and the subsequent electron microscope observation focusing is facilitated. And then rotating the sample stage to rotate the glass slide to a position above the electron microscope, then scanning and shooting the sample by using the electron microscope, and enabling the glass slide and the electron microscope to rotate at a certain angle to facilitate shooting of the sample at a plurality of angles, and combining the shooting results into a pollen image stack and storing the pollen image stack into a hard disk. After the scan is complete, the handling system places the sample slide into the storage cassette. This cartridge has sufficient capacity to hold samples for a full month.

The computer system loads the image stack from the hard disk for image analysis. Separating pollen from other aerosol particles, and identifying pollen type according to morphological characteristics of pollen.

The sampling period may be set by the operator, typically 1-6 hours/day, with each sampling period consuming one slide, which is stored in a reusable storage cartridge.

The electron microscope photographs 120 selected positions of the specimen on each slide, and the photographed results are separated according to the target particles by image preprocessing and constitute an image stack. The graphic analysis module then analyzes the image stack, uses a plurality of automatic classification condition combinations (including length, width, radius, gap number, texture, etc.) in the analysis process, and finally stores the image stack and the analysis result together in a local hard disk.

The image analysis processing flow is as follows:

1) And (5) preprocessing an image. And (3) taking the results of the sample from multiple angles by using an electron microscope, and re-splicing the taken results into an image stack which is convenient for subsequent analysis and processing after the taken results are separated according to target particles by utilizing image preprocessing (mainly comprising image transformation, image enhancement, edge detection and segmentation methods).

2) And (5) extracting characteristics. Feature extraction is carried out on an image stack of target particles (mainly comprising pollen outline features, pollen structure features and pollen texture features), if one feature extraction fails, the particles are judged not to be pollen, and if all the extraction is successful, the extracted parameter values form a feature vector set of each particle object (used for automatic classification in the next step).

3) Automatic (identification) classification. The system uses a multi-stage classifier (comprising three sub-classifiers: an outer contour recognition module, a membership-based internal structure recognition module and a texture recognition module) to classify pollen particles of a target (determine plant species to which pollen corresponds).

4) Counting. Repeating the steps to process all the shooting results of the electron microscope in a period of time to obtain the quantity of pollen of different plant types.

Remarks:

when the automatic (identifying) classification result cannot judge the pollen plant type, the later stage can manually review the image to judge, and the judged plant type is input into the system, and then the system encounters the pollen of the type and can automatically identify.

The following describes in detail the key steps of image preprocessing, feature extraction, and automatic (recognition) classification in the image analysis processing flow:

1. image preprocessing

1.1. Image transformation

The image transformation referred to herein is referred to as two-dimensional orthogonal transformation, which plays an important role in image processing. The image transformation is mainly considered for further image processing, such as that the average value after fourier transformation is proportional to the average value of the image gray scale, and the high frequency component indicates the edge information of the image, and the characteristic value can be extracted from the image by using the properties. In the system, common image transformation methods such as Fourier transformation, cosine transformation, KL transformation, wavelet transformation and the like are designed, and proper transformation methods are used according to different image characteristics.

1.2. Image enhancement

The purpose of image enhancement is to improve the visual effect of the image, to increase the sharpness of the image, or to convert the image into a form more suitable for human eye observation and machine automated analysis. If the sharpening process can highlight the edge contour line of the image, the computer is programmed to track and can perform various feature analyses. Methods for plant pollen image enhancement include trimming of image gray level, image smoothing, image sharpening, etc.

1.3. Edge detection and segmentation

The edges of the image are the most basic features of the image, and the edges can mark out the target object, so that the observer can see the target object at a glance. The edges contain rich intrinsic information such as the direction, the step property, the shape and the like of the image, and are important attributes for extracting image features in image recognition. Essentially, an image edge is a reflection of a discontinuity in the local characteristics of the image (grey scale abrupt change, color abrupt change, etc.), which marks the end of one region and the beginning of another region. Edge extraction first detects discontinuities in the local nature of the image and then concatenates these discontinuous edge pixels into a complete boundary. The nature of the edge is that the pixels along the edge run smoothly, while the pixels perpendicular to the edge run strongly.

Image segmentation is a processing technique that divides an image into meaningful regions. "meaningful" herein refers broadly to "corresponding to an object" or "function of the problem under investigation". For example, if a topographic aerial photograph or a topographic remote sensing image is input, it is necessary to divide mountain areas, plains, water areas, forests, cities, roads, and the like. When processing plant pollen images, it is necessary to detect identifying features such as pollen holes and furrows. These "targets" separated from the image domain are the objects of the image segmentation.

The basis of image segmentation is the similarity and jump between pixels. By "similarity" is meant that pixels have some similar property, such as gray level uniformity, in a certain area; by "jump denaturation" is meant a discontinuity in characteristics, such as abrupt gray value changes. The image segmentation method is classified into a threshold method, a boundary line detection method, a matching method, a tracking method, and the like. In the system, the segmentation of the image is researched by adopting an edge extraction method based on pixel gray mutation and a region generation method based on characteristic similarity and texture analysis.

2. Feature extraction

In order to build a system for identifying objects of different types, it is first necessary to determine which properties of the object should be measured as descriptive parameters. These particular attributes being measured are referred to as the features of the object, and the resulting parameter values constitute the feature vector for each object. Proper selection of a feature is very important because it is the only basis in identifying an object.

In the image recognition problem, a small number of features with good distinguishability, reliability and independence should be extracted. The system selects 3 aspects of characteristics such as the outline, the structure, the texture and the like of pollen.

2.1. Pollen profile feature extraction

The outline of pollen has circular, elliptic, triangle, pillow shape and petal shape, and the noise formed by small sundries around pollen is considered to be larger than the background gray level difference of pollen, so the noise is removed by Gaussian filtering, the obtained outline is more regular, and edge tracking is performed after binarization. The edge tracking obtains a closed curve, and then obtains the chain code of the curve. On the basis of the chain code, the characteristic values of the circumference, the area, the regional roundness and the like of the pollen outline are extracted, the distance from the point on the closed curve to the mass center is calculated to obtain another curve, and the characteristic of the curve (such as a plurality of peaks, whether a straight line is approached or not) is taken as the characteristic of the pollen outline.

2.2. Pollen structural feature extraction

Since pollen of the same family or genus may have different profiles, pollen of different families may have similar profiles, and it is not sufficient to use only profile features. The next step is to find structural features. The structure of pollen is mainly represented by the pores and furrows inside. The grooves are thin and long, and the holes are partial circles. It is an important feature if the hole and groove information can be extracted and described. Because the holes and the grooves are three-dimensionally concave, the grey scale abrupt change is shot under an electron microscope. Gradient operators are used to find their edges.

2.3. Pollen texture feature extraction

The texture reflects some change in the object table, color and gray scale, which in turn is related to the properties of the object itself. For example, the wood of the same tree species has the same or similar texture, and people identify the tree species and the material of the wood by distinguishing the types of the wood grains. The pollen surface is smooth or wavy, and has various carved patterns such as thorns, tumors, grains, stripes, nets and the like on some pollen to form various textures. These textures are important basis for identifying pollen species.

In image recognition, texture is an attribute that reflects the spatial distribution of pixel gray levels in an area. Therefore, the texture measure of an object in an image is of interest when extracting texture features. If the gray level across the image is a constant, or near constant, it is indicated that the object is not textured. The object is textured if the gray level of the image changes significantly but not simply. In order to measure texture, one tries to quantify the nature of the image gray level variation. The texture feature is a value calculated from the image of the object that quantifies the characteristics of the gray level variation of the image of the object. Firstly, the gray level co-occurrence matrix is used for judging the thickness and the fineness of the texture. And extracting statistical parameters such as energy, entropy, contrast, correlation and the like from the gray level co-occurrence matrix as texture characteristic parameters to respectively reflect the uniformity of gray level distribution of the image, the information quantity of the image, the definition of the image and the similarity degree of gray level co-occurrence matrix elements in the row direction or the column direction.

3. Automatic (identification) classification

3.1. Principle of multistage classifier

As shown in FIG. 4, a classifier can be seen as a "machine" consisting of hardware or software. The function of the method is that c discriminant functions gi are calculated first, and then the class corresponding to the maximum value of the discriminant functions is selected as a decision result.

A multi-stage classifier is constructed in the present system. The multi-stage classifier applies different recognition technologies to pollen images with different characteristics, and fully utilizes the advantages of different recognition methods, so that the performance is superior to that of a single-stage recognition system.

In recent years, the combination of multiple classifiers has become a leading-edge research topic in the field of pattern recognition, and has achieved better application effects in many applications of pattern recognition, such as character recognition, object recognition, and the like. Many approaches have been proposed for multi-classifier combinations. Common methods include majority decision, linear weighting, bayesian estimation, and the like.

Aiming at the specificity of the system, a dynamic weight method for multi-classifier result combination is designed. The method does not fix the weight of each classifier, and outputs the candidate set sequence and the corresponding credibility. The confidence level of a candidate set refers to the degree of reliability that an input pattern belongs to this pattern (or class). If the reliability of the classification result a is far higher than that of other classification results in the classifier x, this classifier is assigned a higher weight, i.e. for a single classifier x, the higher the separability of the reliability of its classification result, the higher the weight of this classifier. The benefits of this algorithm are: when the classification result can be basically determined by a single classifier, the classifier is dominant in the multi-classifier combination. When the separation of the credibility of the classification results of all the classifiers is similar, each classifier plays an approximate role in decision making.

3.2. Multistage classifier implementation

Consists of three sub-recognition modules (sub-classifiers):

(1) Outer contour recognition module

(2) Internal structure identification module based on membership degree

(3) Texture recognition module

The correlation between the classifiers for each stage can be divided into two cases: 1. the same aspects of the same object are described, but to a different extent. 2, describing different aspects of the same object. Case 1 reflects the redundant information of the classifier and case 2 reflects the complementary information of the classifier. In the fusion of sensor data, which is a lower layer of information fusion, the condition that a certain sensor fails due to the influence of strong noise needs to be considered, and the stability of a fusion system can be maintained by using redundant information. In the fusion of the high-level classifier of the information fusion, redundant information can cause redundant estimation of the combined result, so that the redundant information needs to be removed.

Although the complementary information describes different aspects of the same object, it is not beneficial for the combination of the classifiers, only the complementary information that can improve the combination effect is useful. The key of classifier combination is how to complement each classifier, so that the combination effect is superior to each individual classifier. The division of the three sub-classifiers of the present system is also due to this consideration, maximizing complementarity and minimizing redundancy.

3.2.1. Outer contour recognition module

The extraction of the outline features comprises perimeter, area, regional roundness, angle points, peak number of the centroid section and variance of the distance from the centroid of the outline. First, the sample image can be divided into two categories according to the region circularity: circles and non-circles. When the outer edge is approximately determined to be a circle, whether the edge curve is regular or not can be determined according to the variance of the regional centroid section curve. If the number of peaks is non-circular, the number of sides of the pollen image can be approximately determined from the number of peaks of the regional centroid profile. When the number of peaks is 2, there may be two cases: oval and pillow-shaped. Since they represent two broad classes of pollen, it is necessary to distinguish them. Experiments prove that the variance of the regional centroid profile has good classification effect when the two classes are distinguished. The variance is large, and the pillow shape is considered. The variance is small and elliptical. However, the ellipse is an ellipse with a major axis greatly different from a minor axis, because the ellipse similar to a circle with a long and short axes has been classified as a circle. When the wave crest number is more than or equal to 3, the outline shape of the pollen image is not easy to determine, and the number of nodes on the edge is combined for consideration. The method can be simply considered that when the number of angular points is small, the edges are smooth, and the number of wave peaks is taken as the number of edges of the outer contour; and when the number of the corner points is large, judging the corner points to be irregular.

A more difficult point is the determination of the threshold. Such as the roundness of the region, the variance and the number of corner points of the centroid section curve of the region, etc. In the system, the method is mainly determined by empirical values, and the classification intervals are determined at the same time when the thresholds are determined. The confidence that the proxy sample belongs to each possible class is determined by the distance between these thresholds and the characteristic value of the proxy sample.

3.2.2. Internal structure identification module based on membership degree

(1) Membership degree and membership function

Concepts that do not explicitly extend are referred to as ambiguous concepts and the totality of the objects in question is referred to as the domain or space. The fuzzy concept does not extend from the common set, the degree to which the element in the universe accords with the concept is not absolute 0 or 1, and the degree can be between 0 and 1. In fuzzy mathematics, the absolute membership of elements to a common set is activated, and the concept of membership is provided. Membership may be described by a membership function.

The fuzzy set a on any argument s= { X } refers to the whole of X, which has an element with a certain property, and is not well-defined, and can be characterized by a membership function F _A (X), wherein the size of F _A (X) reflects the membership degree of X to the fuzzy set a, and the membership degree is in a closed interval [0,1 ]. If the value of F _A (x) is close to 1, it means that x is subordinate to A to a high degree; if the value of F _A (x) is close to 0, it means that x is subordinate to A to a very low degree. For the universe, the universe elements are always clear, only the subset A, B in S is ambiguous, so the ambiguous set is often referred to as an ambiguous subset. Where confusion is not likely, the fuzzy subset is simply referred to as a fuzzy set. For example, S is the number field, if the fuzzy set A represents a real number much larger than 0, i.e., A= { x|x > 0}, then the membership function of A can be written as:

(2) Internal structure classifier design

The method for judging the attribution of the sample directly by calculating the membership degree of the sample is called a membership principle of pattern classification, and also called a direct method of fuzzy pattern classification.

Membership principles:

There are n fuzzy subsets in the universe of questions U And each pair ofAll have membership functions

X ₀ is considered to be affiliated to a _i.

The membership principle is obvious and easy to be acknowledged, but how it works is very dependent on the skills of establishing membership functions of known model classes.

The Gaussian membership function is adopted in the systemThe main reason is that this function has several good characteristics: ① The presentation is simple and does not add much complexity even for multiple input variables. ② Radial symmetry conforms to the nature of a typical objective thing. ③ are positive.

Obtaining a cluster center m _i and a standard deviation sigma _i of each class by cluster analysis to obtain membership functions of each classExtracting feature vector of sample to be detectedSubstituted into the membership functions of each class. The class with the largest membership degree is obtained by the membership principle, namely the class to which the sample to be tested belongs. In this classifier, the confidence level can be simply considered as membership.

3.2.3. Texture recognition module

(1) Texture feature selection by entropy minimization

Entropy (Entropy) is a statistical measure of uncertainty. For a given population of pattern vectors, the class dispersibility is measured by the overall entropy, i.e

H＝-E_p{lnp(x)}

Where p (x) is the probability density of the pattern population and E _p is the desired operation of p (x). Features that reduce uncertainty are considered to have more information when considering the best feature selection. Thus, using entropy as an uncertainty measure, selecting features that minimize the entropy of the pattern class is a reasonable feature selection criterion.

Considering M pattern classes with probability densities p (x|ω ₁),p(x|ω₂),…,p(x|ω_M), respectively, then the entropy of the ith pattern class is

H_i＝-∫_xp(x|ω_i)ln[p(x|ω_i)]dx

The integral domain is the entire pattern space. Obviously, if p (x|ω _i) =1, then H _i =0, at which point there is no uncertainty. It follows that the smaller the entropy, the smaller the uncertainty and the better the separability.

In the texture feature extraction, various texture information features are obtained through a gray level co-occurrence matrix and a Fourier transformation texture analysis method. Because of the large redundancy between features, the entropy of the whole pattern class is large and the separability is poor. And the dimension of the feature vector is high, so that the computational complexity is greatly increased. Thus, the dimension of the feature vector must be reduced on the basis of the principle of entropy reduction.

(2) Detailed description of the invention

In the system, I firstly divide the similar characteristics into one class, and obtain a characteristic quantity describing each class of characteristics according to the indexable of each characteristic by a weighted average method. In this way, redundancy between like features is reduced. Experiments prove that the entropy value of the mode class is greatly reduced after feature selection compared with that before feature selection.

The texture roughness is described as follows: an angular second step f1 extracted from the gray level co-occurrence matrix; the gray difference probability P (d) and the entropy f ₄. R _max resulting from fourier transform. And according to the value of each characteristic, estimating the roughness expressed by each characteristic, and then carrying out weighted average to obtain the uniform roughness C ₀. Since the angular second step f1 and the fourier transform r _max are better separable, a larger weight is assigned. The gray difference probability P (d) entropy f4 assigns a small weight.

Because the similar pollen images have rotations in different directions, the system extracts the strong and weak characteristics of the pollen texture direction instead of the specific direction when extracting the texture direction characteristics. Three features are employed in the system: MAX P _i (r) indicates whether the peaks are evident on the loop feature curve P _i (r), the number of peaks N and the direction feature DIR extracted from the power spectrum matrix. Mainly determined by MAX P _i (r), N and DIR play an auxiliary role. The direction strength I _d is obtained.

Some other texture features have been found in experiments to play a good role in pollen texture classification:

contrast of gray level co-occurrence matrix Reflecting the sharpness of the texture.

Correlation of gray level co-occurrence matrixReflecting the degree of similarity of the elements of the gray co-occurrence matrix in the row direction or the column direction.

After the characteristics are selected, the average value of the characteristics of each pollen in the pollen image library is calculated, the characteristic value of the sample to be detected is compared with the average value, and the sample to be detected is divided into pollen classes with the minimum distance. The distance is normalized and then used as the credibility of each classification.

Example 2

Use pollen monitoring devices, he includes rotating device, rotatory sample platform, slide glass and heating module, rotating device installs on a base, rotating device's pivot is vertical upwards to be set up, the pivot top is connected with rotatory sample platform, be provided with a plurality of recesses on the rotatory sample platform, the recess is circular and its bottom is provided with the through-hole, the slide glass has been placed to detachably in the recess, be provided with the gelatin layer on the slide glass, be provided with the glycerol layer on the gelatin layer, still install heating module on the base, heating module's heating end corresponds rotatory sample platform's recess setting for the slide glass can be heated by heating module's heating end when the recess rotates to heating module's top. When the device is used, pollen is impacted on the glass slide through the impact type sampler and other devices, at the moment, the pollen is stuck by the glycerol layer, and then the rotary sample table is rotated, so that the glass slide in the groove is positioned above the heating module and heated. The pollen is sunk into the melted gelatin layer after heating, and the fixation of the pollen can be realized after cooling. And then the rotary sample stage is rotated, and the pollen sample is observed by using an electron microscope.

The rotating device adopts an index plate. The angle of each rotation can be precisely controlled.

The rotary sample table is made of one of cast iron plates, stainless steel plates or high-temperature-resistant resin plates. Has enough hardness and is not afraid of high temperature.

The heating module is a far infrared heater, and the heating direction of the heating module is vertically upward. As is well known, the far infrared heater has limited heating effect on the air medium, so most heat can directly reach the glass slide, the heating effect is improved, the heat loss is reduced, and the air around the device is not heated greatly.

One of the usage modes is as follows:

The collection cycle should be fixed, for example, for 3 hours, by the impact collector collecting gas from ambient air at a rate of 60m ³/h to impact pollen onto the slide (of course, there may be no pollen in the air, and no pollen on the slide).

The rotary sample stage is rotated to the heating module, the glass slide is heated to 90 ℃, and the viscous surface on the glass slide is melted at high temperature to enable particles to sink into a medium, so that the positions of the particles are fixed and adjusted, and the subsequent observation and focusing of an electron microscope are facilitated. The rotation can be manually controlled or automatically controlled by control equipment (the user installs the device by himself) such as a PLC, and the heating temperature needs to be obtained by the user through limited experiments.

The heated slide, which is the rotating specimen mount after heating, reaches the observation position, where it can be observed and recorded by an electron microscope. After recording is completed, the slide with the pollen sample fixed can be taken out and placed into a storage box. And then carrying out graphic stack analysis so as to realize pollen classification.

The foregoing embodiments and description have been provided merely to illustrate the principles and best modes of carrying out the invention, and various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. The pollen monitoring method is characterized by comprising the following steps of:

1) Sampling air;

2) Separating particles in the air according to the size;

3) Fixing the sampled particles on a glass slide;

4) Imaging by a particle sampling microscope;

5) Analyzing images;

6) Storing the processed slide;

7) The next sampling is carried out, and the analysis result is transmitted to a server according to the set time interval;

In the step 1, surrounding air is collected by using a virtual impactor and is impacted on a glass slide, gelatin and glycerin are coated on the glass slide, and pollen particles in the air are adhered by the glycerin, so that sampling is realized;

in the step 2, gas is collected from the surrounding air at a speed of 60 m/h through a virtual impactor, the virtual impactor takes the central air flow, the virtual impactor firstly separates particles with different sizes in the air, and then the particles with the particle size which accords with the pollen particle size are pushed to a glass slide;

in the step 3, the gelatin is melted by heating, pollen particles which are adhered by glycerol are immersed in the gelatin, and then the pollen is permanently fixed in the gelatin by cooling;

scanning and shooting a sample on a glass slide by using an electron microscope to obtain an image stack and storing the image stack in a hard disk;

The glass slide and the electron microscope can rotate for a certain degree to facilitate shooting of samples at a plurality of angles, the electron microscope shoots the samples at 120 positions on each glass slide, the shooting results are separated again according to target particles by using image preprocessing and spliced into an image stack, and the image stack is stored in a hard disk;

In the step 5, the image stack obtained in the step 4 is exported from a hard disk for image analysis, and pollen classification and identification are further carried out on the image stack through big data and a machine learning algorithm, so that pollen classification and statistics counting are realized on the basis;

The method is implemented by adopting a pollen monitoring device, the pollen monitoring device comprises a rotating device, a rotating sample table, a glass slide and a heating module, the rotating device is arranged on a base, a rotating shaft of the rotating device is vertically upwards arranged, the top of the rotating shaft is connected with the rotating sample table, a plurality of grooves are formed in the rotating sample table, the grooves are round, through holes are formed in the bottoms of the grooves, the glass slide is detachably placed in the grooves, a gelatin layer is arranged on the glass slide, a glycerin layer is arranged on the gelatin layer, the heating module is further arranged on the base, and the heating end of the heating module corresponds to the groove of the rotating sample table, so that the glass slide can be heated by the heating end of the heating module when the grooves rotate above the heating module.