Pollen viability detection model construction method, pollen viability detection method, pollen viability detection device and pollen viability detection equipment

Abstract

The invention relates to the technical field of electronics, and discloses pollen viability detection model construction and pollen viability detection methods, devices and equipments.The pollen viability detection model construction method comprises the steps that pollen particle detection is conducted on pollen image data in a pollen data set, and multiple pieces of target pollen particle data in the corresponding pollen image data are obtained; performing visual feature extraction on each piece of target pollen particle data to obtain target visual index data corresponding to the target pollen particles, constructing an associated data set according to the target visual index data of each piece of target pollen particle data and the vitality label data, and training a preset machine learning model by using the associated data set, and finally, a pollen viability detection model is obtained, and the pollen viability detection model can output corresponding viability label data according to the target visual index data of the single pollen, so that rapid and accurate detection of the pollen viability is realized.

Description

Technical Field

[0001] The present invention relates to the field of electronic technology, and in particular to pollen viability detection model construction and pollen viability detection methods, devices and equipment. Background Art

[0002] Pollen viability, a key indicator of reproductive capacity and genetic potential, is directly related to pollination success, seed set, and ultimately crop yield. The close correlation between pollen viability and breeding is particularly noteworthy. Poor pollen viability directly hinders breeding progress and affects the selection and promotion of superior varieties. Therefore, strengthening pollen viability monitoring and assessment is crucial for improving breeding standards and ensuring stable and efficient corn production.

[0003] Researchers and agricultural practitioners rely on traditional pollen viability testing methods, such as triphenyltetrazolium chloride (TTC), acetic acid carmine staining, I2-KI staining, and pollen tube germination. These methods involve treating pollen and visually assessing its viability. While these methods can provide a certain degree of pollen viability, they suffer from limitations such as cumbersome procedures, time-consuming processes, and subjective interpretation of results, making them unable to meet the urgent needs of modern agriculture for efficient and precise management. Summary of the Invention

[0004] In view of this, the present invention provides a pollen viability detection model construction and a pollen viability detection method, device and equipment to solve the problem of low detection efficiency in traditional pollen viability detection methods.

[0005] In a first aspect, the present invention provides a method for constructing a pollen vitality detection model, the method comprising: obtaining multiple pollen image data sets, the pollen image data sets comprising multiple pollen image data; performing pollen particle detection on each pollen image data to obtain multiple target pollen particle data in the corresponding pollen image data; obtaining vitality label data corresponding to different target pollen particle data in each pollen image data; performing visual feature extraction on each target pollen particle data using preset analysis software to obtain target visual index data corresponding to the target pollen particle data; associating the target visual index data of each target pollen particle data in each pollen image data with the vitality label data to obtain an associated data set; and training a preset machine learning model using the associated data set until the model accuracy meets preset conditions to obtain a pollen vitality detection model.

[0006] The pollen vitality detection model construction method provided by the present invention performs pollen particle detection on each pollen image data in a pollen data set to obtain multiple target pollen particle data in the corresponding pollen image data, performs visual feature extraction on each target pollen particle data to obtain target visual index data of the corresponding target pollen particle, constructs an associated data set based on the target visual index data and vitality label data of each target pollen particle data, and uses the associated data set to train a preset machine learning model to finally obtain a pollen vitality detection model. The pollen vitality detection model can output corresponding vitality label data based on the target visual index data of a single pollen grain, thereby realizing rapid and accurate detection of pollen vitality, and solving the problem of low detection efficiency in pollen vitality detection performed by traditional pollen vitality detection methods in related technologies.

[0007] In an optional embodiment, the step of performing pollen grain detection on each pollen image data to obtain multiple target pollen grain data in the corresponding pollen image data includes: using a preset feature detection algorithm to detect each pollen image data to obtain multiple first pollen grain data in the corresponding pollen image data; performing contour optimization on each first pollen grain data to obtain second pollen grain data; performing denoising and boundary smoothing processing on each second pollen grain data to obtain third pollen grain data; and performing segmentation processing on different third pollen grain data of each pollen image data to obtain multiple target pollen grain data corresponding to the pollen image data.

[0008] The method provided in this optional embodiment detects pollen images, identifies pollen grain data in the pollen images, optimizes the contours of the pollen grain data, thereby restoring and optimizing the true contours of the pollen grains, performs denoising and boundary smoothing processing on the pollen grain data after contour optimization, thereby effectively removing noise from the image data, making the boundaries of the pollen grains smoother, and improving the overall quality of the image, and performs segmentation processing on the pollen grain data after denoising and boundary smoothing processing to separate closely connected pollen grains, so that each pollen grain is independently marked, providing an accurate basis for subsequent pollen analysis and counting.

[0009] In an optional embodiment, the step of using preset analysis software to extract visual features of each target pollen grain data to obtain target visual index data corresponding to the target pollen grain data includes: using the preset analysis software to extract visual features of each target pollen grain data to obtain a first number of visual indicators corresponding to the pollen grain data; performing correlation analysis on each visual indicator and target pollen grain data of different categories to obtain correlation values ​​between the corresponding visual indicators and the target pollen grain categories, and the vitality label data corresponding to target pollen grain data of different categories are different; based on the correlation values ​​of each visual indicator, the first number of visual indicators are screened to obtain a second number of visual indicators; and the second number of visual indicators are used as target visual indicator data.

[0010] The method provided in this optional embodiment performs correlation analysis on each visual indicator and target pollen grain data of different categories to obtain correlation values ​​between the corresponding visual indicators and the target pollen grain categories. Based on the correlation values, machine vision indicators with high contribution are screened out, providing accurate training data for the subsequent construction of a pollen vitality detection model.

[0011] In an optional embodiment, the method further includes: obtaining a preset test set; testing the pollen vigor detection model using the preset test set to obtain test results; and evaluating the pollen vigor detection accuracy of the pollen vigor detection model based on the test results.

[0012] In a second aspect, the present invention also provides a pollen vitality detection method, which includes: obtaining target visual index data of each target pollen particle data in the image data of the pollen to be detected; inputting the target visual index data of each pollen particle data in the image data of the pollen to be detected into a pre-constructed pollen vitality detection model, so that the pollen vitality detection model outputs vitality label data corresponding to the target pollen particle data in the pollen image data to be detected, and the pollen vitality detection model is constructed according to the pollen vitality detection model construction method of the above-mentioned first aspect or any corresponding embodiment thereof; determining the vitality information of the pollen to be detected based on the vitality label data corresponding to different target pollen particle data in the image data of the pollen to be detected.

[0013] The pollen viability detection method provided by the present invention inputs the target visual indicator data of each target pollen particle data into a pre-built pollen viability detection model. The pollen viability detection model outputs the viability label data corresponding to each target pollen particle data. The viability information of the pollen to be detected is determined based on the viability label data corresponding to different target pollen particle data, thereby realizing rapid and accurate detection of pollen viability, and solving the problem of low detection efficiency of pollen viability detection using traditional pollen viability detection methods in related technologies.

[0014] In a third aspect, the present invention provides a pollen vitality detection model construction device, which includes: a first acquisition module for acquiring multiple pollen image data sets, the pollen image data sets including multiple pollen image data; a detection module for performing pollen particle detection on each pollen image data to obtain multiple target pollen particle data in the corresponding pollen image data; a second acquisition module for obtaining vitality label data corresponding to different target pollen particle data in each pollen image data; an extraction module for performing visual feature extraction on each target pollen particle data using preset analysis software to obtain target visual index data corresponding to the target pollen particle data; an association module for associating the target visual index data of each target pollen particle data in each pollen image data with the vitality label data to obtain an associated data set; a training module for training a preset machine learning model using the associated data set until the model accuracy meets the preset conditions to obtain a pollen vitality detection model.

[0015] In a fourth aspect, the present invention provides a pollen vitality detection model construction device, which includes: a third acquisition module, used to obtain target visual index data of each target pollen particle data in the image data of the pollen to be detected; a determination module, used to input the target visual index data of each pollen particle data in the image data of the pollen to be detected into a pre-constructed pollen vitality detection model, so that the pollen vitality detection model outputs vitality label data corresponding to the target pollen particle data in the pollen image data to be detected, and the pollen vitality detection model is constructed according to the pollen vitality detection model construction method of the above-mentioned first aspect or any corresponding embodiment; the vitality information of the pollen to be detected is determined based on the vitality label data corresponding to different target pollen particle data in the image data of the pollen to be detected.

[0016] In a fifth aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to thereby execute the pollen viability detection model construction method of the above-mentioned first aspect or any corresponding embodiment thereof, or execute the pollen viability detection method of the above-mentioned second aspect.

[0017] In a sixth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the pollen viability detection model construction method of the first aspect or any corresponding embodiment thereof, or to execute the pollen viability detection method of the second aspect.

[0018] In a seventh aspect, the present invention provides a computer program product comprising computer instructions, the computer instructions being used to enable a computer to execute the pollen viability detection model construction method of the first aspect or any corresponding embodiment thereof, or to execute the pollen viability detection method of the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 is a flow chart of a method for constructing a pollen viability detection model according to an embodiment of the present invention; Figure 2 is a flow chart of another method for constructing a pollen viability detection model according to an embodiment of the present invention; Figure 3 is a schematic diagram of a specific example of a pollen microscopic image segmentation algorithm according to an embodiment of the present invention; Figure 4 is a schematic diagram of a machine vision mechanism process for single pollen grain detection according to an embodiment of the present invention; Figure 5 is a schematic flow chart of a pollen viability detection method according to an embodiment of the present invention; Figure 6 is a structural block diagram of a pollen vitality detection model construction device according to an embodiment of the present invention; Figure 7 is a structural block diagram of a pollen vitality detection model construction device according to an embodiment of the present invention; Figure 8 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0022] In related technologies, researchers and agricultural practitioners rely on traditional pollen viability detection methods, such as TTC (triphenyltetrazolium chloride) staining, acetic acid magenta staining, I2-KI staining and pollen tube germination. Although these methods can reflect the activity state of pollen to a certain extent, they generally have limitations such as cumbersome operation, long time consumption and strong subjectivity in the interpretation of results. They are difficult to meet the urgent needs of modern agriculture for efficient and precise management.

[0023] In view of this, a pollen vitality detection model construction method provided in an embodiment of the present application can be applied to a server to realize the construction of a pollen vitality detection model. The method provided in an embodiment of the present application performs pollen grain detection on each pollen image data in a pollen dataset to obtain multiple target pollen grain data in the corresponding pollen image data, performs visual feature extraction on each target pollen grain data to obtain target visual index data of the corresponding target pollen grain, constructs an associated dataset based on the target visual index data and vitality label data of each target pollen grain data, and uses the associated dataset to train a preset machine learning model to finally obtain a pollen vitality detection model. The pollen vitality detection model can output corresponding vitality label data based on the target visual index data of a single pollen grain, thereby realizing rapid and accurate detection of pollen vitality, solving the problem of low detection efficiency of pollen vitality detection using traditional pollen vitality detection methods in the related art.

[0024] According to an embodiment of the present invention, an embodiment of a method for constructing a pollen viability detection model is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0025] In this embodiment, a method for constructing a pollen vitality detection model is provided, which can be used in the above-mentioned server. Figure 1 FIG. 1 is a flow chart of a method for constructing a pollen viability detection model according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps: Step S101: Acquire multiple pollen image datasets, where the pollen image datasets include multiple pollen image data.

[0026] Exemplarily, the pollen image data may be a microscopic image of pollen obtained by processing a pollen sample using a preset pollen determination method. The preset pollen determination method may include, but is not limited to, triphenyltetrazolium chloride (TTC) staining. In this embodiment of the present application, the pollen may be corn pollen. Corn plants that are robust, pest-free, and have begun to shed pollen at one-third of the tassel are selected as sampling plants. 0.01g of pollen is then removed for pollen viability measurement. Pollen viability is measured using TTC staining, which has been proven to be the simplest, fastest, and most accurate traditional pollen viability testing method. A small amount of dye solution is applied using a rubber-tipped dropper and dropped onto a glass slide. A microscopic image of the stained pollen is then captured using a microscope system. A microscopic image of the stained pollen is captured using a portable microscope system. Three images are taken in three different fields of view, corresponding to the pollen image data of the pollen sample.

[0027] Step S102 : performing pollen particle detection on each pollen image data to obtain a plurality of target pollen particle data in the corresponding pollen image data.

[0028] For example, in an embodiment of the present application, in order to remove complex backgrounds to enhance application scenarios and improve the detection results of high-density and low-quality pollen images, the machine vision mechanism of pollen detection is analyzed, and the pollen grain images in each pollen image data are detected to obtain the target pollen grain data of each pollen.

[0029] Step S103: Obtaining vitality label data corresponding to different target pollen particle data in each pollen image data.

[0030] For example, the viability label data for each target pollen grain can be obtained through manual annotation, with different viability label data used to characterize different viability states of the pollen. In the embodiments of the present application, the pollen viability label data may include, but is not limited to, three types of data: high-vigor pollen (HVP), low-vigor pollen (LVP), and non-viable pollen (NVP). The embodiments of the present application do not limit the specific content of the pollen viability label data, and those skilled in the art can determine it as needed.

[0031] Step S104 : Using preset analysis software, visual feature extraction is performed on each target pollen grain data to obtain target visual index data corresponding to the target pollen grain data.

[0032] Exemplarily, the target visual indicator data is used to characterize the visual feature information of pollen particles. The preset analysis software may include but is not limited to the Zhinong Cloud Core Seed Intelligent Analysis System (AIseed). Its core function is to achieve high-throughput phenotypic extraction, quality inspection and processing parameter optimization of seeds through artificial intelligence technology, and is widely used in agricultural scientific research, seed production and other fields. In the embodiment of the present application, AIseed software is used to extract machine vision indicators such as pollen morphology, color and texture. Morphological indicators include length (Length), width (Width), length-to-width ratio (LWR), area (Area), perimeter (Perimeter), and roundness (Roundness). Color metrics include red channel mean (Rmean), red channel standard deviation (Rstd), green channel mean (Gmean), green channel standard deviation (Gstd), blue channel mean (Bmean), blue channel standard deviation (Bstd), L channel mean (Imean), L channel standard deviation (Istd), a channel mean (amean), a channel standard deviation (astd), b channel mean (bmean), b channel standard deviation (bstd), hue mean (Hmean), hue standard deviation (Hstd), saturation mean (Smean), saturation standard deviation (Sstd), lightness mean (V_mean), lightness standard deviation (Vstd), grayscale mean (Graymean), and grayscale standard deviation (Graystd). Texture metrics include red channel R, green channel G, blue channel B, and grayscale contrast (Con), dissimilarity (Dis), homogeneity (Hom), energy (Ene), correlation (Cor), angular second moment (ASM), and entropy (Ent), for a total of 54 machine vision metrics. This embodiment of the application does not limit the number and specific content of machine vision metrics. Those skilled in the art can determine them based on their needs, as long as they are reasonable.

[0033] Step S105 , associating the target visual index data of each target pollen grain data in each pollen image data with the vitality label data to obtain an associated data set.

[0034] For example, the embodiment of the present application does not limit the manner of associating the target visual indicator data with the vitality tag data, and those skilled in the art can determine it according to their needs, as long as it is reasonable.

[0035] Step S106: Use the associated data set to train the preset machine learning model until the model accuracy meets the preset conditions, thereby obtaining a pollen viability detection model.

[0036] For example, the preset machine learning models may include, but are not limited to, convolutional neural networks (CNNs), support vector machines (SVMs), long short-term memory networks (LSTMs), and optimized ensemble learning (OEL) algorithms. The preset conditions may be determined based on past experience.

[0037] In this embodiment, the Kennard-Stone algorithm was used to select samples from a machine learning dataset in a 3:1 ratio, splitting it into training and test sets. The Kennard-Stone algorithm uses Euclidean distance as a metric to extract the most representative sample points from the dataset. Different algorithms can produce significantly different results due to their different principles. A CNN primarily consists of an input layer, a convolutional layer, an activation layer, a pooling layer, and a fully connected layer. The convolutional layer extracts local features from the input image, the activation layer adds nonlinearity, the pooling layer performs downsampling to reduce dimensionality and computational complexity, and finally, the fully connected layer combines the outputs of the previous layers to output the classification result. Its primary advantage is its powerful feature extraction capabilities, with the convolutional layer automatically extracting hierarchical features from the input data. An LSTM consists of a forget gate, an input gate, and an output gate. The forget gate controls the degree to which the previous cell state is forgotten, the input gate controls how much new information is incorporated, and the output gate controls how much of the current cell state is filtered before output. These gating mechanisms enable LSTM to selectively memorize and forget information, effectively processing long sequences of data and effectively addressing the vanishing gradient problem. Support Vector Machine (SVM) is a classic machine learning classification method based on statistical learning theory. Its goal is to find an optimal hyperplane to separate two types of data. By introducing a kernel function to map data into a high-dimensional space and then searching for the optimal hyperplane within that space, it effectively processes high-dimensional data. Its structure is relatively simple and computationally inefficient. OEL is an algorithm based on bagging and boosting. By constructing and combining multiple machine learning models to complete learning tasks, it achieves significantly better generalization performance than a single model. Combined with other ensemble learning methods, it significantly reduces model complexity and improves model efficiency.

[0038] The pollen viability detection model construction method provided in this embodiment performs pollen grain detection on each pollen image data in a pollen dataset to obtain multiple target pollen grain data in the corresponding pollen image data, performs visual feature extraction on each target pollen grain data to obtain target visual index data of the corresponding target pollen grain, constructs an associated dataset based on the target visual index data and vitality label data of each target pollen grain data, and uses the associated dataset to train a preset machine learning model to ultimately obtain a pollen viability detection model. The pollen viability detection model can output corresponding vitality label data based on the target visual index data of a single pollen grain, thereby achieving rapid and accurate detection of pollen viability, and solving the problem of low detection efficiency in pollen viability detection using traditional pollen viability detection methods in related technologies.

[0039] In this embodiment, a method for constructing a pollen vitality detection model is provided, which can be used in the above-mentioned server. Figure 2 FIG. 1 is a flow chart of a method for constructing a pollen viability detection model according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps: Step S201: Acquire multiple pollen image datasets, where the pollen image datasets include multiple pollen image data. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0040] Step S202 : performing pollen particle detection on each pollen image data to obtain a plurality of target pollen particle data in the corresponding pollen image data.

[0041] Specifically, the above step S202 includes: Step S2021: Detect each pollen image data using a preset feature detection algorithm to obtain a plurality of first pollen grain data corresponding to the pollen image data.

[0042] Exemplarily, the preset feature detection algorithm may include but is not limited to the circular Hough transform. In the embodiment of the present application, the circular area in the pollen image data is detected as pollen grains based on the circular Hough transform, and a plurality of first pollen grain data are determined. The circular Hough transform is performed using the image circle detection (imfindcircles) function. This function is based on the edge information in the image, and performs cumulative voting in the parameter space by calculating the gradient direction and position of the edge point. When the number of votes obtained by a combination of the center coordinates and the radius of a circle exceeds a preset threshold, it is considered that a circle is detected. Figure 3 As shown in part a of FIG, after the circular Hough transform detection, all the approximately circular pollen in the microscopic image are preliminarily marked. The present embodiment does not limit the specific content of the preset threshold value, and those skilled in the art can determine it according to needs.

[0043] Step S2022: Optimize the contour of each first pollen grain data to obtain second pollen grain data.

[0044] For example, in the embodiment of the present application, the circular area detected by the circular Hough transform is used as the initial contour of the active contour model. Then, the external force is defined based on the edge information, gradient information, etc. of the image to move the contour toward the true boundary of the pollen; at the same time, the internal force is defined to maintain the smoothness and continuity of the contour. The position of the contour is updated iteratively until the energy function reaches the minimum value. Figure 3 As shown in part b of the figure, after active contour model optimization, foreign matter that does not belong to pollen (such as dust and debris) is removed, and the true contour of the pollen grain is restored and optimized, ensuring the accuracy and smoothness of the contour.

[0045] Step S2023: Denoise and smooth the edges of each second pollen grain data to obtain third pollen grain data.

[0046] For example, in the embodiment of the present application, the image optimized by the active contour model is subjected to morphological processing by first dilation and then erosion. The dilation operation can expand the boundaries of the pollen grains outward and fill some small depressions; the erosion operation can remove some isolated noise points and make the boundaries smoother. Figure 3 As shown in part c of the figure, after morphological processing, the noise in the image is effectively removed, the boundaries of the pollen particles are smoother, and the overall quality of the image is improved.

[0047] Step S2024 : Segment the different third pollen grain data of each pollen image data to obtain a plurality of target pollen grain data corresponding to the pollen image data.

[0048] For example, in the embodiment of the present application, the gradient calculation is performed on the morphologically processed image to obtain the edge information of the image. Then, the gradient image is regarded as a topographic map, and different water basins are formed by continuously injecting water into low-lying areas. When adjacent water basins meet, a watershed line is constructed to divide the image into different areas. Figure 3 As shown in part d in Figure 3, after segmentation using the watershed algorithm, the closely connected pollen grains were successfully segmented, and each pollen grain was independently marked, providing an accurate basis for subsequent pollen analysis and counting.

[0049] In the embodiment of the present application, the algorithm for performing a series of processing on each pollen image data to obtain data corresponding to multiple target pollen particles can be defined as a pollen microscopic image segmentation algorithm (PMISA). This algorithm can remove complex backgrounds and segment connected pollen while retaining the morphological structure of the pollen to the greatest extent. PMISA is integrated into a functional function for easy use. The parameters used by the PMISA algorithm are shown in Table 1. In addition, PMISA is compared with advanced segmentation algorithms (Segment Anything Model, SAM) and Nested U-Net models. SAM is a large visual model based on prompts that can segment any target in a given image and has strong zero-sample generalization capabilities. Nested U-Net enhances the ability of feature extraction and fusion by introducing mechanisms such as nested convolution blocks and jump connections, and has excellent performance in processing high-resolution images and images with complex structures.

[0050] Table 1

[0051] Step S203: Obtain vitality label data corresponding to different target pollen grain data in each pollen image data.

[0052] Step S204 : Using preset analysis software, visual feature extraction is performed on each target pollen grain data to obtain target visual index data corresponding to the target pollen grain data.

[0053] Specifically, the above step S204 includes: Step S2041 : Using preset analysis software, visual features are extracted from each target pollen grain data to obtain a first quantity of visual indicators corresponding to the pollen grain data.

[0054] Exemplarily, the preset analysis software may be AIseed software. For the specific visual feature extraction process, please refer to the description of the relevant content in the above embodiment, which will not be repeated here.

[0055] Step S2042 , performing correlation analysis on each visual indicator and target pollen grain data of different categories to obtain correlation values ​​between the corresponding visual indicators and target pollen grain categories. Target pollen grain data of different categories have corresponding vitality label data that is different.

[0056] For example, in the embodiments of the present application, in order to screen machine vision indicators with high contribution, the statistical analysis software (Statistical Package for the Social Sciences 23, SPSS23) was used to perform Spearman correlation analysis on the machine vision indicators of pollen and different categories of pollen to obtain the correlation value between each visual indicator and the pollen particle category.

[0057] Step S2043: Filter the first number of visual indicators based on the correlation value of each visual indicator to obtain a second number of visual indicators.

[0058] Illustratively, in an embodiment of the present application, among the first number of visual indicators, visual indicators with correlation values ​​less than a preset threshold are eliminated to obtain a second number of visual indicators.

[0059] Step S2044: using the second number of visual indicators as target visual indicator data.

[0060] Illustratively, in an embodiment of the present application, the second number of visual indicators is used as visual indicator data for subsequent model construction.

[0061] Step S205: Associating the target visual index data of each target pollen particle data in each pollen image data with the vitality label data to obtain an associated data set. Figure 1 Step S105 of the illustrated embodiment will not be described in detail here.

[0062] Step S206: Use the associated data set to train the preset machine learning model until the model accuracy meets the preset conditions, thereby obtaining a pollen vitality detection model. Figure 1 Step S106 of the illustrated embodiment will not be described in detail here.

[0063] In some optional embodiments, the method further comprises: Step a1: Obtain a preset test set.

[0064] For example, in an embodiment of the present application, the preset test set can be obtained by dividing the associated data set. The embodiment of the present application does not limit the specific content of the preset test set, and those skilled in the art can determine it according to needs.

[0065] Step a2: Testing the pollen vitality detection model using a preset test set to obtain test results.

[0066] Illustratively, in an embodiment of the present application, the target visual index data of each pollen in a preset test set is input into a pollen vitality detection model, and the test results are obtained according to the output of the model.

[0067] Step a3: Evaluate the pollen viability detection accuracy of the pollen viability detection model based on the test results.

[0068] For example, in the embodiments of this application, the data processing environment is as follows: Hardware: Processor: AMD Ryzen 76800H with Radeon Graphics 3.20 GHz, 16GB of RAM. Software: Python 3.9, PyTorch 2.0.1, MATLAB 2023b; Operating system: Windows 11.

[0069] To evaluate the model's overall classification performance for all pollen categories, the Accuracy metric is used. To evaluate the model's performance for classifying a single pollen category, the Precision and Recall metrics are used. To evaluate the model's comprehensive classification performance for a single pollen category, the F1 score metric is used. These metrics are derived from the confusion matrix, which records the consistency between the actual category and the predicted category. To evaluate whether pollen can be detected from the background, the detection rate is used. The formula is as follows:

[0070]

[0071]

[0072]

[0073]

[0074] Here, "TP" refers to the number of samples classified as positive with a label of positive. "TN" refers to the number of samples classified as negative with a label of negative. "FP" refers to the number of samples classified as positive with a label of negative. "FN" refers to the number of samples classified as negative with a label of positive. "TT" refers to the total number of labels.

[0075] The method provided in this application example, the machine vision mechanism process of single pollen detection is as follows Figure 4 The original pollen image at high density and the image after PMISA segmentation are shown in Figure 4As shown in parts a and b of the figure, the analysis results show that the algorithm achieved a 98.7% accuracy rate in correctly segmenting pollen, meaning that only 1.3% of pollen samples were unsegmented or incorrectly segmented, fully demonstrating the algorithm's high precision and reliability. Notably, the algorithm achieved a 99.8% pollen detection rate, ensuring that nearly all pollen was effectively extracted, laying the foundation for subsequent work. Compared to the SAM and Nested U-Net models, while all three had detection rates exceeding 99%, PMISA achieved significantly higher accuracy. Nested U-Net had the lowest accuracy, at only 82.7%, primarily due to the inability to separate many connected pollen grains, making it unsuitable for pollen segmentation. Finally, machine vision metrics for 3,000 pollen grains (1,000 grains of each type) were extracted using AIseed software.

[0076] Table 2

[0077] Furthermore, in order to analyze the machine vision mechanism of pollen detection, Spearman correlation analysis was performed between the extracted machine vision indicators and the three types of pollen. The results of the correlation analysis between size, color and texture indicators and pollen vitality are as follows: Figure 4 As shown in part c of Figure 2. In terms of size characteristics, although all size indicators are correlated with pollen vitality, the correlation coefficient of length (-0.222) is the most prominent, and its probability density distribution is shown in the figure below. Figure 4 As shown in the e part of the figure. However, the overall discrimination of size indicators is relatively low, suggesting that pollen with different vitality states do not differ significantly in morphology. In terms of color characteristics, color characteristics perform well in distinguishing pollen vitality. Among them, the correlation of green mean (Gmean) is as high as 0.833, far exceeding other color indicators, becoming the strongest color factor for evaluating pollen vitality. Its probability density distribution is shown in the figure below. Figure 4 This finding not only highlights the core value of color information in pollen vitality detection, but also emphasizes the unique importance of green components as an indicator of pollen health status; texture features also provide valuable information, especially the gray homogeneity (GrayHom) index, whose correlation coefficient of 0.677 proves its key role in evaluating pollen vitality. Its probability density distribution is shown in the figure below. Figure 4 As shown in section g of the figure, although slightly inferior to color features, texture features still serve as a complementary tool, enhancing the comprehensiveness and accuracy of the overall assessment system. Furthermore, based on the above analysis, we selected machine vision indicators with significant correlations greater than 0.5 to construct a more efficient and accurate pollen viability assessment model.

[0078] In order to compare the overall performance of different models, CNN, LSTM, SVM and OEL networks were tested on the same test set in the same test environment, as shown in Table 3.

[0079] Table 3

[0080] The results show that under the high-density pollen dataset (HD), LSTM has the best overall performance with an accuracy of 98.5%. The confusion matrix is ​​as follows Figure 4 As shown in part d of the figure, the CNN model performed second best, with an accuracy of 87.9%, while the OEL model had the worst overall performance. On the low-quality pollen dataset (LQ), the LSTM model achieved the best overall performance, with an accuracy of 96.5%. The CNN model performed second best, with an accuracy of 88.8%, while the OEL model had the worst overall performance. To further compare the classification performance of different models for individual pollen viability categories, the models were evaluated using the precision, recall, and F1-score metrics. On the high-density pollen dataset, for HVP classification, the LSTM model achieved the highest F1-score of 99.1%, followed by the CNN model with an F1-score of 88.4%. For LVP classification, the LSTM model achieved the highest F1-score of 97.8%, followed by the CNN model with an F1-score of 81.9%. For NVP classification, the LSTM model achieved the highest F1-score of 98.7%, followed by the CNN model with an F1-score of 92.1%. In a low-quality pollen dataset, the LSTM model achieved the highest F1-Score of 98.6% for HVP classification, followed by the CNN model with an F1-Score of 96.5%. For LVP classification, the LSTM model achieved the highest F1-Score of 88.7%, followed by the CNN model with an F1-Score of 72.8%. For NVP classification, the LSTM model achieved the highest F1-Score of 95.6%, followed by the SVM model with an F1-Score of 88.7%. The main error rate among the three models occurred in LVP, primarily due to LVP being an intermediate class with some similarities to HVP and NVP, further highlighting the challenge of borderline samples. In summary, deep learning models outperformed machine learning models in pollen detection, with the LSTM model in particular demonstrating excellent performance, achieving evaluation metrics above 95% for all three pollen types.

[0081] In this embodiment, a pollen vitality detection method is provided, which can be used in the above-mentioned server. Figure 5 FIG. 1 is a flow chart of a pollen viability detection method according to an embodiment of the present invention. Figure 5 As shown, the process includes the following steps: Step S501 : obtaining target visual index data of each target pollen particle in the image data of the pollen to be detected.

[0082] Illustratively, the image data of the pollen to be detected may be pollen image data requiring pollen vitality detection. The specific process of determining the target visual index data is described in the above embodiments and will not be repeated here.

[0083] In step S502, the target visual indicator data of each pollen particle in the pollen image data to be detected is input into a pre-constructed pollen viability detection model, so that the pollen viability detection model outputs viability label data corresponding to the target pollen particle data in the pollen image data to be detected. The pollen viability detection model is constructed according to the pollen viability detection model construction method of the above embodiment.

[0084] For example, the construction process of the pollen vigor detection model is described in the above embodiment, which will not be repeated here. The vigor label data is used to characterize the vigor of the pollen grains.

[0085] Step S503 : determining the vitality information of the pollen to be detected based on the vitality label data corresponding to different target pollen particle data in the image data of the pollen to be detected.

[0086] For example, in an embodiment of the present application, the vitality information of the pollen to be detected is determined based on the vitality label data corresponding to different target pollen particle data. The vitality information can be used to evaluate the overall vitality status of the pollen to be detected. The embodiment of the present application does not limit the specific determination process of the vitality information, and those skilled in the art can determine it according to needs.

[0087] The pollen viability detection method provided in this embodiment inputs the target visual indicator data of each target pollen particle data into a pre-built pollen viability detection model. The pollen viability detection model outputs the viability label data corresponding to each target pollen particle data. The viability information of the pollen to be detected is determined based on the viability label data corresponding to different target pollen particle data, thereby achieving rapid and accurate detection of pollen viability. This solves the problem of low detection efficiency of pollen viability detection using traditional pollen viability detection methods in related technologies.

[0088] This embodiment also provides a pollen viability detection model construction device for implementing the aforementioned embodiments and preferred embodiments. Details already described will not be repeated. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. While the devices described in the following embodiments are preferably implemented using software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0089] This embodiment provides a pollen vitality detection model construction device, such as Figure 6 Shown, including: A first acquisition module 601 is configured to acquire a plurality of pollen image datasets, wherein the pollen image datasets include a plurality of pollen image data; The detection module 602 is used to perform pollen particle detection on each pollen image data to obtain a plurality of target pollen particle data in the corresponding pollen image data; The second acquisition module 603 is used to obtain vitality label data corresponding to different target pollen grain data in each pollen image data; Extraction module 604, configured to extract visual features from each target pollen grain data using preset analysis software to obtain target visual index data corresponding to the target pollen grain data; An association module 605 is configured to associate the target visual index data of each target pollen particle data in each pollen image data with the vitality label data to obtain an associated data set; The training module 606 is used to train the preset machine learning model using the associated data set until the model accuracy meets the preset conditions, thereby obtaining a pollen vitality detection model.

[0090] In some optional implementations, the detection module 602 includes: a detection submodule, configured to detect each pollen image data using a preset feature detection algorithm to obtain a plurality of first pollen particle data corresponding to the pollen image data; an optimization submodule, configured to perform contour optimization on each first pollen grain data to obtain second pollen grain data; A first processing submodule is used to perform denoising and boundary smoothing processing on each second pollen grain data to obtain third pollen grain data; The second processing submodule is used to segment the different third pollen grain data of each pollen image data to obtain a plurality of target pollen grain data corresponding to the pollen image data.

[0091] In some optional implementations, the extraction module 604 includes: an extraction submodule, configured to extract visual features from each target pollen grain data using preset analysis software to obtain a first quantity of visual indicators corresponding to the pollen grain data; The analysis submodule is used to perform correlation analysis between each visual indicator and target pollen grain data of different categories to obtain the correlation value between the corresponding visual indicator and the target pollen grain category. Different categories of target pollen grain data correspond to different vitality label data. a screening submodule, configured to screen the first number of visual indicators based on correlation values ​​of the visual indicators to obtain a second number of visual indicators; The determination submodule is configured to use the second number of visual indicators as target visual indicator data.

[0092] In some optional embodiments, the above device further includes: The fourth acquisition module is used to obtain a preset test set; A testing module is used to test the pollen vitality detection model using a preset test set to obtain test results; The evaluation module is used to evaluate the pollen vitality detection accuracy of the pollen vitality detection model based on the test results.

[0093] This embodiment also provides a pollen vitality detection device, such as Figure 7 As shown, including: The third acquisition module 701 is used to obtain target visual index data of each target pollen particle data in the image data of the pollen to be detected; A first determining module 702 is configured to input target visual indicator data of each pollen particle in the pollen image data to be detected into a pre-built pollen vitality detection model, so that the pollen vitality detection model outputs vitality label data corresponding to the target pollen particle data in the pollen image data to be detected. The pollen vitality detection model is constructed according to the pollen vitality detection model construction method of the above embodiment; The second determining module 703 is configured to determine the vitality information of the pollen to be detected based on the vitality label data corresponding to different target pollen particle data in the image data of the pollen to be detected.

[0094] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0095] The pollen viability detection model construction device and the pollen viability detection device in this embodiment are presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit) circuits, processors and memories that execute one or more software or fixed programs, and / or other devices that can provide the above-mentioned functions.

[0096] The embodiment of the present invention also provides a computer device having the above Figure 6 The pollen vitality detection model construction device shown, or having the above Figure 7 The pollen vitality detection device shown.

[0097] See also Figure 8 , Figure 8 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 8As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of a GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 8 A processor 10 is taken as an example.

[0098] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0099] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0100] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0101] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0102] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0103] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0104] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0105] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for constructing a pollen viability detection model, characterized in that: The method comprises: Acquire a plurality of pollen image data sets, wherein the pollen image data sets include a plurality of pollen image data; Performing pollen particle detection on each pollen image data to obtain a plurality of target pollen particle data in the corresponding pollen image data; Obtaining vitality label data corresponding to different target pollen grain data in each pollen image data; Use the preset analysis software to extract visual features of each target pollen grain data to obtain target visual index data corresponding to the target pollen grain data; Associating the target visual index data of each target pollen particle data in each pollen image data with the vitality label data to obtain an associated data set; The associated data set is used to train a preset machine learning model until the model accuracy meets the preset conditions, thereby obtaining a pollen viability detection model.

2. The method according to claim 1, characterized in that The step of performing pollen particle detection on each pollen image data to obtain a plurality of target pollen particle data corresponding to the pollen image data includes: Detecting each pollen image data using a preset feature detection algorithm to obtain a plurality of first pollen particle data corresponding to the pollen image data; performing contour optimization on each first pollen grain data to obtain second pollen grain data; performing denoising and boundary smoothing processing on each second pollen grain data to obtain third pollen grain data; The different third pollen grain data of each pollen image data are segmented to obtain a plurality of target pollen grain data corresponding to the pollen image data.

3. The method according to claim 1 or 2, characterized in that The steps of extracting visual features from each target pollen grain data using preset analysis software to obtain target visual index data corresponding to the target pollen grain data include: Using preset analysis software to extract visual features from each target pollen grain data to obtain a first quantity of visual indicators corresponding to the pollen grain data; Correlation analysis was performed between each visual indicator and target pollen grain data of different categories to obtain the correlation value between the corresponding visual indicator and the target pollen grain category. Different categories of target pollen grain data correspond to different vitality label data. Filtering the first number of visual indicators based on correlation values ​​of the visual indicators to obtain a second number of visual indicators; The second number of visual indicators is used as target visual indicator data.

4. The method according to claim 3, characterized in that The method further comprises: Get the preset test set; Using the preset test set to test the pollen vitality detection model, and obtain a test result; The pollen viability detection accuracy of the pollen viability detection model is evaluated based on the test results.

5. A pollen viability detection method, characterized in that: The method comprises: Obtaining target visual index data of each target pollen particle data in the image data of the pollen to be detected; inputting target visual indicator data of each pollen particle in the image data of the pollen to be detected into a pre-constructed pollen viability detection model, so that the pollen viability detection model outputs viability label data corresponding to the target pollen particle data in the image data of the pollen to be detected, wherein the pollen viability detection model is constructed according to the pollen viability detection model construction method according to any one of claims 1 to 4; The vitality information of the pollen to be detected is determined based on the vitality label data respectively corresponding to different target pollen particle data in the image data of the pollen to be detected.

6. A pollen vitality detection model construction device, characterized in that: The device comprises: A first acquisition module is used to acquire a plurality of pollen image data sets, wherein the pollen image data sets include a plurality of pollen image data; A detection module is used to perform pollen particle detection on each pollen image data to obtain a plurality of target pollen particle data in the corresponding pollen image data; The second acquisition module is used to obtain the vitality label data corresponding to different target pollen particle data in each pollen image data; An extraction module is used to extract visual features from each target pollen grain data using preset analysis software to obtain target visual index data corresponding to the target pollen grain data; an association module, configured to associate the target visual index data of each target pollen particle data in each pollen image data with the vitality label data to obtain an associated data set; The training module is used to train the preset machine learning model using the associated data set until the model accuracy meets the preset conditions, thereby obtaining a pollen vitality detection model.

7. A pollen vitality detection device, characterized in that: The device comprises: The third acquisition module is used to obtain target visual index data of each target pollen particle data in the image data of the pollen to be detected; a first determining module, configured to input target visual indicator data of each pollen particle in the image data of the pollen to be detected into a pre-constructed pollen viability detection model, so that the pollen viability detection model outputs viability label data corresponding to the target pollen particle data in the image data of the pollen to be detected, wherein the pollen viability detection model is constructed according to the pollen viability detection model construction method according to any one of claims 1 to 4; The second determining module is configured to determine the vitality information of the pollen to be detected based on the vitality label data respectively corresponding to different target pollen particle data in the image data of the pollen to be detected.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the pollen viability detection model construction method according to any one of claims 1 to 4, or the pollen viability detection method according to claim 5, by executing the computer instructions.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, which are used to enable a computer to execute the pollen viability detection model construction method according to any one of claims 1 to 4, or the pollen viability detection method according to claim 5.

10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the method for constructing a pollen viability detection model according to any one of claims 1 to 4, or the method for detecting pollen viability according to claim 5.

Images