Method and system for counting pollen germination rate and germination length of camellia oleifera in vitro

Abstract

The present application relates to a kind of statistical camellia pollen germination rate and germination length method and system, method includes: S1, collection camellia pollen and in vitro culture, obtain the micrograph of camellia pollen tube, the micrograph is divided into training image and to be measured image;S2, different marking mode is handled to training image, obtain the detection label for statistics germination rate and the segmentation label for statistics germination length;S3, using the training image after marking, detection label and segmentation label training obtain the artificial intelligence model for detection and segmentation;S4, to be measured image is input in the artificial intelligence model for detection and segmentation, obtain pollen grain count result and germination pollen tube true length;S5, according to pollen grain count result and germination pollen tube true length, determine and statistics camellia pollen germination rate and germination length.Solve the image processing tedious, workload after collecting camellia pollen tube germination and other technical problems.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of agricultural biology combined with artificial intelligence, and particularly relates to a method and system for statistically determining the germination rate and germination length of oil tea pollen in vitro. BACKGROUND

[0002] Camellia oleifera is one of the most important edible oil tree species in China, widely distributed in southern China. It has rich oil yield, significant economic value, high nutritional and medicinal value, and is a pure natural green health food with antibacterial and anti-inflammatory, antioxidant, and blood sugar and lipid lowering effects. The domestic oil tea planting area exceeds 4.4 million hm 2 , and the annual production of oil tea seed oil is about 600,000 tons. Oil tea is one of the economic tree species with the largest cultivation area and the widest distribution in China.

[0003] In production, oil tea generally has the phenomena of more flowers and less fruits, and low fruit setting rate. An important reason for this phenomenon is that oil tea has the characteristics of late-acting self-incompatibility (LSI), that is, although the self-pollination pollen tube can pass through the style, it finally stops in the ovary and cannot reach the embryo sac to release sperm and complete double fertilization. Therefore, the self-pollination seed setting rate of most varieties is low, and reasonable allocation of pollination trees or artificial auxiliary pollination measures must be taken during the flowering period to obtain the expected yield and quality. The project group previously determined through plant microscopic observation, multi-omics and other technologies that the LSI of oil tea is a process of programmed cell death (PCD) of self-pollination pollen tube, in which abscisic acid (ABA), salicylic acid (SA), jasmonic acid (JA), indole acetic acid (IAA), gibberellic acid (GA3), cytokinin (CTK) and other hormones are involved. Therefore, in order to explore the specific role of plant hormones in the growth process of oil tea pollen tube, the project group carried out oil tea pollen in vitro culture experiments under the action of various exogenous hormones.

[0004] This study used pollen tube germination rate and germination length as reference indicators to determine the effects of different hormones on the growth of Camellia oleifera pollen tubes. To calculate the germination rate and germination length, the growth status of pollen tubes under each treatment was observed and photographed using a microscope. To ensure data uniformity, at least five fields of view were collected for each treatment, with at least 30 pollen cells counted in each field of view. The current experimental method uses the Cell Counter tool in ImageJ software to obtain Camellia oleifera pollen germination rate data, and the line marking function in ImageJ software to obtain the pixel length of Camellia oleifera pollen tubes, which is then converted to the actual length according to the scale bar to obtain the germination length data. Generally, the field of view of a microscope is relatively small, and Camellia oleifera pollen grains are numerous and the pollen tubes are relatively long, so image processing and analysis involves a large amount of tedious and repetitive work, greatly consuming researchers' time and slowing down the research progress. Therefore, there is an urgent need for an efficient and accurate method to statistically analyze the germination rate and germination length of isolated Camellia oleifera pollen. Summary of the Invention

[0005] This invention provides a method and system for statistically analyzing the germination rate and germination length of isolated Camellia oleifera pollen, solving the technical problems of cumbersome and labor-intensive image processing after collecting germinating Camellia oleifera pollen tubes. By adopting a deep learning approach, artificial intelligence technology is used to automatically detect and segment microscopic images of pollen tubes, thereby obtaining data on the germination rate and germination length of pollen after in vitro culture.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] Firstly, a method for statistically analyzing the germination rate and germination length of isolated Camellia oleifera pollen is provided, including:

[0008] S1. Collect camellia pollen and culture it in vitro to obtain microscopic images of camellia pollen tubes. Divide the microscopic images into training images and test images.

[0009] S2, different annotation methods are used to process the training images to obtain detection labels for statistical germination rate and segmentation labels for statistical germination length;

[0010] S3 uses labeled training images, detection labels, and segmentation labels to train an artificial intelligence model for detection and segmentation;

[0011] S4. Input the image to be tested into the artificial intelligence model used for detection and segmentation to obtain the number of germinated pollen grains, the number of non-germinated pollen grains, and the pixel length of the germinated pollen tube. Obtain the pollen grain count result based on the number of germinated and non-germinated pollen grains, and calculate the true length of the germinated pollen tube based on the pixel length of the germinated pollen tube.

[0012] S5, according to the pollen grain count result and the true length of the germinated pollen tube, the germination rate and the germination length of the oil tea pollen are measured and counted.

[0013] Further, in step S1, the oil tea pollen is collected and cultured in vitro to obtain microscopic images of the oil tea pollen tube, and the microscopic images are divided into training images and to-be-measured images, including:

[0014] S11, collecting oil tea pollen and storing it at low temperature;

[0015] S12, spreading the oil tea pollen in the pollen culture dish prepared with the oil tea in vitro pollen suspension culture solution, mixing thoroughly, and then placing it in a constant temperature incubator for dark culture;

[0016] S13, placing the current pollen culture dish under a microscope, adjusting the focal length of the microscope so that the pollen grains and pollen tubes in the current pollen culture dish can be clearly imaged under the microscope, moving the current pollen culture dish to take multiple field images until the microscopic images of all central regions of the current pollen culture dish are collected;

[0017] S14, replacing the pollen culture dish that has not been completed, repeating step S13, and collecting the microscopic images of all pollen culture dishes;

[0018] S15, orderly storing the microscopic images of all pollen culture dishes, and randomly selecting the microscopic images to be divided into training images and to-be-measured images.

[0019] Further, in step S2, different labeling methods are used to process the training images to obtain detection labels for counting the germination rate and segmentation labels for counting the germination length, including:

[0020] S21, using a rectangular frame tool to frame and label the pollen grains in the training images using an image labeling software, setting the label categories of ungerminated pollen grains and germinated pollen grains respectively, and obtaining the detection labels for counting the germination rate;

[0021] S22, using a polygon tool to enclose and label the pollen tubes in the training images using an image labeling software, setting the label categories of the pollen tubes that are clearly imaged and not crossed with other pollen tubes, the label categories of the pollen tubes that are clearly imaged and crossed with other pollen tubes, and the label categories of the pollen tubes that are not clearly imaged or extend beyond the focal length of the lens, and obtaining the segmentation labels.

[0022] Further, in step S3, the artificial intelligence model for detection and segmentation is trained using the labeled training images, detection labels and segmentation labels, including:

[0023] S31, constructing a germination rate detection model for counting the pollen germination rate;

[0024] S32, input the labeled training image and the detection label into the germination rate detection model for training, and save the best detection model weight for outputting the pollen tube detection result;

[0025] S33, construct an instance segmentation model for counting pollen tube length;

[0026] S34, input the labeled training image and the segmentation label into the instance segmentation model for training, and save the best segmentation model weight for outputting the pollen tube segmentation result;

[0027] S35, randomly select part of the to-be-tested images, input them into the models trained in steps S32 and S34 to obtain new detection labels and new segmentation labels; after adjusting the detection labels and the segmentation labels according to the new detection labels and the new segmentation labels, continue to input them into the models in steps S32 and S34 for training;

[0028] S36, repeat step S35 until the germination rate detection model and the instance segmentation model reach the preset model performance, and obtain the artificial intelligence model for detection and segmentation.

[0029] Further, in step S4, input the to-be-tested image into the artificial intelligence model for detection and segmentation to obtain the number of germinated pollen grains, the number of un-germinated pollen grains and the pixel length of the germinated pollen tube, obtain the pollen grain counting result according to the number of germinated pollen grains and the number of un-germinated pollen grains, and calculate the real length of the germinated pollen tube according to the pixel length of the germinated pollen tube, including:

[0030] S41, input the to-be-tested image into the artificial intelligence model for detection and segmentation to obtain the target detection label and the target segmentation label of the to-be-tested image;

[0031] S42, screen the target detection label and the target segmentation label according to the preset label specification;

[0032] S43, count the screened target detection label to obtain the number of germinated pollen grains and the number of un-germinated pollen grains, and obtain the pollen grain counting result according to the number of germinated pollen grains and the number of un-germinated pollen grains;

[0033] S44, select the target pollen tubes with label categories of normal and cross in the screened target segmentation label, draw a mask image of each target pollen tube using a polygon label, and obtain the pixel length of each target pollen tube by processing the mask image using a thinning algorithm;

[0034] S45, according to the proportion of the image pixels of the to-be-tested image and the real length scale, the pixel length is converted to obtain the real length of each target pollen tube.

[0035] In a second aspect, a system for counting the germination rate and germination length of Camellia oleifera pollen in vitro is provided, comprising:

[0036] A microscopic image acquisition module is configured to acquire Camellia oleifera pollen and cultivate it in vitro, and obtain microscopic images of Camellia oleifera pollen tubes, and divide the microscopic images into training images and to-be-tested images.

[0037] A microscopic image annotation processing module is configured to process the training images by using different annotation methods to obtain detection labels for counting the germination rate and segmentation labels for counting the germination length.

[0038] A model training module is configured to train an artificial intelligence model for detection and segmentation by using the annotated training images, the detection labels and the segmentation labels.

[0039] A calculation module based on the artificial intelligence model is configured to input the to-be-tested images into the artificial intelligence model for detection and segmentation to obtain the number of germinated pollen grains, the number of un-germinated pollen grains and the pixel length of the germinated pollen tubes, obtain a pollen grain counting result according to the number of germinated pollen grains and the number of un-germinated pollen grains, and calculate the real length of the germinated pollen tubes according to the pixel length of the germinated pollen tubes.

[0040] A germination rate and germination length counting module is configured to determine and count the germination rate and the germination length of Camellia oleifera pollen according to the pollen grain counting result and the real length of the germinated pollen tubes.

[0041] Further, the microscopic image acquisition module comprises:

[0042] A Camellia oleifera pollen acquisition unit is configured to acquire Camellia oleifera pollen and store it at low temperature.

[0043] An in-vitro cultivation unit is configured to spread the Camellia oleifera pollen in a pollen culture dish containing a prepared Camellia oleifera pollen suspension culture solution, mix thoroughly and then place it in a constant-temperature incubator for dark culture.

[0044] A microscopic image shooting unit is configured to place the current pollen culture dish under a microscope, adjust the focal length of the microscope so that the pollen grains and pollen tubes in the current pollen culture dish can be clearly imaged under the microscope, move the current pollen culture dish to shoot multiple fields of view until the microscopic images of all central regions of the current pollen culture dish are acquired, replace the pollen culture dish that has not been shot, and acquire the microscopic images of all the pollen culture dishes.

[0045] The micro-image division unit is configured to orderly store micro-images of all pollen culture dishes and randomly select the micro-images to divide the micro-images into training images and to-be-tested images.

[0046] Further, the micro-image labeling processing module comprises:

[0047] The pollen grain labeling processing unit is configured to use a rectangular frame tool to frame and label pollen grains in the training images by using image labeling software, set label categories of ungerminated pollen grains and germinated pollen grains respectively, and obtain detection labels for counting germination rates.

[0048] The pollen tube labeling processing unit is configured to use a polygon tool to surround and label pollen tubes in the training images by using image labeling software, set label categories of pollen tubes that are clearly imaged and not crossed with other pollen tubes as normal, label categories of pollen tubes that are clearly imaged and crossed with other pollen tubes as cross, and label categories of pollen tubes that are not clearly imaged or extend beyond the focal length of the lens as error, and obtain segmentation labels.

[0049] Further, the model training module comprises:

[0050] The germination rate detection model construction unit is configured to construct a germination rate detection model for counting pollen germination rates.

[0051] The germination rate detection model training unit is configured to input the labeled training images and the detection labels into the germination rate detection model for training, save the best detection model weight for outputting a detection result of the pollen tubes, and save the best detection model weight for outputting a detection result of the pollen tubes.

[0052] The instance segmentation model construction unit is configured to construct an instance segmentation model for counting lengths of the pollen tubes.

[0053] The instance segmentation model training unit is configured to input the labeled training images and the segmentation labels into the instance segmentation model for training, save the best segmentation model weight for outputting a segmentation result of the pollen tubes, and save the best segmentation model weight for outputting a segmentation result of the pollen tubes.

[0054] The model optimization unit is configured to randomly select part of the to-be-tested images, input the to-be-tested images into the trained models in the germination rate detection model training unit and the instance segmentation model training unit, obtain new detection labels and new segmentation labels, continue to input the new detection labels and the new segmentation labels into the models in the germination rate detection model training unit and the instance segmentation model training unit after adjusting the detection labels and the segmentation labels according to the new detection labels and the new segmentation labels, and repeatedly execute until the germination rate detection model and the instance segmentation model reach a preset model performance to obtain an artificial intelligence model for detection and segmentation.

[0055] Further, the model optimization unit is configured to randomly select part of the to-be-tested images, input the to-be-tested images into the trained models in the germination rate detection model training unit and the instance segmentation model training unit, obtain new detection labels and new segmentation labels, continue to input the new detection labels and the new segmentation labels into the models in the germination rate detection model training unit and the instance segmentation model training unit after adjusting the detection labels and the segmentation labels according to the new detection labels and the new segmentation labels, and repeatedly execute until the germination rate detection model and the instance segmentation model reach a preset model performance to obtain an artificial intelligence model for detection and segmentation.

[0056] a label obtaining unit, configured to input a to-be-tested image into an artificial intelligence model for detection and segmentation to obtain target detection labels and target segmentation labels of the to-be-tested image;

[0057] a label screening unit, configured to screen the target detection labels and the target segmentation labels according to preset label specifications;

[0058] a pollen grain counting unit, configured to count the screened target detection labels to obtain a number of germinated pollen grains and a number of ungerminated pollen grains, and obtain a pollen grain counting result according to the number of germinated pollen grains and the number of ungerminated pollen grains;

[0059] a pollen tube pixel length counting unit, configured to select target pollen tubes with label categories of normal and cross in the screened target segmentation labels, draw a mask image of each target pollen tube using a polygon label, and process the mask image using a thinning algorithm to obtain a pixel length of each target pollen tube;

[0060] a pollen tube real length calculating unit, configured to convert the pixel length according to a scale of image pixels to real length of the to-be-tested image to obtain a real length of each target pollen tube.

[0061] The present application has the following advantages:

[0062] The oil tea pollen is collected and cultured in vitro, a microscopic image of the pollen tube is obtained, and the microscopic image is divided into a training image and a to-be-tested image; different labeling methods are adopted to process the training image to obtain detection labels for counting the germination rate and segmentation labels for counting the germination length; the artificial intelligence model for detection and segmentation is trained using the labeled training image, the detection labels and the segmentation labels; the to-be-tested image is input into the artificial intelligence model for detection and segmentation to obtain the number of germinated pollen grains, the number of ungerminated pollen grains and the pixel length of the germinated pollen tube, the pollen grain counting result is obtained according to the number of germinated pollen grains and the number of ungerminated pollen grains, and the real length of the germinated pollen tube is calculated according to the pixel length of the germinated pollen tube; the germination rate and the germination length of the oil tea pollen are determined and counted according to the pollen grain counting result and the real length of the germinated pollen tube; the technical problems of complicated image processing and large workload after the pollen tube of the oil tea pollen is collected and cultured in vitro are solved. By adopting the deep learning scheme, the artificial intelligence technology is used to automatically detect and segment the pollen tube microscopic image, so that the germination rate and the germination length data of the pollen cultured in vitro are obtained. BRIEF DESCRIPTION OF DRAWINGS

[0063] Figure 1 The flowchart of the method for counting the germination rate and the germination length of the oil tea pollen cultured in vitro according to the present application;

[0064] Figure 2This is a schematic diagram illustrating the label rectangle marking and classification of the present invention;

[0065] Figure 3 This is a schematic diagram illustrating the content contained in the detection tag according to the present invention;

[0066] Figure 4 This is a schematic diagram illustrating the segmentation labeling and classification of the present invention;

[0067] Figure 5 This is a structural diagram of the system for statistically analyzing the germination rate and germination length of isolated Camellia oleifera pollen according to the present invention;

[0068] Figure 6 This is a structural diagram of the microscopic image acquisition module of the present invention;

[0069] Figure 7 This is a structural diagram of the model training module of the present invention;

[0070] Figure 8 This is a structural diagram of the computation module based on the artificial intelligence model of the present invention. Detailed Implementation

[0071] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0072] like Figure 1 As shown, this embodiment of the invention provides a method for statistically analyzing the germination rate and germination length of isolated Camellia oleifera pollen, comprising:

[0073] S1. Collect camellia pollen and culture it in vitro to obtain microscopic images of camellia pollen tubes. Divide the microscopic images into training images and test images.

[0074] In this embodiment, the specific process of in vitro culture and photography of camellia pollen is as follows:

[0075] S11, collect pollen from Camellia oleifera and preserve it at low temperature; collect pollen from Camellia oleifera 'Asus' during its peak flowering period, select flower buds at the white stage, gently remove the anthers with alcohol-sterilized pointed tweezers, lay them flat on sulfuric acid paper, dry them at room temperature (26℃) for 14 h, and collect the pollen in 1.5 mL centrifuge tubes after the anthers have fully dehisced, and store them in a -80℃ refrigerator.

[0076] S12, spread the camellia pollen in the pollen culture dish prepared with the camellia in vitro pollen suspension culture solution, mix well and place in a constant temperature incubator for dark culture; prepare the camellia in vitro pollen suspension culture system, the components of which include 10% sucrose and 0.015% boric acid, pH is 5.8; take out the pollen collected in step S11 in advance, thaw at room temperature for 2h; add different amounts of 10mM abscisic acid (ABA) stock solution to the culture solution, pour 10mL culture solution into each culture dish, so that the final concentration of ABA treatment is: 0μM, 10μM, 30μM, 50μM, 70μM, 90μM, each treatment has 3 repeats, use a brush to dip the camellia pollen and evenly shake it in the culture solution of different treatment concentrations, mix well and place in a 25℃ constant temperature incubator for dark culture for 2h;

[0077] S13, place the current pollen culture dish under a microscope, adjust the focal length of the microscope so that the pollen grains and pollen tubes in the current pollen culture dish can be clearly imaged under the microscope, move the current pollen culture dish to take multiple field shots until all the central area images of the current pollen culture dish are collected;

[0078] After the pollen in the culture dish of the constant temperature incubator is terminated, turn on the microscope power, use OLYMPUS DP71 lens, adjust the ISO of the camera to "200", and the exposure time to "auto" mode, place the pollen culture dish after in vitro culture in step S12 under the microscope, adjust the focal length of the microscope so that the pollen and pollen tubes can be clearly imaged under the microscope, move the current pollen culture dish to ensure that there are more than 30 pollen grains in each field of view, click the "shooting" button to take a picture, move the current pollen culture dish to get a new field of view after the shooting is completed, repeat the shooting step until all the central area images of the current culture dish are collected;

[0079] S14, replace the pollen culture dish that has not completed shooting, repeat step S13 to collect the microscopic images of all the pollen culture dishes;

[0080] S15, orderly store the microscopic images of all the pollen culture dishes, randomly select to divide the microscopic images into training images and test images;

[0081] According to the experimental conditions of culture: camellia varieties, hormone types, hormone concentrations, store in folders in three dimensions, and rename the image files according to the naming rules of "camellia variety_hormone type_hormone concentration", for example, the above culture method can be named as "Huashuo ABA 70μM". After the microscopic images of all the pollen culture dishes are orderly stored, the microscopic images are divided into training images for training the model and test images for testing the model by random selection.

[0082] S2, different labeling methods are adopted for the training images to obtain detection labels for statistical germination rate and segmentation labels for statistical germination length;

[0083] In this embodiment, different labeling methods are adopted for germination rate and germination length:

[0084] S21, the pollen grains in the training images are labeled by using a rectangular frame tool, and the label category of the ungerminated pollen grains is changed to "0" and the label category of the germinated pollen grains is changed to "1" by using an image labeling software (Labelme), a detection annotation file in YOLO format is exported, and detection labels for statistical germination rate are obtained; as shown in Figure 2 , the label of the microscopic image should include: the label category of each pollen grain and the resolution size of the image; as shown in Figure 3 , the width-height ratio of the upper left position of the rectangular frame of each pollen grain relative to the full image and the width-height ratio of the rectangular frame relative to the width-height ratio of the full image;

[0085] S22, the pollen tubes in the training images are labeled by using a polygon tool, and the label category of the pollen tube that is clearly imaged and does not cross with other pollen tubes is set to normal, the label category of the pollen tube that is clearly imaged and crosses with other pollen tubes is set to cross, and the label category of the pollen tube that is not clearly imaged or extends beyond the focal length of the lens is set to error, and the segmentation labels are obtained; as shown in Figure 4 , wherein the pollen tube that extends out of the field of view of the lens also changes the label category to normal or cross according to the situation, and a COCO format instance segmentation annotation file is exported, wherein the label of each microscopic image includes the resolution size of the image, the label category of the pollen tube, and the absolute coordinates of each point of the polygon frame of the pollen tube.

[0086] S3, the training images, the detection labels and the segmentation labels are used to train an artificial intelligence model for detection and segmentation;

[0087] In this embodiment, the specific process of training the model is:

[0088] S31, using python programming language as the basis, calling pytorch and ultralytics third-party python libraries, and constructing an improved YOLOv8x detection model for statistical pollen germination rate (i.e. germination rate detection model);

[0089] S32, input the labeled training image and the detection label in YOLO format into the germination rate detection model for training, and the specific training parameters are 500 cycles until the training loss value converges, and the best detection model weight is saved for outputting the pollen tube detection result;

[0090] S33, use python programming language as the basis, call pytorch and detectron2 and other third-party python libraries, and build an improved EVA instance segmentation model (i.e. instance segmentation model) for counting the length of pollen tube;

[0091] S34, input the labeled training image and the segmentation label in COCO format into the instance segmentation model for training, and the specific training parameters are 200,000 cycles until the training loss value converges, and the best segmentation model weight is saved for outputting the segmentation result of pollen tube;

[0092] S35, randomly select part of the images to be tested, input the trained model in step S32 and step S34, and get new detection labels and new segmentation labels; after adjusting the detection labels and segmentation labels manually using the labelme software according to the new detection labels and new segmentation labels, continue to input the model in step S32 and step S34 for training;

[0093] S36, repeat step S35, and record the final P-R curve of the germination rate detection model in each training process, wherein P is the accuracy and R is the recall, , wherein TP represents the number of correct detection and pollen tube germination, FP represents the number of false detection and pollen tube germination, and FN represents the number of false detection and pollen tube non-germination; when the area surrounded by the P-R curve is close to 1, it indicates that the germination rate detection model has excellent performance;

[0094] record the final average intersection over union value of the instance segmentation model in each training process, and the formula is:

[0095] ,

[0096] wherein TP is correct and the pixel belongs to the current pollen tube, FN is false and the pixel does not belong to the current pollen tube, FP is false and the pixel belongs to the current pollen tube, and k is the total number of pollen tubes of all categories in the current microscopic image; when the average intersection over union is closer to 1, it indicates that the instance segmentation model has excellent performance.

[0097] S4, input the to-be-tested image into the artificial intelligence model for detection and segmentation, to obtain the number of germinated pollen grains, the number of ungerminated pollen grains, and the pixel length of the germinated pollen tube, obtain the pollen grain counting result according to the number of germinated pollen grains and the number of ungerminated pollen grains, and calculate the real length of the germinated pollen tube according to the pixel length of the germinated pollen tube;

[0098] In this embodiment, the specific execution steps are as follows:

[0099] S41, input the to-be-tested image into the artificial intelligence model for detection and segmentation, to obtain the target detection label and the target segmentation label of the to-be-tested image;

[0100] S42, screen the target detection label and the target segmentation label according to the preset label specification;

[0101] The specific screening includes removing the rectangular frame abutting the image edge in the detection label, removing the pollen tube label with the label "error" in the instance segmentation label, the pollen tube label with the polygon point coordinates abutting the image edge, and the pollen tube label with the pixel length less than 1 times the pixel diameter of the pollen image, so as to reduce the interference of the labels not meeting the specification on the result;

[0102] S43, count the screened target detection label to obtain the number of germinated pollen grains and the number of ungerminated pollen grains, and obtain the pollen grain counting result according to the number of germinated pollen grains and the number of ungerminated pollen grains;

[0103] S44, select the target pollen tube with the label class of normal and cross in the screened target segmentation label, draw a mask image of each target pollen tube using the polygon label, and process the mask image using a thinning algorithm to obtain the pixel length of each target pollen tube;

[0104] S45, according to the scale of the image pixel of the to-be-tested image and the real length, the pixel length is converted to obtain the real length of each target pollen tube;

[0105] The calculation formula of the real length is:

[0106] ;

[0107] The scale of the pixel length and the real length is calculated and obtained by measuring the pixel length of the micrometer scale from the micrometer scale photo taken by the microscope.

[0108] In steps S43 and S45 above, the pollen grain counting result and the data of the real length of pollen tube are obtained, and for the convenience of intuitive observation, a display image can also be generated using third-party python libraries such as opencv-python, Pillow, numpy, etc., and a rectangular frame is drawn according to the detection label to superimpose the original image, the rectangular frame is divided into two groups according to the label, and the rectangular frame with the label of “germinated” is drawn in a special color (for example, red) and is pasted with a counting mark, and the rectangular frame with the label of “not germinated” is drawn in a special color (for example, blue) and is pasted with a counting mark, and the output is a picture.

[0109] S5, according to the pollen grain counting result and the real length of the germinated pollen tube, the germination rate and the germination length of the oil tea pollen are determined and counted.

[0110] In this embodiment, the germination rate calculation formula is:

[0111] ;

[0112] The obtained germination rate and pollen tube germination length are subjected to data screening, and after removing outliers, the average value and variance are calculated as the final result, which is convenient for researchers to use for subsequent data analysis.

[0113] The beneficial effects of the embodiment of the present application are:

[0114] The oil tea pollen is collected and cultured in vitro, the microscopic image of the oil tea pollen tube is obtained, the microscopic image is divided into training images and to-be-tested images; the training images are processed by adopting different labeling methods to obtain detection labels for counting the germination rate and segmentation labels for counting the germination length; the labeled training images, the detection labels and the segmentation labels are used to train an artificial intelligence model for detection and segmentation; the to-be-tested images are input into the artificial intelligence model for detection and segmentation to obtain the number of germinated pollen grains, the number of ungerminated pollen grains and the pixel length of the germinated pollen tube, the pollen grain counting result is obtained according to the number of germinated pollen grains and the number of ungerminated pollen grains, and the real length of the germinated pollen tube is calculated according to the pixel length of the germinated pollen tube; according to the pollen grain counting result and the real length of the germinated pollen tube, the germination rate and the germination length of the oil tea pollen are determined and counted; and the technical problems of complicated image processing and large workload after collecting the oil tea pollen tube germination are solved. By adopting the deep learning scheme, the artificial intelligence technology is used to automatically detect and segment the pollen tube microscopic image, so as to obtain the germination rate and the germination length data of the pollen after in vitro culture.

[0115] In combination with the method for counting the germination rate and the germination length of the in vitro oil tea pollen described in the above embodiments, the system for counting the germination rate and the germination length of the in vitro oil tea pollen is described below by way of examples.

[0116] As Figure 5 shown, the embodiment of the present application provides a system for counting the germination rate and germination length of Camellia oleifera pollen in vitro, comprising:

[0117] The microscopic image acquisition module 501 is configured to collect Camellia oleifera pollen and culture it in vitro, obtain microscopic images of Camellia oleifera pollen tubes, and divide the microscopic images into training images and images to be tested.

[0118] The microscopic image labeling processing module 502 is configured to process the training images by using different labeling methods to obtain detection labels for counting the germination rate and segmentation labels for counting the germination length.

[0119] The model training module 503 is configured to train an artificial intelligence model for detection and segmentation by using the labeled training images, the detection labels and the segmentation labels.

[0120] The artificial intelligence model-based calculation module 504 is configured to input the images to be tested into the artificial intelligence model for detection and segmentation to obtain the number of pollen grains that have germinated, the number of pollen grains that have not germinated, and the pixel length of the germinated pollen tubes, obtain a pollen grain counting result according to the number of pollen grains that have germinated and the number of pollen grains that have not germinated, and calculate the real length of the germinated pollen tubes according to the pixel length of the germinated pollen tubes.

[0121] The germination rate and germination length counting module 505 is configured to determine and count the germination rate and germination length of Camellia oleifera pollen according to the pollen grain counting result and the real length of the germinated pollen tubes.

[0122] In combination with Figure 5 shown, preferably, as Figure 6 shown, in some embodiments of the present application, the microscopic image acquisition module 501 comprises:

[0123] The Camellia oleifera pollen collection unit 601 is configured to collect Camellia oleifera pollen and store it at low temperature.

[0124] The in vitro culture unit 602 is configured to spread the Camellia oleifera pollen in a pollen culture dish containing a Camellia oleifera pollen suspension culture solution prepared in vitro, mix thoroughly, and then place it in a constant-temperature incubator for dark culture.

[0125] The microscopic image shooting unit 603 is configured to place the current pollen culture dish under a microscope, adjust the focal length of the microscope so that the pollen grains and pollen tubes in the current pollen culture dish can be clearly imaged under the microscope, move the current pollen culture dish to shoot multiple fields of view until the microscopic images of all central regions of the current pollen culture dish are collected, replace the pollen culture dishes that have not been shot, and collect the microscopic images of all the pollen culture dishes.

[0126] The microscopic image division unit 604 is configured to sequentially store the microscopic images of all pollen culture dishes, and randomly select the microscopic images as training images and to-be-tested images.

[0127] In combination Figure 5 In the embodiment shown, preferably, the microscopic image labeling processing module 502 in some embodiments of the application comprises:

[0128] The pollen grain labeling processing unit is configured to use a rectangular frame tool to frame and label the pollen grains in the training images by using image labeling software, set label categories of ungerminated pollen grains and germinated pollen grains respectively, and obtain detection labels for counting the germination rate.

[0129] The pollen tube labeling processing unit is configured to use a polygon tool to surround and label the pollen tubes in the training images by using image labeling software, set label categories of pollen tubes that are clearly imaged and not crossed with other pollen tubes as normal, label categories of pollen tubes that are clearly imaged and crossed with other pollen tubes as cross, and label categories of pollen tubes that are not clearly imaged or extend beyond the focal length of the lens as error, and obtain segmentation labels.

[0130] In combination Figure 5 In the embodiment shown, preferably, as Figure 7 shown, the model training module 503 in some embodiments of the application comprises:

[0131] The germination rate detection model construction unit 701 is configured to construct a germination rate detection model for counting the pollen germination rate.

[0132] The germination rate detection model training unit 702 is configured to input the labeled training images and detection labels into the germination rate detection model for training, save the best detection model weight for outputting the detection result of the pollen tubes, and output the detection result of the pollen tubes.

[0133] The instance segmentation model construction unit 703 is configured to construct an instance segmentation model for counting the lengths of the pollen tubes.

[0134] The instance segmentation model training unit 704 is configured to input the labeled training images and segmentation labels into the instance segmentation model for training, save the best segmentation model weight for outputting the segmentation result of the pollen tubes, and output the segmentation result of the pollen tubes.

[0135] The model optimization unit 705 is configured to randomly select part of the to-be-tested images, input the trained model in the germination rate detection model training unit and the instance segmentation model training unit, and obtain new detection labels and new segmentation labels; after adjusting the detection labels and the segmentation labels according to the new detection labels and the new segmentation labels, continue to input the model in the germination rate detection model training unit and the instance segmentation model training unit for training; and repeat the execution until the germination rate detection model and the instance segmentation model reach a preset model performance, and obtain the artificial intelligence model for detection and segmentation.

[0136] In combination Figure 5 with the embodiments shown in the drawings, preferably, as Figure 8 shown in the drawings, in some embodiments of the present application, the calculation module 504 based on the artificial intelligence model comprises:

[0137] The label acquisition unit 801 is configured to input the to-be-tested images into the artificial intelligence model for detection and segmentation, and obtain target detection labels and target segmentation labels of the to-be-tested images.

[0138] The label screening unit 802 is configured to screen the target detection labels and the target segmentation labels according to a preset label specification.

[0139] The pollen grain statistical unit 803 is configured to count the screened target detection labels, and obtain the number of pollen grains that have germinated and the number of pollen grains that have not germinated, and obtain a pollen grain counting result according to the number of pollen grains that have germinated and the number of pollen grains that have not germinated.

[0140] The pollen tube pixel length statistical unit 804 is configured to select target pollen tubes with label categories of normal and cross in the screened target segmentation labels, draw a mask image of each target pollen tube using a polygon label, and process the mask image using a thinning algorithm to obtain the pixel length of each target pollen tube.

[0141] The pollen tube real length calculation unit 805 is configured to convert the pixel length according to the scale of the image pixels of the to-be-tested images and the real length, and obtain the real length of each target pollen tube.

[0142] In summary, the system for statistically counting the germination rate and germination length of the in-vitro camellia pollen according to the embodiments of the present application has the following beneficial effects:

[0143] The trained artificial intelligence model uses a predetermined rule automatic program to process image data, reduces the interference of human factors, and improves the accuracy and reliability of the data.

[0144] The existing method is limited to manual identification and data acquisition of human, so that image processing and analysis contain a large amount of repetitive labor, greatly occupy the research time of researchers, and slow down the research progress; the present application can realize automatic acquisition of experimental data through program after image acquisition, greatly improve the data analysis efficiency of researchers engaged in related research;

[0145] The present application processes larger data volume in less time, and the traditional method may select a smaller sample size due to time and resource limitations when processing a large amount of data, or even may select and screen image data that is troublesome to process in order to reduce workload, thereby affecting the representativeness and depth of statistical results; the present application can quickly process a large amount of data, so that researchers can analyze based on a larger sample size, thereby obtaining more comprehensive and in-depth statistical results.

[0146] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0147] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks. Figure 1 The functions specified in one or more flows and / or blocks.

[0148] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks. Figure 1 The functions specified in one or more flows and / or blocks.

[0149] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices, to generate computer-implemented processes, thus the instructions executed on the computer or other programmable data processing devices provide processes for implementing the functions specified in the flowchart Figure 1 one flow or a plurality of flows and / or the functions specified in the flowchart Figure 1 one flow or a plurality of flows and / or the functions specified in the flowchart

[0150] The above merely describes the embodiments of the present application, and is not intended to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of the claims of the present application.

Claims

1. A method for statistically determining the germination rate and germination length of Camellia oleifera pollen in vitro, characterized by, The method comprises the following steps: S1, collecting oil tea pollen and culturing in vitro to obtain microscopic images of oil tea pollen tubes, and dividing the microscopic images into training images and test images; S2, processing the training images by adopting different labeling methods to obtain detection labels for counting germination rates and segmentation labels for counting germination lengths; S3, training an artificial intelligence model for detection and segmentation by using the labeled training images, the detection labels, and the segmentation labels; S4, inputting the test images into the artificial intelligence model for detection and segmentation to obtain the number of germinated pollen grains, the number of ungerminated pollen grains, and the pixel length of the germinated pollen tubes, obtaining pollen grain counting results according to the number of germinated pollen grains and the number of ungerminated pollen grains, and calculating the true length of the germinated pollen tubes according to the pixel length of the germinated pollen tubes; S5, determining and counting the germination rate and the germination length of the oil tea pollen according to the pollen grain counting results and the true length of the germinated pollen tubes; In the step S4, the test images are input into the artificial intelligence model for detection and segmentation to obtain the number of germinated pollen grains, the number of ungerminated pollen grains, and the pixel length of the germinated pollen tubes, the pollen grain counting results are obtained according to the number of germinated pollen grains and the number of ungerminated pollen grains, and the true length of the germinated pollen tubes is calculated according to the pixel length of the germinated pollen tubes, which comprises the following steps: S41, inputting the test images into the artificial intelligence model for detection and segmentation to obtain target detection labels and target segmentation labels of the test images; S42, screening the target detection labels and the target segmentation labels according to preset label specifications; S43, counting the screened target detection labels to obtain the number of germinated pollen grains and the number of ungerminated pollen grains, and obtaining the pollen grain counting results according to the number of germinated pollen grains and the number of ungerminated pollen grains; S44, selecting target pollen tubes with label categories of normal and cross from the screened target segmentation labels, drawing a mask image of each target pollen tube using a polygon label, and processing the mask image to obtain the pixel length of each target pollen tube by using a thinning algorithm; S45, converting the pixel length according to the scale of the image pixels and the true length of the test images to obtain the true length of each target pollen tube.

2. The method for determining the germination rate and germination length of Camellia oleifera pollen ex situ according to claim 1, characterized in that, In the step S1, the oil tea pollen is collected and cultured in vitro to obtain microscopic images of oil tea pollen tubes, and the microscopic images are divided into training images and test images, which comprises the following steps: S11, collecting oil tea pollen and storing it at low temperature; S12, spreading the oil tea pollen in a pollen culture dish containing an oil tea in vitro pollen suspension culture solution, thoroughly mixing, and then placing it in a constant temperature incubator for dark culture. S13, place the current pollen culture dish under a microscope, adjust the focal length of the microscope so that the pollen grains and pollen tubes in the current pollen culture dish can be clearly imaged under the microscope, move the current pollen culture dish to take multiple field of view shots until all the central regions of the current pollen culture dish are collected micrograph images; S14, replace the pollen culture dish that has not been completed, repeat the step S13, collect the micrograph images of all the pollen culture dishes; S15, orderly store the micrograph images of all the pollen culture dishes, randomly select the micrograph images to be divided into training images and test images.

3. The method for determining the germination rate and germination length of Camellia oleifera pollen ex situ according to claim 1, characterized in that, The step S2, the training images are processed by different labeling methods to obtain detection labels for statistical germination rate and segmentation labels for statistical germination length, including: S21, using a rectangular frame tool to frame and label the pollen grains in the training images by using an image labeling software, setting the label categories of ungerminated pollen grains and germinated pollen grains respectively, to obtain detection labels for statistical germination rate; S22, using a polygon tool to enclose and label the pollen tubes in the training images by using the image labeling software, setting the label categories of the pollen tubes that are clearly imaged and not crossed with other pollen tubes as normal, the label categories of the pollen tubes that are clearly imaged and crossed with other pollen tubes as cross, and the label categories of the pollen tubes that are not clearly imaged or longer than the focal length of the lens as error, to obtain segmentation labels.

4. The method for determining the germination rate and germination length of Camellia oleifera pollen ex situ according to claim 3, characterized in that, The step S3, using the labeled training images, the detection labels and the segmentation labels to train an artificial intelligence model for detection and segmentation, including: S31, constructing a germination rate detection model for statistical pollen germination rate; S32, inputting the labeled training images and the detection labels into the germination rate detection model for training, saving the best detection model weight for outputting the detection result of the pollen tube; S33, constructing an instance segmentation model for statistical pollen tube length; S34, inputting the labeled training images and the segmentation labels into the instance segmentation model for training, saving the best segmentation model weight for outputting the segmentation result of the pollen tube; S35, randomly selecting part of the test images, inputting them into the models trained in steps S32 and S34 to obtain new detection labels and new segmentation labels; after adjusting the detection labels and the segmentation labels according to the new detection labels and the new segmentation labels, continue to input them into the models in steps S32 and S34 for training; S36, repeating step S35 until the germination rate detection model and the instance segmentation model reach the preset model performance, to obtain an artificial intelligence model for detection and segmentation.

5. A system for statistically determining the germination rate and germination length of Camellia oleifera pollen in vitro, characterized in that, Including: A micrograph image acquisition module for collecting and culturing oil tea pollen in vitro, obtaining micrograph images of oil tea pollen tubes, and dividing the micrograph images into training images and test images; The micro-image labeling processing module is configured to process the training images in different labeling manners to obtain a detection label for counting the germination rate and a segmentation label for counting the germination length. The model training module is configured to train an artificial intelligence model for detection and segmentation by using the labeled training images, the detection label, and the segmentation label. The artificial intelligence model-based calculation module is configured to input the to-be-tested images into the artificial intelligence model for detection and segmentation to obtain the number of germinated pollen grains, the number of un-germinated pollen grains, and the pixel length of the germination pollen tube, to obtain a pollen grain counting result according to the number of germinated pollen grains and the number of un-germinated pollen grains, and to calculate a real length of the germination pollen tube according to the pixel length of the germination pollen tube. The germination rate and germination length statistical module is configured to determine and count the germination rate and the germination length of the oil tea pollen according to the pollen grain counting result and the real length of the germination pollen tube. The artificial intelligence model-based calculation module includes: The label acquisition unit is configured to input the to-be-tested images into the artificial intelligence model for detection and segmentation to obtain target detection labels and target segmentation labels of the to-be-tested images. The label screening unit is configured to screen the target detection labels and the target segmentation labels according to a preset label specification. The pollen grain statistical unit is configured to count the screened target detection labels to obtain the number of germinated pollen grains and the number of un-germinated pollen grains, and to obtain a pollen grain counting result according to the number of germinated pollen grains and the number of un-germinated pollen grains. The pollen tube pixel length statistical unit is configured to select target pollen tubes with label categories of normal and cross from the screened target segmentation labels, to draw a mask image of each target pollen tube using a polygon label, and to process the mask image by using a thinning algorithm to obtain a pixel length of each target pollen tube. The pollen tube real length calculation unit is configured to convert the pixel length according to a scale of image pixels to real length of the to-be-tested images to obtain a real length of each target pollen tube.

6. The system for statistically determining the germination rate and germination length of Camellia oleifera pollen outside the body according to claim 5, wherein, The micro-image acquisition module includes: The oil tea pollen acquisition unit is configured to acquire oil tea pollen and store the oil tea pollen at a low temperature. The in-vitro culture unit is configured to spread the oil tea pollen in a pollen culture dish containing an oil tea in-vitro pollen suspension culture solution, mix thoroughly, and place the pollen culture dish in a constant-temperature incubator for dark culture. The micro-image shooting unit is configured to place a current pollen culture dish under a microscope, adjust the focal length of the microscope so that the pollen grains and pollen tubes in the current pollen culture dish can be clearly imaged under the microscope, move the current pollen culture dish to shoot multiple fields of view until micro-images of all central regions of the current pollen culture dish are acquired, replace the pollen culture dish that has not been shot, and acquire micro-images of all pollen culture dishes. The micro-image division unit is configured to orderly store the micro-images of all pollen culture dishes and randomly select the micro-images to divide the micro-images into training images and to-be-tested images.

7. The system for statistically determining the germination rate and germination length of Camellia oleifera pollen outside the body according to claim 5, wherein, The microscopic image labeling processing module comprises: The pollen grain labeling processing unit is configured to use a rectangular frame tool to frame and label the pollen grains in the training images by using image labeling software, set label categories of ungerminated pollen grains and germinated pollen grains respectively, and obtain detection labels for counting germination rates. The pollen tube labeling processing unit is configured to use a polygon tool to surround and label the pollen tubes in the training images by using the image labeling software, set label categories of pollen tubes that are clearly imaged and not crossed with other pollen tubes as normal, label categories of pollen tubes that are clearly imaged and crossed with other pollen tubes as cross, and label categories of pollen tubes that are not clearly imaged or extend beyond the focal length of the lens as error, and obtain segmentation labels.

8. The system for statistically determining the germination rate and germination length of Camellia oleifera pollen outside the body according to claim 7, characterized in that, The model training module comprises: The germination rate detection model construction unit is configured to construct a germination rate detection model for counting pollen germination rates. The germination rate detection model training unit is configured to input the labeled training images and the detection labels into the germination rate detection model for training, save the best detection model weight for outputting a detection result of pollen tubes, and output the detection result of pollen tubes. The instance segmentation model construction unit is configured to construct an instance segmentation model for counting lengths of pollen tubes. The instance segmentation model training unit is configured to input the labeled training images and the segmentation labels into the instance segmentation model for training, save the best segmentation model weight for outputting a segmentation result of pollen tubes, and output the segmentation result of pollen tubes. The model optimization unit is configured to randomly select part of the to-be-tested images, input the to-be-tested images into the trained models of the germination rate detection model training unit and the instance segmentation model training unit, obtain new detection labels and new segmentation labels, continue to input the new detection labels and the new segmentation labels into the models of the germination rate detection model training unit and the instance segmentation model training unit after adjusting the detection labels and the segmentation labels according to the new detection labels and the new segmentation labels, and repeatedly execute until the germination rate detection model and the instance segmentation model reach a preset model performance, to obtain an artificial intelligence model for detection and segmentation.

Images