Poplar leaf pore detection method based on microscopic image

Abstract

The invention discloses a poplar leaf pore detection method based on microscopic images, and belongs to the technical field of machine vision. The method comprises the following steps: constructing a special data set containing poplar leaf stoma microscopic images and annotation information under different moisture treatments; training a deep learning detection model integrated with a feature enhancement module to reinforce the features of the stomata region and suppress background interference; a two-stage detection process is adopted, in the first stage, a pore bounding box and feature point coordinates are output through a model, and in the second stage, morphological parameters such as pore length, width, opening and density are calculated based on feature points; and finally, outputting a detection result containing the visual annotation and the structured parameters. The method realizes automation and high precision of poplar pore detection, significantly improves the detection efficiency, effectively eliminates artificial errors, is suitable for poplar stress resistance research and breeding screening under drought stress, can be expanded to other plant species and environmental stress scenes, and has wide application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine vision, in particular to a poplar leaf stomata detection method based on microscopic images. BACKGROUND

[0002] Stomata are the key structures for gas (CO2, O2) and water exchange between plant leaves and the external environment. Their morphological parameters (such as length, width, density, and opening degree) directly reflect the photosynthetic efficiency, water use efficiency, and stress resistance (especially drought stress response) of plants. Accurate detection of stomata morphology is of great significance for plant physiology research and breeding work.

[0003] Traditional stomata detection methods rely on manual microscope observation and manual measurement, which have the following significant defects: single stomata measurement takes more than 10 seconds, and large-scale samples (such as 400 stomata) take 1 hour, which cannot meet the high-throughput analysis requirements. The manual annotation of recognition features is greatly influenced by the experience of the operator, and systematic errors are easily produced.

[0004] In the prior art, although there are attempts to detect stomata based on deep learning, most of them are aimed at crops (such as wheat and soybeans), and are not optimized for the specificity of poplar stomata (such as morphological variation under drought stress and complex leaf texture), resulting in insufficient detection accuracy and robustness. Therefore, there is an urgent need for an efficient, accurate, and automated detection method for poplar leaf stomata. SUMMARY

[0005] (I) Technical problems solved

[0006] To overcome the shortcomings of the prior art, the present application provides a poplar leaf stomata detection method based on microscopic images, which realizes automatic recognition, positioning, and morphological parameter calculation of stomata by constructing a special data set, optimizing a deep learning model, and designing an automated process, thereby improving detection efficiency and accuracy and meeting the needs of large-scale sample analysis.

[0007] (II) Technical solutions

[0008] To achieve the above purpose, the present application is implemented by the following technical solutions: a poplar leaf stomata detection method based on microscopic images, comprising:

[0009] Step 1: Construct a special microscopic image data set, which contains microscopic images for stomata detection and corresponding annotation information;

[0010] Step 2: Train a deep learning detection model to realize the recognition and positioning of stomata and feature point extraction;

[0011] Step 3: Detect the input microscopic image through the deep learning detection model, and output the boundary box and feature point coordinates of the stomata;

[0012] Step four: calculating the morphological parameters of stomata based on the coordinates of the landmarks;

[0013] Step five: outputting the detection results containing the morphological parameters of stomata.

[0014] Preferably, the construction of the special microscopic image dataset in step one includes:

[0015] Step A: collecting microscopic images of the stomata of poplar leaves, the images being obtained by an optical microscope;

[0016] Step B: labeling the stomata in the microscopic images, the labeling content including the bounding box of the stomata and the coordinates of the feature points;

[0017] Step C: converting the labeling information format to be compatible with the deep learning model.

[0018] Preferably, in step A, the poplar leaf is the hybrid seedling leaf of the hybrid offspring of Populus deltoids and Populus nigra, the microscopic images cover stomata samples under different water treatment conditions, and the water treatment conditions include a normal water supply control group with a soil relative water content of 75%-80% and a drought stress group with a soil relative water content of 35%-40%.

[0019] Preferably, in step A, the collection of the microscopic images of the stomata adopts the epidermal imprinting method, and the specific steps are as follows:

[0020] Apply transparent imprinting material to the lower epidermis of the leaf avoiding the main vein area, and after drying, peel off to form a stomata imprint sample, prepare a temporary mount from the sample, and collect images through an optical microscope with a 40x eyepiece and a 10x objective lens imaging unit, with an image resolution of 1920x1080 pixels.

[0021] Preferably, in step two, the deep learning detection model integrates a feature enhancement module, which can specifically enhance the features of the stomata region and suppress background interference, realizing automatic processing from image input to morphological parameter output, the feature enhancement module of the deep learning detection model is an attention mechanism module, which can enhance the subtle features of the stomata region by dynamically adjusting the feature weight and suppress the interference of the background texture of the leaf stomata recognition.

[0022] Preferably, the deep learning detection model is improved based on the YOLOv8-pose model, and the attention mechanism module is a SimAM module integrated in the head network of the model.

[0023] Preferably, the feature points in step three are the four vertices of the stomatal outline, corresponding to the upper left, upper right, lower right and lower left positions of the stomata respectively, and the morphological parameters in step four include: the stomatal length calculated by the straight-line distance between the upper left vertex and the lower right vertex, the stomatal width calculated by the straight-line distance between the upper right vertex and the lower left vertex, the stomatal opening calculated by the formula "length x width x π / 4" based on the length and width, and the stomatal density calculated by the number of stomata in a unit area.

[0024] Preferably, the detection results in step five include: a visualization annotated image of superimposing a stomatal bounding box on the original microscopic image, different feature points using differentiated color marking, and morphological parameter text; and structured data recording the serial number, coordinates, confidence and morphological parameters of each stomata in table form.

[0025] Preferably, the method supports single sample detection and batch sample detection: real-time output of results in single sample detection, and processing of multiple images through parallel processing mechanism to generate summary results in batch sample detection.

[0026] (Three) beneficial effects

[0027] The present application provides a poplar leaf stomata detection method based on microscopic images. It has the following beneficial effects:

[0028] 1. The present application realizes the detection of poplar stomata through the synergistic optimization of special data set and improved model, and optimizes the characteristics of small size, complex background texture and morphological variation of stomata, which can accurately identify and locate the stomatal boundary and key feature points, and effectively reduce the false detection and missed detection problems caused by manual judgment or insufficient adaptability of general model in traditional methods. This high-precision feature is especially suitable for detecting the subtle morphological changes of stomata under environmental stress such as drought stress, and can accurately capture the morphological adjustment of stomata in stress response.

[0029] 2. The whole process from image input to parameter output of the present application does not need human intervention, which shortens the detection time of single sample and batch sample, solves the problem of "time-consuming and laborious, difficult to deal with large-scale samples" in traditional manual measurement. It enables researchers to quickly process multiple genotypes, multiple treatment groups and multiple repeated experimental samples, meets the demand of high-throughput phenotype screening in forest tree breeding, and provides guarantee for the selection of poplar clones with excellent drought resistance.

[0030] 3.The application avoids subjective errors caused by factors such as operator experience and fatigue level in traditional manual measurement, ensuring the objectivity and repeatability of data. At the same time, the automated process simplifies the operation steps, and the operator does not need to have professional microscope operation skills or stomatal morphology knowledge, but only needs to complete the detection through simple software operation, reducing the dependence on professionals and making the technology easier to popularize and apply. In addition, the output data is saved in a standardized format and can be directly used for subsequent statistical analysis, solving the problem of data format confusion and integration difficulty in traditional manual recording.

[0031] 4.The application specially adapts to the uniqueness of poplar stomata through special data set construction and model customization. The data set covers the stomatal morphological characteristics of poplar under different water conditions, especially the specific changes under drought stress, so that the model can accurately identify atypical stomatal morphology. At the same time, the model strengthens the attention to the stomatal area through a feature enhancement mechanism, effectively suppressing background interference such as leaf epidermis texture and impurities, solving the problem of insufficient recognition ability of general models for densely distributed and significantly different size stomata in poplar leaf images, and ensuring stable detection in complex scenes.

[0032] 5.The core framework of the application has good expansibility. For other plant species, only the corresponding special data set needs to be replaced, and the same model training process can be used for detection adaptation; for other environmental stresses, samples under corresponding stress conditions can be included by expanding the data set, so that the model has the ability to detect different stress responses. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 is the flowchart of the whole application;

[0034] Figure 2 is the flowchart of the application for constructing a special microscopic image data set;

[0035] Figure 3 is the flowchart of the application for training a deep learning detection module;

[0036] Figure 4 is the flowchart of the application for detection and calculation, result output. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0038] Please refer to Figure 1The application provides a technical scheme: a poplar leaf stomatal detection method based on microscopic images, comprising:

[0039] Step one: constructing a special microscopic image dataset, which contains microscopic images for stomatal detection and corresponding annotation information;

[0040] Step two: training a deep learning detection model for realizing stomatal recognition, positioning and feature point extraction;

[0041] Step three: detecting the input microscopic image through the deep learning detection model, and outputting the boundary box and feature point coordinates of the stomata;

[0042] Step four: calculating the morphological parameters of the stomata based on the feature point coordinates;

[0043] Step five: outputting the detection results containing the stomatal morphological parameters.

[0044] The construction of the special microscopic image dataset in step one comprises:

[0045] Step A: collecting the stomatal microscopic images of poplar leaves, which are obtained through an optical microscope;

[0046] Step B: annotating the stomata in the microscopic images, and the annotation content includes the boundary box and feature point coordinates of the stomata;

[0047] Step C: converting the annotation information format to be compatible with the deep learning model.

[0048] In step A, the poplar leaves are the hybrid offspring seedling leaves of small black poplar and European black poplar, and the microscopic images cover stomatal samples under different water treatment conditions, including the control group with a soil relative water content of 75%-80% and the drought stress group with a soil relative water content of 35%-40%.

[0049] In step A, the stomatal microscopic images are collected by using the epidermal imprinting method, and the specific steps are as follows:

[0050] The transparent imprinting material is applied to the lower epidermis of the leaf avoiding the main vein area, and after drying, the stomatal imprint sample is peeled off to form a temporary mount, and the image is collected through an optical microscope with a 40x eyepiece and a 10x objective lens, with an image resolution of 1920x1080 pixels.

[0051] In step two, the deep learning detection model integrates a feature enhancement module, which can specifically enhance the features of the stomatal region and suppress the background interference, realize the automatic processing from image input to morphological parameter output, and the feature enhancement module of the deep learning detection model is an attention mechanism module, which can enhance the subtle features of the stomatal region by dynamically adjusting the feature weight and suppress the interference of the leaf background texture.

[0052] The deep learning detection model is improved based on a YOLOv8-pose model, and an attention mechanism module is a SimAM module, which is integrated in a head network of the model.

[0053] The feature points in step three are four vertices of the stomatal outline, respectively corresponding to the upper left, upper right, lower right, and lower left positions of the stomata.

[0054] The detection results in step five include: a visualization labeled image of superimposing a stomatal bounding box on an original microscopic image, differentiating color labels for different feature points, and text of morphological parameters; and structured data in a table form recording a serial number, coordinates, confidence, and morphological parameters of each stomata.

[0055] The method supports single sample detection and batch sample detection: real-time output of results in single sample detection, and parallel processing of multiple images and generation of summary results in batch sample detection.

[0056] First, prepare 30-day-old small black poplar and European black poplar hybrid offspring cuttings; control the cultivation conditions as follows: potting in a greenhouse (soil is humus soil: perlite = 3:1), temperature 25±2℃, light 16h / day, using an Olympus BX53 optical microscope equipped with a 40x eyepiece, a 10x objective lens, and an imaging unit as the imaging unit;

[0057] Then, construct a special dataset, in which:

[0058] Control group: daily watering to maintain soil relative water content of 75% to 80% (monitored by soil moisture meter);

[0059] Drought stress group: watering only at the beginning of the treatment to 35% to 40%, and then naturally water consumption to maintain this range; the treatment period is 30 days, and the growth state of the seedlings is recorded during the period.

[0060] Select the 3rd functional leaf of each seedling, and at the lower epidermis of the leaf, 1 / 3 away from the leaf tip, avoid the main vein, apply transparent nail polish, and after 20 minutes when the nail polish is completely dry, gently tear off the imprint with tweezers, place it on a glass slide to make a temporary mount, and prepare the stomata sample.

[0061] The microscope was focused to clearly show the stomata. Three fields of view were randomly selected from each slide and photographed, saved as 1920x1080 pixel PNG images. The image was opened using the labelme software, and the stomata boundary was labeled with a rectangular frame. Click on the four vertices to label k1-k4. Generate a JSON file. Convert the JSON file to YOLO format using a Python script, normalize the coordinates to the range of 0-1, and finally form a dataset of 3000 images, with a training set:validation set = 8:2. Image acquisition and labeling are completed.

[0062] The basic model is a YOLOv8-pose model containing Backbone, Neck, and Head. The SimAM attention layer is inserted into the Head module. The feature weight is dynamically calculated by the formula, and the stomata region response is strengthened. The model structure is adjusted. The specific calculation formula is as follows:

[0063] where z c (i,j) is the feature image pixel value, μ c is the mean, is the variance, and ε is a constant.

[0064] The specific training process is as follows: 80% of the images in the dataset are used for training, and 20% are used for verification. The learning rate is 0.01, the batch size is 16, and the iteration is 500 times. The model weight is saved every 100 iterations. Combined with the boundary box loss (CIoU), key point loss (MSE), and classification loss (cross-entropy).

[0065] When detecting a single image, the user imports a 1280x720 pixel poplar leaf microscopic image into the software. After preprocessing (normalization, Resize), the image is input into the YOLOv8-SimAM model, and the boundary box and k1-k4 coordinates are output. The software calls the parameter algorithm to automatically calculate the length, width, opening, and density. The interface displays the labeled image (including boundary box, colored key points, and parameters), and the table shows the detailed data.

[0066] When detecting a batch of images, the user selects a folder containing multiple images. The software automatically loads the images in a loop and performs detection and parameter calculation in parallel. A folder is generated, containing the labeled image (PNG) of each image and the summary Excel table (containing all stomata parameters).

[0067] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and implementations, it is to be understood that the terminology used is for the purpose of descriptive clarity and that it is intended to be limited only by the words recited in the appended claims. It is to be understood that the terms "including", "comprising", "consisting" and variations thereof do not preclude the addition of further integers to the claimed combination of integers. It is to be understood that the terms "including", "comprising", "consisting" and variations thereof encompass the various features of the embodiments described herein, and are not restricted to the use of only the most preferred embodiments. It is further to be understood that the use of relational terms such as first and second, and the like, do not denote a physical or logical order or relationship among the various elements, but are used simply for distinguishing between various elements for clarity. It is to be understood that the terms "comprising", "including", and "having" and variations thereof are intended to be equivalent and open-ended, and include the various embodiments of the present application as recited in the claims. It is to be understood that the terms "including", "comprising", "consisting" and variations thereof encompass the various features of the embodiments described herein, and are not restricted to the use of only the most preferred embodiments. It is further to be understood that the use of relational terms such as first and second, and the like, do not denote a physical or logical order or relationship among the various elements, but are used simply for distinguishing between various elements for clarity. It is to be understood that the terms "comprising", "including", and "having" and variations thereof are intended to be equivalent and open-ended, and include the various embodiments of the present application as recited in the claims.

[0068] While the embodiments of the application have been shown and described herein, it is understood that modifications, substitutions, changes, and alterations can be made by those of ordinary skill in the art without departing from the spirit and scope of the present application, which is defined by the following claims and their equivalents.

Claims

1. A poplar leaf stomatal detection method based on microscopic images, characterized by, The application relates to a method for detecting the stomata of a poplar leaf, and belongs to the technical field of plant stomata detection. The method comprises the following steps: Step 1: constructing a special microscopic image dataset, wherein the dataset comprises microscopic images for stomata detection and corresponding annotation information; Step 2: training a deep learning detection model for realizing the identification and positioning of stomata and the extraction of feature points; Step 3: detecting inputted microscopic images through the deep learning detection model to output the boundary box of the stomata and the coordinates of the feature points; Step 4: calculating the morphological parameters of the stomata based on the coordinates of the feature points; 2. The method of claim 1, wherein, Step 5: outputting the detection results comprising the morphological parameters of the stomata. The construction of the special microscopic image dataset in step 1 comprises the following steps: Step A: collecting the stomata microscopic images of poplar leaves, wherein the images are obtained through an optical microscope; Step B: annotating the stomata in the microscopic images, and the annotation content comprises the boundary box of the stomata and the coordinates of the feature points; 3. The method according to claim 2, wherein, Step C: converting the annotation information format into a format compatible with the deep learning model.

4. The method according to claim 2, wherein, In step A, the poplar leaves are the hybrid seedling leaves of Populus deltoides and Populus nigra, the microscopic images cover stomata samples under different water treatment conditions, and the water treatment conditions comprise a normal water supply control group with a soil relative water content of 75% to 80% and a drought stress group with a soil relative water content of 35% to 40%. In step A, the stomata microscopic images are collected by using the epidermis imprinting method, and the specific steps are as follows:

5. The method according to claim 1, wherein, A transparent imprinting material is applied to the lower epidermis of the leaf to avoid the main vein area, the imprinting sample is peeled off after drying to form a stomata imprinting sample, the sample is made into a temporary mounting, and the image is collected through an optical microscope with a 40-fold eyepiece and a 10-fold objective lens imaging unit, and the image resolution is 1920*1080 pixels.

6. The method according to claim 5, wherein, In step 2, the deep learning detection model integrates a feature enhancement module, can specifically strengthen the features of the stomata region and inhibit background interference, realizes automatic processing from image input to morphological parameter output, the feature enhancement module of the deep learning detection model is an attention mechanism module, can strengthen the subtle features of the stomata region by dynamically adjusting the feature weight, and inhibit the interference of the leaf background texture.

7. The method according to claim 1, wherein, The deep learning detection model is improved based on a YOLOv8-pose model, the attention mechanism module is a SimAM module, and the module is integrated in the head network of the model.

8. The method according to claim 1, wherein, In step 3, the feature points are four vertices of the stomata contour, and correspond to the upper left, upper right, lower right and lower left positions of the stomata respectively. In step 4, the morphological parameters comprise the stomata length calculated through the straight-line distance between the upper left vertex and the lower right vertex, the stomata width calculated through the straight-line distance between the upper right vertex and the lower left vertex, the stomata opening calculated through the formula "length x width x pi / 4" based on the length and the width, and the stomata density calculated through the number of stomata in a unit area. In step 5, the detection results comprise a visualized annotation image in which the boundary box of the stomata, different feature points and morphological parameter texts are marked with different colors on the original microscopic image, and the structured data of the serial number, coordinates, confidence and morphological parameters of each stomata are recorded in a table form.

9. The method according to claim 1, wherein, The method supports single sample detection and batch sample detection; in single sample detection, a result is output in real time, and in batch sample detection, multiple images are processed through a parallel processing mechanism, and an aggregated result is generated.

Images