Image recognition-based grape powdery mildew evaluation method
By combining uniform inoculation with a settling tower and a machine learning model, the problems of strong subjectivity and insufficient accuracy in the assessment of grape powdery mildew have been solved, achieving high-throughput, stable and repeatable disease assessment, which is suitable for automated processing of large-scale samples.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SANYA RES INST OF CHINESE ACAD OF TROPICAL AGRI
- Filing Date
- 2026-03-31
- Publication Date
- 2026-06-26
Smart Images
Figure CN122290100A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer vision and image recognition technology, and specifically relates to an image recognition-based method for assessing powdery mildew in grapes. Background Technology
[0002] Grapes (Vitis vinifera L.) are an important economic fruit crop worldwide, widely used for fresh consumption, winemaking, and processed products. In grape production, diseases are a key factor affecting yield and fruit quality. Powdery mildew (Erysiphe necator) is a widespread and frequently occurring fungal disease. Under natural field conditions, the typical symptom of powdery mildew infection on grape leaves and bunches is a white powdery mold layer covering the leaves and fruit surfaces. This disease can spread by wind or rain, infecting leaves, young shoots, inflorescences, and fruits. In the early stages, it is not easily detected by the naked eye. In the middle stages, a white powdery mold layer gradually appears on the leaf surface. If timely control and prevention are not implemented to stop its spread, it can cause large-scale epidemics in the vineyard later, leading to premature leaf senescence, hindered photosynthesis, decreased fruit quality, and even significant yield reduction. Therefore, utilizing and breeding powdery mildew-resistant grape varieties is considered an effective way to reduce control costs and yield losses. Accurate and objective assessment of the degree of powdery mildew infection is an important foundation for the breeding of disease-resistant varieties, the evaluation of disease control effectiveness, and the study of disease mechanisms.
[0003] Although field surveys and manual visual assessment are widely used in existing technologies for evaluating powdery mildew, these methods are significantly affected by various uncontrollable factors in practical applications. The occurrence and development of grape powdery mildew under natural conditions are easily influenced by factors such as climate, wind direction, local humidity, and uneven distribution of pathogens. It often exhibits characteristics of localized outbreaks or scattered distribution. Even within the same plot, the degree of infection varies significantly between different plants and even between different parts of the same plant, making survey results susceptible to accidental interference and difficult to accurately reflect the variety's inherent resistance. Furthermore, the intensity of powdery mildew occurrence exhibits obvious seasonal and annual fluctuations. Survey results from a single year are insufficient as a stable and reliable basis for evaluating resistance; repeated observations over several years are often required to obtain relatively credible conclusions, significantly increasing the experimental cycle and labor costs. In addition, manual visual assessment relies heavily on the experience and judgment of the surveyors. Different personnel often lack completely consistent scoring standards, and even experienced surveyors may give different evaluations for the same leaf or plant, resulting in poor consistency and insufficient repeatability of assessment results. In addition, manual surveys are time-consuming and labor-intensive, making it difficult to achieve high-throughput assessment of large-scale samples. Furthermore, traditional disease grading methods often use discrete grading, which makes it difficult to reflect the continuous changes in the infected area and is not conducive to conducting detailed genetic analysis and quantitative trait studies.
[0004] Currently, in grape powdery mildew research and production practices, disease phenotypic assessment mainly relies on manual field surveys and visual scoring methods. For example, investigators randomly select a certain number of plants or leaves in the field and perform empirical grading and statistical analysis of disease severity based on the proportion of powdery mildew spots or disease severity standards (such as VIVC OIV455). However, this type of method is highly dependent on the investigators' experience and judgment, resulting in low consistency in assessment results among different personnel. Furthermore, disease severity grading often uses discrete methods, making it difficult to reflect the continuous changes in infection severity. In addition, manual surveys are time-consuming and labor-intensive, making them difficult to implement under conditions of large-scale germplasm resource evaluation or continuous monitoring at multiple time points, and assessment results from different years or under different environmental conditions lack good comparability. Therefore, traditional field statistical methods are insufficient in terms of accuracy, repeatability, and throughput to meet the needs of refined disease phenotypic research.
[0005] With the development of computer vision and image recognition technologies, the quantitative assessment of powdery mildew using image analysis methods has gradually attracted attention. Some studies have collected leaf or plant images and combined them with threshold segmentation, color feature extraction, or machine learning models to identify powdery mildew lesion areas, thereby estimating the degree of disease occurrence. Although such methods reduce the influence of subjective human factors to some extent and improve assessment efficiency, they still have problems such as sensitivity to changes in light conditions and background, insufficient accuracy in identifying lightly or moderately infected areas, and some methods rely on a limited number of training samples or fixed parameter settings, showing weak adaptability and transferability under different varieties or different infection levels. It is still difficult to form a stable, repeatable assessment scheme applicable to multiple scenarios.
[0006] Some existing studies attempt to use image analysis for quantitative assessment of powdery mildew to reduce subjective human interference and improve assessment efficiency. However, existing image recognition methods still have certain limitations in practical applications. For example, they are sensitive to lighting conditions, shooting angle, and background environment, and it is difficult to maintain stable recognition results under natural conditions. Some methods rely on a limited number of training samples or fixed threshold settings, resulting in insufficient generalization and transfer capabilities across different varieties, infection levels, or shooting conditions. For lightly or moderately infected areas, the recognition accuracy remains limited when lesion boundaries are unclear. Furthermore, existing image assessment methods are mostly used in laboratories or in single scenarios, and a unified automated assessment process that balances accuracy, stability, and high-throughput processing capabilities has not yet been established. This makes it difficult to meet the practical needs of standardized, reproducible, and large-scale assessment of powdery mildew phenotypes in scientific research and production practice.
[0007] Therefore, there is an urgent need for an image recognition-based method for assessing powdery mildew in grapes that can accurately identify and quantify powdery mildew-infected areas while reducing subjective human interference, and possess good stability, repeatability, and applicability to meet the practical needs of accurate phenotypic assessment of powdery mildew in scientific research and production practice. Summary of the Invention
[0008] The purpose of this invention is to address the problems commonly found in existing methods for assessing powdery mildew in grapes, such as high subjectivity, insufficient accuracy, low throughput, poor repeatability, and difficulty in achieving refined quantitative analysis. This invention provides an image recognition-based method for assessing powdery mildew in grapes, which can achieve objective, accurate, and high-throughput assessment of the degree of powdery mildew infection.
[0009] This invention addresses two core issues: consistency of pathogen inoculation and objectivity of phenotypic assessment. It systematically optimizes the assessment process for grape powdery mildew, with key technical points including: 1. A uniform inoculation system using a sedimentation tower: Powdery mildew spores are naturally and uniformly deposited onto the leaf disc surface using a sedimentation tower, achieving synchronous and equal-intensity inoculation. This significantly reduces spatial heterogeneity and randomness under natural infection conditions, improving the consistency of infection intensity among different samples. 2. A standardized leaf disc culture system: By standardizing leaf sampling sites, leaf disc specifications, and culture conditions, individual differences and environmental interference are reduced, improving comparability between samples and providing a stable basis for accurate phenotypic assessment. 3. Automatic lesion segmentation and quantitative phenotypic extraction based on machine learning: The ImageJ-Weka classification model is used to automatically identify and calculate the area of infected regions. The infection ratio is used as a continuous phenotypic indicator, replacing manual rating and achieving objective and quantifiable disease assessment. 4. A traceable assessment system based on image data: All assessments are completed based on standardized image data, allowing for repeated calculation and verification of results, improving experimental repeatability and verifiability. 5. Standardized process design suitable for high-throughput screening: The overall process is modular and parameterized, and can process large-scale samples in batches, making it suitable for germplasm resource evaluation and disease resistance screening of breeding populations.
[0010] To achieve the above objectives, this invention discloses an image recognition-based method for assessing grape powdery mildew, which specifically includes the following steps:
[0011] Step 1: Sample collection and leaf disc preparation:
[0012] Functional leaves with consistent growth status and no mechanical damage were collected from the grapevines to be evaluated. Leaf discs with a diameter of 3 cm were prepared on the leaves using a standard perforator. At least 4 to 6 complete leaf discs were prepared for each sample as biological replicates. The leaf discs were then placed evenly on agar plates containing nutrient culture medium with the upper surface facing up to maintain leaf viability and facilitate subsequent uniform inoculation. The agar plates containing the leaf discs were then placed in a climate incubator for pre-incubation to reduce the stress resistance to powdery mildew infection caused by mechanical damage and reduce the differences in initial state between samples.
[0013] Step 2: Settling tower assembly and powdery mildew inoculation:
[0014] A settling tower inoculation device was constructed, with a nylon net fixed at the top of the settling tower. Infected leaves carrying a large number of powdery mildew spores were placed in the top area of the settling tower, and gently and evenly brushed onto the nylon net with a brush. Then, the brush was used to evenly spread the powdery mildew on the surface of the nylon net, and the net was tapped and shaken. The pathogen spores were evenly distributed and settled onto the leaf discs in the petri dish through natural settling, thus achieving synchronous and uniform inoculation of powdery mildew. After inoculation, the petri dish was placed under constant temperature and humidity conditions for continued cultivation, so that the disease could occur and develop stably under standardized conditions.
[0015] Step 3: Image acquisition and image preprocessing:
[0016] Fifteen days after inoculation and culture, uniform image acquisition was performed on the leaf discs in the culture dish. High-resolution images were obtained using a fixed light source, fixed shooting height and shooting angle to reduce the impact of lighting changes and background differences on image recognition results. Subsequently, the original images were preprocessed. Through the above processing flow, combined with the batch processing function of Photoshop, the leaf disc region was initially segmented to obtain a standardized image dataset suitable for model training and batch analysis.
[0017] Step 4: Model training and phenotypic extraction based on ImageJ-Weka:
[0018] Eight representative leaf disc images were selected from the preprocessed images as training samples and imported into the Weka image segmentation plugin (Advanced Weka Segmentation) in the ImageJ software platform. The Fast Random Forest classification model was trained and its parameters optimized by manually labeling powdery mildew-infected and non-infected areas, ultimately constructing a classification model for automatic lesion area identification. Using the trained classification model, the remaining 222 images were batch-processed using ImageJ macro commands to automatically identify and segment powdery mildew-infected areas. The area of the infected area and the total area of the leaf discs were calculated to obtain the percentage of powdery mildew infection as a quantitative phenotypic value. This phenotypic value can be directly used for subsequent applications such as comparing disease resistance among different varieties, genetic analysis, and breeding screening.
[0019] Preferably, in step 1, when collecting leaves, avoid obviously aged or severely diseased parts, and prepare at least 5 complete leaf discs for each sample as biological replicates. The nutrient medium used is 1.5% agarose medium.
[0020] Preferably, in step 1, the petri dish containing the leaf discs is placed in a climate incubator and pre-cultured for 24 hours under conditions of 12 hours of alternating light and dark and a temperature of 28°C.
[0021] Preferably, in step 2, the dimensions of the sedimentation tower inoculation device are length × width × height = 50cm × 30cm × 30cm. The nylon mesh is 150 mesh with an aperture of 110μm. The entire inoculation process is carried out in a clean, relatively enclosed environment to avoid the impact of airflow disturbance and environmental pollution on the uniformity of spore sedimentation.
[0022] Preferably, step 3 involves preprocessing the original image, including image inversion, extraction of leaf disc regions based on color range, reverse deletion of the culture dish background, image grayscale processing, contrast enhancement, and image cropping and storage.
[0023] Preferably, in step 4, 222 independent samples were used as the verification objects. The automatically extracted phenotypic values and the manually observed results showed a highly consistent monotonic correlation, with a Spearman correlation coefficient of ρ = −0.935 (P = 5.564 × 10⁻¹). 0 ¹), thus stably reflecting the relative differences in the degree of powdery mildew infection among different samples.
[0024] Beneficial effects
[0025] This invention reduces the interference of randomness and spatial heterogeneity of natural field infection on assessment results by introducing standardized sample preparation and unified inoculation methods. This makes the disease occurrence process more controllable among different materials, thereby improving the stability and comparability of phenotypic data. Simultaneously, image recognition and machine learning technologies are used to automatically identify and calculate the area of powdery mildew-infected regions, avoiding the reliance on the experience of assessors in traditional manual visual grading. This enables a continuous quantitative description of the degree of powdery mildew infection, improving the consistency and accuracy of assessment results.
[0026] Furthermore, this invention constructs a large-scale image analysis workflow that can be used for batch processing, significantly improving the efficiency of disease assessment and reducing the workload of manual investigation. This makes the method applicable to application scenarios requiring the processing of large numbers of samples, such as germplasm resource evaluation and breeding material screening. Through the standardized design of image acquisition, preprocessing, and model recognition processes, the assessment method maintains good stability and versatility under different varieties, infection levels, and experimental conditions, thus forming a replicable and scalable automated powdery mildew assessment scheme.
[0027] By achieving the above technical objectives, this invention provides a stable, objective, repeatable, and large-scale application-appropriate assessment method for evaluating grape powdery mildew resistance, offering reliable phenotypic data support for resistant variety breeding, disease control effectiveness evaluation, and related mechanism research. The image recognition-based grape powdery mildew assessment method proposed in this invention demonstrates good predictive performance and stability in actual samples. Using 222 independent samples as validation subjects, the automatically extracted phenotypic values and manually observed results showed a highly consistent monotonic correlation, with a Spearman correlation coefficient of ρ = −0.935 (P = 5.564 × 10⁻¹). 0 ¹), indicating that the method can stably reflect the relative differences in powdery mildew infection levels among different samples. Further, after fitting the predicted and observed values using an exponential model, the resulting model determination coefficient R² was 0.872, indicating that the method can explain approximately 87.2% of the phenotypic variation, demonstrating strong fitting and phenotypic explanatory power. Meanwhile, the prediction error remained at a low level, with a mean absolute error (MAE) of 2.72 and a root mean square error (RMSE) of 4.79, indicating that the method has good quantitative accuracy across different infection levels.
[0028] To evaluate the model's generalization ability across different sample subsets, this invention conducted a five-fold cross-validation analysis on all samples. The mean RMSE of the cross-validation was 4.82, and the standard deviation was 1.02, indicating that the model performs stably under different training and testing partitions, does not depend on specific sample combinations, and has good robustness and generalizability.
[0029] In summary, the results of this invention demonstrate that the method not only enables continuous quantitative assessment of powdery mildew infection levels but also significantly outperforms traditional manual visual grading methods in terms of accuracy, stability, and repeatability. Furthermore, this method can automate the processing of large batches of samples, making it suitable for high-throughput disease resistance screening and breeding material evaluation, thus providing a reliable and efficient phenotypic assessment tool for grape powdery mildew resistance research and variety breeding. Attached Figure Description
[0030] Figure 1 The structure of powdery mildew spores and hyphae observed under a stereomicroscope shows the morphology of the pathogen's attachment and spread on the leaf surface.
[0031] Figure 2 A schematic diagram of powdery mildew disease severity classification based on the VIVC OIV455 standard, with levels 1 to 9 representing the degree of infection from mild to severe, used for field artificial disease resistance assessment and variety classification.
[0032] Figure 3 Grape powdery mildew inoculation using the sedimentation tower method: A uniformly growing and infectious spore source was obtained through single spore live culture, with leaf disc diameter of 3.0 cm and 5 replicates per sample.
[0033] Figure 4 After inoculation, the spores were observed under a stereomicroscope to be evenly distributed.
[0034] Figure 5 ImageJ-Weka image segmentation model training and prediction: The training set consists of 8 representative samples with different infection resistances, and the test set consists of 222 natural population samples.
[0035] Figure 6 The formula for calculating the phenotypic value of powdery mildew and the final prediction results are presented.
[0036] Figure 7 Regression relationship between image recognition prediction and manual evaluation of powdery mildew: Images of leaf discs from 222 samples were collected 15 days after inoculation. The proportion of infected area was automatically calculated using a machine learning model and manually evaluated according to the VIVCOIV 455 standard. The results showed that there was significant consistency between the two evaluation methods, with a goodness of fit R² of 0.87, indicating that the method of the present invention has good prediction accuracy. Detailed Implementation
[0037] The technical solution of the present invention will be further described below with reference to the embodiments, but the present invention is not limited to the following embodiments. Various modifications, equivalent substitutions, or improvements can be made by those skilled in the art without departing from the spirit and substance of the present invention, and all such modifications, substitutions, or improvements should fall within the protection scope of the present invention.
[0038] Example 1: A method for assessing powdery mildew in grapes based on image recognition, the specific steps of which are as follows:
[0039] Step 1: Sample collection and leaf disc preparation: Functional leaves with consistent growth status and no mechanical damage were collected from 30 main grape varieties, avoiding aging and diseased parts; leaf discs were prepared using a standard 3cm diameter punch, with 5 complete leaf discs prepared for each variety as biological replicates; the leaf discs were placed evenly in petri dishes containing 1.5% agarose medium with the upper surface facing up; the petri dishes were placed in a climate incubator and pre-cultured for 24 hours under 12h light-dark alternation and 28℃ temperature.
[0040] Step 2: Settling tower device construction and powdery mildew inoculation: Construct a settling tower inoculation device with dimensions of 50cm×30cm×30cm (length×width×height), and fix a 150-mesh nylon net with an aperture of 110μm on the top; In a clean and relatively sealed environment, gently sweep the infected leaf carrying powdery mildew spores onto the nylon net, spread it, and tap and shake it to allow the spores to settle naturally and evenly onto the leaf disc surface to complete the inoculation; After inoculation, place the petri dish under constant temperature and humidity conditions and continue to incubate for 15 days.
[0041] Step 3: Image acquisition and image preprocessing: High-resolution images of leaf discs are acquired using a fixed light source, fixed shooting height and angle; the original images are then subjected to the following steps in sequence: image inversion, extraction of leaf disc regions based on color range, reverse deletion of the culture dish background, grayscale processing, contrast enhancement, image cropping and storage, to obtain a standardized image dataset.
[0042] Step 4: Model training and phenotypic extraction based on ImageJ-Weka: Eight representative leaf disc images with different infection levels were selected as training samples and imported into the Weka image segmentation plugin of ImageJ software. Infected and non-infected areas were manually labeled, and the random forest classification model was trained and optimized. Using the trained model, all images were processed in batches using the ImageJMarco macro command to automatically identify and segment infected areas. The ratio of infected area to total leaf disc area was calculated to obtain the percentage of powdery mildew infection as a quantitative phenotypic value, which was used for comparison of variety resistance and breeding screening.
[0043] Example 2: A method for assessing powdery mildew in grapes based on image recognition, the specific steps of which are as follows:
[0044] Step 1: Sample collection and leaf disc preparation: 50 wild and cultivated grape germplasm samples were selected. Undamaged functional leaves with uniform growth were collected and leaf discs with a diameter of 3 cm were prepared according to the method in Example 1. Five biological replicates were made for each sample. The samples were placed in 1.5% agarose medium and pre-cultured at 28°C for 12 hours with alternating light and dark for 24 hours to unify the initial physiological state of the samples.
[0045] Step 2: Spore inoculation was carried out using a sedimentation tower of the same size as in Example 1 and a 150-mesh nylon net. The spores were uniformly settled and inoculated in a closed environment without airflow interference. After inoculation, the spores were cultured at a constant temperature and humidity for 15 days to ensure the stable occurrence and development of the disease.
[0046] Step 3: Image acquisition and preprocessing: Obtain high-resolution images with fixed shooting parameters, and use Photoshop's batch processing function to complete the initial segmentation and standardization preprocessing of the leaf disc region to eliminate the influence of lighting and background differences on the recognition results.
[0047] Step 4: Model Prediction and Accuracy Verification: Using the classification model trained in Example 1, 222 sample images were processed in batches to automatically extract the phenotypic value of powdery mildew infection percentage. After testing, the Spearman correlation coefficient between the automatic phenotypic value and the manual observation results was ρ=-0.935, the coefficient of determination R²=0.872, the mean absolute error MAE=2.72, and the root mean square error RMSE=4.79. This can accurately quantify the degree of powdery mildew infection in different germplasm resources and be used for genetic analysis of disease resistance.
[0048] Example 3: A method for assessing powdery mildew in grapes based on image recognition, the specific steps of which are as follows:
[0049] Step 1: Sample collection and leaf disc preparation: Healthy functional leaves were collected from 10 grape hybrid breeding progeny materials. Leaf discs were prepared using a standard 3cm diameter punch, with 5 complete leaf discs for each sample as biological replicates. The discs were placed in 1.5% agarose medium and pre-cultured at 28℃ for 24h to reduce resistance differences caused by mechanical damage.
[0050] Step 2: Spore inoculation was carried out using the same sedimentation tower and 150-mesh nylon netting as in Example 1, and the spores were uniformly inoculated in a clean and sealed environment. After inoculation, the spores were cultured at a constant temperature and humidity for 15 days to ensure consistent infection conditions.
[0051] Step 3: Rapid Image Preprocessing: The original captured image can be quickly inverted, background removed, grayscale converted, contrast enhanced, and cropped and stored. Standardization processing of a single image is completed within 5 seconds.
[0052] Step 4: Rapid Phenotypic Extraction: Call the trained ImageJ-Weka classification model and process images in batches with one click using macro commands. Phenotypic extraction of a single sample is completed within 1 minute, and the proportion of powdery mildew infection area is automatically output. No manual visual grading is required throughout the process, making it suitable for high-throughput and rapid disease resistance identification scenarios.
[0053] Example 4: A method for assessing grape powdery mildew based on image recognition, with the following steps: A. Standardized inoculation of powdery mildew using the sedimentation tower method. Spore sources with consistent growth and stable infectivity were obtained through single-spore isolation and culture. Leaf discs with a diameter of 3.0 cm were placed in petri dishes for simultaneous inoculation, with 5 biological replicates per sample. B. Post-inoculation observation under a stereomicroscope showed that powdery mildew spores were uniformly distributed on the leaf disc surface, indicating consistent inoculation intensity. C. Image segmentation model training and batch prediction based on ImageJ-Weka. Eight representative leaf disc images with different infection levels were selected as the training set. Diseased and non-disease areas were manually labeled, and the model was trained. Automated phenotypic prediction was then performed on 222 natural population samples. D. Calculation method and prediction results of powdery mildew phenotypic values. Continuous quantitative phenotypic indicators were obtained by calculating the ratio of the infected area to the total leaf disc area, used for subsequent disease resistance assessment and genetic analysis.
[0054] It should be understood that the above-described embodiments are merely preferred embodiments of the present invention, used to illustrate the technical solutions of the present invention, and not to limit the scope of protection of the present invention. For those skilled in the art, various modifications, equivalent substitutions, or improvements can be made to the above embodiments without departing from the essence and principle of the technical solutions of the present invention, and such modifications, equivalent substitutions, or improvements should all fall within the scope of protection of the present invention.
Claims
1. A method for assessing powdery mildew in grapes based on image recognition, characterized in that: The image recognition-based method for assessing grape powdery mildew specifically includes the following steps: Step 1: Sample collection and leaf disc preparation: Functional leaves with consistent growth status and no mechanical damage were collected from the grapevines to be evaluated. Leaf discs with a diameter of 3 cm were prepared on the leaves using a standard perforator. At least 4 to 6 complete leaf discs were prepared for each sample as biological replicates. The leaf discs were then placed evenly on a petri dish containing nutrient medium with the upper surface facing up. The petri dish containing the leaf discs was then placed in a climate incubator for pre-culture. Step 2: Settling tower assembly and powdery mildew inoculation: A sedimentation tower inoculation device was constructed, with a nylon net fixed at the top of the sedimentation tower. Infected leaves carrying a large number of powdery mildew spores were placed in the top area of the sedimentation tower, and the spores were gently and evenly brushed onto the nylon net. Then, the brush was used to evenly spread the spores on the surface of the nylon net, and the net was tapped and shaken. The pathogen spores were evenly distributed and settled onto the surface of the leaf discs in the culture dish through natural sedimentation, thereby achieving synchronous and uniform inoculation of powdery mildew. After inoculation, the culture dish was placed under constant temperature and humidity conditions for further cultivation. Step 3: Image acquisition and image preprocessing: Fifteen days after inoculation and culture, uniform image acquisition was performed on the leaf discs in the culture dish. High-resolution images were obtained using a fixed light source, fixed shooting height and shooting angle. The original images were then preprocessed, and the leaf disc regions were initially segmented using Photoshop's batch processing function to obtain a standardized image dataset suitable for model training and batch analysis. Step 4: Model training and phenotypic extraction based on ImageJ-Weka: Eight representative leaf disc images were selected from the preprocessed images as training samples and imported into the Weka image segmentation plugin in the ImageJ software platform. The random forest classification model was trained and its parameters were optimized by manually labeling the powdery mildew-infected and non-infected areas. Finally, a classification model for automatic identification of lesion areas was constructed. Using the trained classification model, the remaining 222 images were batch-processed using ImageJ Marco macros to achieve automatic identification and segmentation of powdery mildew-infected areas. The area of the infected area and the total area of the leaf discs were calculated separately to obtain the percentage of powdery mildew infection as a quantitative phenotypic value.
2. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: In step 1, avoid collecting leaves from parts that are obviously old or have serious disease.
3. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: In step 1, at least 5 complete leaf discs are prepared for each sample as biological replicates.
4. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: In step 1: the nutrient culture medium is 1.5% agarose medium by mass.
5. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: In step 1, the petri dish containing the leaf discs was placed in a climate incubator and pre-cultured for 24 hours under 12-hour light-dark alternation and a temperature of 28°C.
6. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: In step 2, the dimensions of the settling tower inoculation device are length × width × height = 50cm × 30cm × 30cm.
7. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: In step 2, the nylon mesh is a 150-mesh nylon mesh with an aperture of 110μm.
8. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: In step 2, the entire inoculation process is carried out in a clean, relatively enclosed environment.
9. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: Step 3 involves preprocessing the original image, including image inversion, extraction of leaf disc regions based on color range, reverse deletion of the culture dish background, image grayscale conversion, contrast enhancement, and image cropping and storage.
10. The image recognition-based method for assessing powdery mildew in grapes according to claim 1, characterized in that: In step 4, using 222 independent samples as validation subjects, the automatically extracted phenotypic values and the manually observed results showed a highly consistent monotonic correlation, with a Spearman correlation coefficient of ρ = −0.935 (P = 5.564 × 10⁻¹). 0 ¹), thus stably reflecting the relative differences in the degree of powdery mildew infection among different samples.