Boundary sample quantile based adaptive recognition method for grape downy mildew

Abstract

The application discloses a grape powdery mildew adaptive recognition method based on boundary sample quantile, relates to the technical field of agricultural lesion recognition, and comprises the following steps: acquiring a grape leaf original image, mapping and converting the grape leaf original image into a leaf grayscale image through a grayscale mapping method; performing boundary recognition on the leaf grayscale image to obtain leaf lesion boundary information; determining a lesion threshold of the leaf lesion boundary information based on a sample quantile extraction method; determining pixels less than the lesion threshold as lesions according to the lesion threshold of each pixel point of the leaf grayscale image, and obtaining an image lesion recognition result; and calculating a lesion proportion of the image lesion recognition result based on a statistical method. The application solves the technical problems of slow lesion recognition speed and low recognition accuracy of grape powdery mildew in the prior art, and achieves the technical effects of improving the lesion recognition speed and recognition accuracy of grape powdery mildew.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural disease spot recognition, and particularly relates to a grape downy mildew self-adaptive recognition method based on boundary sample quantile. BACKGROUND

[0002] Grape downy mildew is a common disease affecting grape yield, and its pathogen is grape peronospora, which is a kind of grape specific parasitic fungus and can circulate multiple times. During the growth of grape leaves, if the conditions are suitable, grape peronospora will multiply rapidly and cause widespread damage, which even affects the development of the grape industry. The propagation of grape downy mildew on leaves is continuous and process-based, which is reflected in the derivation of disease spots on leaves. Monitoring the size of disease spots is a way to estimate the degree of downy mildew.

[0003] However, the current disease spot monitoring mostly uses traditional machine learning methods, which still has the problems of slow calculation speed and low recognition accuracy. SUMMARY

[0004] The present application provides a grape downy mildew self-adaptive recognition method and system based on boundary sample quantile, which is used to solve the technical problems of slow disease spot recognition speed and low recognition accuracy of grape downy mildew in the prior art.

[0005] In a first aspect, the present application provides a grape downy mildew self-adaptive recognition method based on boundary sample quantile, which comprises: obtaining a grape leaf original image, and mapping and converting the grape leaf original image into a leaf grayscale image through a grayscale mapping method; performing boundary recognition on the leaf grayscale image to obtain leaf disease spot boundary information; determining a disease spot threshold of the leaf disease spot boundary information based on a sample quantile extraction method; determining pixels less than the disease spot threshold as disease spots according to the disease spot threshold to obtain an image disease spot recognition result; and calculating a disease spot proportion of the image disease spot recognition result based on a statistical method.

[0006] In a second aspect of the present application, an adaptive powdery mildew recognition system based on boundary sample quantile is provided, and the system comprises: a leaf gray image acquisition module, which is configured to acquire a grape leaf original image, and convert the grape leaf original image into a leaf gray image through a gray mapping method; a leaf lesion boundary information acquisition module, which is configured to perform boundary recognition on the leaf gray image to obtain leaf lesion boundary information; a leaf lesion boundary threshold determination module, which is configured to determine a lesion threshold of the leaf lesion boundary information based on a sample quantile extraction method; an image lesion recognition result acquisition module, which is configured to determine pixels less than the lesion threshold as lesions according to the lesion threshold to obtain an image lesion recognition result; and a lesion proportion acquisition module, which is configured to calculate a lesion proportion of the image lesion recognition result based on a statistical method.

[0007] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0008] The adaptive powdery mildew recognition method based on boundary sample quantile provided in the present application relates to the technical field of agricultural lesion recognition, converts a grape leaf original image into a leaf gray image through a gray mapping method, performs boundary recognition on the leaf gray image to obtain leaf lesion boundary information, determines a lesion threshold of the leaf lesion boundary information based on a sample quantile extraction method, determines pixels less than the lesion threshold as lesions to obtain an image lesion recognition result, and finally calculates a lesion proportion of the image lesion recognition result based on a statistical method, thereby solving the technical problems of slow lesion recognition speed and low recognition accuracy of powdery mildew in the prior art, and achieving the technical effect of improving the lesion recognition speed and recognition accuracy of powdery mildew. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort.

[0010] Figure 1 The adaptive powdery mildew recognition method based on boundary sample quantile provided in the present application relates to the technical field of agricultural lesion recognition, converts a grape leaf original image into a leaf gray image through a gray mapping method, performs boundary recognition on the leaf gray image to obtain leaf lesion boundary information, determines a lesion threshold of the leaf lesion boundary information based on a sample quantile extraction method, determines pixels less than the lesion threshold as lesions to obtain an image lesion recognition result, and finally calculates a lesion proportion of the image lesion recognition result based on a statistical method, thereby solving the technical problems of slow lesion recognition speed and low recognition accuracy of powdery mildew in the prior art, and achieving the technical effect of improving the lesion recognition speed and recognition accuracy of powdery mildew.

[0011] Figure 2A schematic diagram illustrating the process of mapping the original image of grape leaves into a grayscale image in the adaptive identification method for grape downy mildew based on boundary sample quantiles provided in this application embodiment;

[0012] Figure 3 A flowchart illustrating the process of obtaining leaf lesion boundary information in the adaptive identification method for grape downy mildew based on boundary sample quantiles provided in this application embodiment;

[0013] Figure 4 A schematic diagram of the structure of the adaptive identification system for grape downy mildew based on boundary sample quantiles provided in this application embodiment.

[0014] Figure labeling: Leaf grayscale image acquisition module 11, Leaf lesion boundary information acquisition module 12, Leaf lesion boundary threshold determination module 13, Image lesion recognition result acquisition module 14, Lesion proportion acquisition module 15. Detailed Implementation

[0015] This application provides an adaptive identification method for grape downy mildew based on boundary sample quantiles, which solves the technical problems of slow identification speed and low identification accuracy of grape downy mildew lesions in the prior art.

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] Example 1

[0019] like Figure 1 As shown, this application provides an adaptive identification method for grape downy mildew based on boundary sample quantiles, the method comprising:

[0020] S100: acquire a grape leaf original image, and map the grape leaf original image to a leaf gray scale image by a gray scale mapping method;

[0021] Specifically, the grape leaf original image is captured by using a camera or the like, and the leaf original image is an RGB color image. Each image of the RGB color image is composed of three gray scale images, so that it is difficult to carry out digital analysis on the color image. In order to simplify the identification problem and facilitate the research on disease spot identification, the grape leaf original image is converted to a leaf gray scale image by using a gray scale mapping method,

[0022] Further, as shown in Figure 2 the embodiment of the present application further comprises:

[0023] S110: acquire RGB color space description information of the grape leaf original image;

[0024] S120: construct a gray scale image conversion formula: Gray=R*0.299+G*0.587+B*0.114, wherein Gray is a pixel gray scale value, and R, G and B are red, green and blue color coordinates, respectively;

[0025] S130: calculate and convert the RGB color space description information based on the gray scale image conversion formula, and generate the leaf gray scale image.

[0026] Specifically, the grape leaf original image is an RGB color image. For the RGB color image, red, green and blue are three basic primary colors, which are defined as a space according to the color recognition of the human eye, and can represent most colors. The red, green and blue channels are set as X, Y and Z axes in a Cartesian coordinate system, the positions of all color points of the grape leaf original image in the coordinate system are found, and the RGB color space description information of the grape leaf original image is obtained.

[0027] Since the perception of color by the human body is derived from the visual cortex and associative areas of the brain, the three types of cone cells, short (S), medium (M) and long (L) cone cells, have different sensitivities to three different spectra, so the perceived level of RGB color is not the same. Therefore, different weights need to be given to different colors in the gray scale conversion process. Based on the visual sensitivity theory, a psychological formula Gray=R*0.299+G*0.587+B*0.114 is used as a gray scale image conversion formula to convert the color image. In the gray scale image conversion formula, Gray is a pixel gray scale value, and R, G and B are red, green and blue color coordinates, respectively. Based on the gray scale image conversion formula, the three colors of the leaf color image are weighted and calculated respectively, and are mapped to a one-dimensional numerical pixel representation of the gray scale image in the gray scale space, which is used as the leaf gray scale image to facilitate subsequent identification of the leaf disease spot boundary.

[0028] S200: boundary recognition is performed on the leaf gray image to obtain leaf lesion boundary information;

[0029] Specifically, due to the differences in brightness, contrast and other factors between different images, there are also differences in gray values after different images are converted into gray images, and it is difficult to find a fixed gray threshold as the basis for lesion determination. Therefore, the application adopts a boundary recognition method to perform boundary recognition on the leaf gray image, finds the boundary between the lesion and the normal leaf tissue in the leaf gray image, and the boundary is the transition layer between the normal tissue and the lesion, which is also between the two types of tissues in the gray value and can be used as the basis for subsequent lesion threshold determination.

[0030] Further, as shown in Figure 3 the step S200 of the embodiment of the application further comprises:

[0031] S210: obtaining a Gaussian filter, and performing edge recognition on the leaf gray image by using the Gaussian filter to obtain a leaf edge recognition result;

[0032] S220: optimizing the leaf edge recognition result by using a non-maximum suppression method to obtain a leaf edge optimized image;

[0033] S230: screening the leaf edge optimized image based on a double threshold method to obtain the leaf lesion boundary information.

[0034] Specifically, a Gaussian filter is obtained, and edge recognition is performed on the leaf gray image by using the Gaussian filter. The Gaussian filter is a denoising filter algorithm. The gray values of a pixel point and its neighborhood points to be filtered are weighted and averaged according to the parameter rules generated by the Gaussian formula, and the high-frequency noise superimposed in the image is filtered. The Gaussian filter is a smoothing filter that selects weights according to a Gaussian function, and is essentially a weighted filter. For example, the Gaussian filter used in this application has a pyramid structure in numerical distribution. The value of the filter can be understood as a weight. The larger the value, the greater the weight of the corresponding pixel point, and the greater the proportion. The gray values of the pixel points and their neighborhood points in the leaf gray image are weighted and averaged by using the Gaussian filter, the high-frequency noise superimposed in the image is filtered, the edge of the lesion is recognized, and the leaf edge recognition result is obtained. Further, the local gray maximum value of the leaf edge pixel point is found by using the non-maximum suppression method, the pixel of the maximum value is set to 1, and the pixels of other non-maximum values are set to 0, so as to achieve the purpose of sharpening the edge and reducing the interference, and obtain the leaf edge optimization image. The leaf edge optimization image is screened by using a double-threshold method, the pixel value of the pixel point with a too small gray value is set to 0, the pixel value of the pixel point with a relatively large gray value is set to 1, and the pixel point with a relatively large gray value is retained as the leaf lesion boundary information, so as to remove the interference caused by the pseudo edge on the edge determination.

[0035] Further, the step S220 of the embodiment of the application further includes:

[0036] S221: Obtain a target suppression point and connected neighborhood points of the leaf edge recognition result;

[0037] S222: Calculate the target suppression point based on an angle image function to obtain a gradient direction;

[0038] S223: Obtain a gray value size comparison result of the target suppression point and the connected neighborhood points;

[0039] S224: When the gray value size comparison result is that the gray value of the target suppression point is the maximum, first and second gradient intersection points of the gradient direction are obtained;

[0040] S225: If the gray value of the target suppression point is greater than the gray values of the first and second gradient intersection points, the target suppression point is a maximum value point, the pixel of the maximum value point is set to 1, and the pixels of non-maximum value points are set to 0, to obtain the leaf edge optimization image.

[0041] Specifically, a target suppression point is selected from the leaf edge recognition result, and eight connected neighborhood points in the neighborhood are found. The target suppression point is a point that is currently not suppressed by the non-maximum value. Based on an angle image function The target suppression point is calculated to obtain a gradient direction, that is, a non-maximum suppression direction. The gray value of the target suppression point and the connected neighborhood points are compared. When the comparison result is that the gray value of the target suppression point is the largest, the first gradient intersection point and the second gradient intersection point in the gradient direction are extracted. Then, the gray value of the target suppression point and the first gradient intersection point and the second gradient intersection point are compared. When the gray value of the target suppression point is greater than the gray value of the first gradient intersection point and the second gradient intersection point, the target suppression point is determined as a maximum value point. The pixel of the maximum value point is set to 1, and the pixel of a non-maximum value point is set to 0, so as to achieve the purposes of sharpening edges and reducing interference. The leaf edge optimization image is obtained, and the problems of coarse, wide and much interference of the edges after the leaf spot edge is recognized by using the Gaussian filter can be solved.

[0042] Further, the step S230 of the embodiment of the present application further includes:

[0043] S231: selecting an edge low threshold value and an edge high threshold value;

[0044] S232: setting the pixels smaller than the edge low threshold value in the leaf edge optimization image to 0 and setting the pixels greater than the edge high threshold value to 1 to obtain the leaf spot boundary information.

[0045] Specifically, two edge gray threshold values are selected. The selected edge gray threshold values should make the best separation between different classes. Exemplarily, first, the occurrence probability of each segmentation characteristic value is obtained based on a histogram, and the segmentation characteristics are divided into two classes by a threshold variable. Then, the intra-class variance and the inter-class variance of each class are calculated. The edge low threshold value and the edge high threshold value are selected as the best threshold values when the intra-class variance is the smallest or the inter-class variance is the largest. The edge gray values in the leaf edge optimization image are compared with the edge low threshold value and the edge high threshold value one by one. If the edge gray value is smaller than the edge low threshold value, it is determined as a false edge, and the pixel value is set to 0. If the edge gray value is greater than the edge high threshold value, it is determined as a strong edge, and the pixel value is set to 1. Finally, the edge connection method based on the double threshold values is used to connect the edges into a contour until the entire image is closed, so as to obtain the leaf spot boundary information, and the interference caused by the false edges on the edge determination can be removed.

[0046] S300: determining a spot threshold value of the leaf spot boundary information based on a sample quantile extraction method;

[0047] Specifically, the sample quantile refers to a sample feature corresponding to the population quantile, and can reflect a mathematical feature of statistical data of a concentrated position of a certain proportion of data. In order to suppress the interference of the existing stripes and boundaries of the blade on the determination of the threshold value, the sample quantile extraction method is used to find the gray value that can most represent the difference between the normal tissue and the disease spot of the blade in probability, as the disease spot threshold value.

[0048] Further, the step S300 of the embodiment of the application further comprises:

[0049] S310: obtaining an order statistic of the leaf disease spot boundary information;

[0050] S320: setting a boundary sample quantile;

[0051] S330: dividing and extracting the order statistic based on the boundary sample quantile to determine the disease spot threshold value.

[0052] Specifically, let X1X2,…,X n be a sample from the leaf disease spot boundary information, X (1) ≤X (2) ≤…≤X (n) be an order statistic thereof. The p quantile of the sample is defined as: m p =X (k) , where k=[np]+1, and [np] is an integer operation.

[0053] Let X1X2,…,X n be a sample from a population with density function f(x), given p∈(0,1), f(x) is continuous at the p quantile of the population ξ p , and f(ξ p )>0. Define k, such that Then the kth order statistic X (k) of the sample has:

[0054]

[0055] where, represents a high-order infinitesimal, and L represents a distribution convergence.

[0056] The identified boundary pixels are statistically analyzed to obtain a numerical distribution of the boundary points that tends to approach, a boundary sample quantile is set, that is, a sample data distribution ratio is selected, a lesion threshold is determined according to the sample quantile, and an exemplary preferred 50% quantile is selected. The order statistics before 50% are extracted based on the boundary sample quantile, and a pixel gray value corresponding to the 50% order statistics is taken as a lesion threshold. The pixel points corresponding to the order statistics before 50% are the lesions. The lesion threshold can be used to distinguish the gray value of the normal tissue of the leaf and the lesion.

[0057] S400: determining each pixel point of the leaf gray scale image according to the lesion threshold, determining the pixel point less than the lesion threshold as a lesion, and obtaining an image lesion recognition result;

[0058] Specifically, the gray value of each pixel point of the leaf gray scale image is compared with the lesion threshold one by one, and the pixel point with a gray value less than the lesion threshold in the leaf gray scale image is screened out. The image area composed of the screened pixel points is determined as a lesion area, which is used as an image lesion recognition result and can be used to calculate the lesion proportion of the leaf.

[0059] S500: calculating the lesion proportion of the image lesion recognition result based on a statistical method.

[0060] Specifically, based on a statistical algorithm, the area of the lesion in the image lesion recognition result of all grape leaf samples is divided by the total area of the leaf to obtain the area proportion of the lesion of each grape leaf sample, which is used as a lesion proportion and can be used to evaluate the health status of the plant.

[0061] Further, the embodiment of the present application further includes step S600, and step S600 further includes:

[0062] S610: randomly identifying and verifying the grape leaf original image by using a Monte Carlo method to obtain image lesion recognition verification information;

[0063] S620: determining the grape leaf original image by using an artificial calibration method to obtain image lesion calibration information;

[0064] S630: comparing the image lesion recognition verification information and the image lesion calibration information to obtain image lesion recognition accuracy.

[0065] Specifically, the Monte Carlo method is also called statistical simulation method, statistical test method. It is a numerical simulation method taking probability phenomenon as the research object, and is a calculation method for estimating unknown characteristic quantities by obtaining statistical values through sampling survey method. For example, by means of random number generation, N pixel points are randomly selected on the picture, combined with artificial experience, it is determined whether the N points are disease spots or not, and the position information of the image disease spots and normal is obtained by the method and the control method respectively. By comparing the determination results of N points by artificial determination and identification method, the recognition accuracy is calculated.

[0066] Further, in order to compare the effectiveness of the gray threshold determination method proposed in this paper, this application adopts random threshold determination, median threshold determination, threshold determination based on normal tissue and threshold determination based on disease tissue as a control to compare the influence of different threshold determination methods on disease recognition accuracy. The recognition accuracy results of different methods are shown in Table 1. From the results in Table 1, it can be found that the disease recognition threshold determination method proposed in this application has obviously higher accuracy than the control group, which shows that using the number of disease boundary samples as the threshold for disease recognition can adaptively extract the difference information between disease and normal tissue, thereby providing a basis for disease determination.

[0067] Table 1: Recognition accuracy longitudinal comparison

[0068]

[0069] Further, in order to compare the effectiveness of the gray threshold determination method proposed in this paper, this application adopts random threshold determination, median threshold determination, threshold determination based on normal tissue and threshold determination based on disease tissue as a control to compare the influence of different threshold determination methods on disease recognition accuracy. The recognition accuracy results of different methods are shown in Table 1. From the results in Table 1, it can be found that the disease recognition threshold determination method proposed in this application has obviously higher accuracy than the control group, which shows that using the number of disease boundary samples as the threshold for disease recognition can adaptively extract the difference information between disease and normal tissue, thereby providing a basis for disease determination.

[0070] Table 2: Recognition accuracy horizontal comparison

[0071] Method Recognition accuracy Average time (s) The method of the present application 92.2％ 1.410 BP neural network 78.6％ 8.763 Convolutional neural network 91.8％ 12.348 VGG16 model 94.3％ 15.588 Support vector machine 88.7％ 6.533 ResNet50 model 96.26％ 20.317

[0072] In summary, the embodiments of the present application have at least the following technical effects:

[0073] The present application maps the original grape leaf image into a leaf grayscale image by a grayscale mapping method, and identifies the boundary to obtain leaf lesion boundary information. Based on sample quantile extraction, the lesion threshold of the leaf lesion boundary information is determined, and the pixel points less than the lesion threshold are determined as lesions to obtain an image lesion identification result. Finally, the lesion proportion of the image lesion identification result is calculated based on a statistical method.

[0074] The technical effect of improving the lesion identification speed and accuracy of grape downy mildew is achieved.

[0075] Embodiment Two

[0076] Based on the same inventive concept as the adaptive identification method of grape downy mildew based on boundary sample quantiles in the foregoing embodiments, as shown in Figure 4 The present application provides an adaptive identification system of grape downy mildew based on boundary sample quantiles, and the system and method embodiments in the present application are based on the same inventive concept. The system comprises:

[0077] A leaf grayscale image acquisition module 11 is configured to acquire an original grape leaf image, and map the original grape leaf image into a leaf grayscale image by a grayscale mapping method.

[0078] A leaf lesion boundary information acquisition module 12 is configured to identify the boundary of the leaf grayscale image to obtain leaf lesion boundary information.

[0079] A leaf lesion boundary threshold determination module 13 is configured to determine the lesion threshold of the leaf lesion boundary information based on sample quantile extraction.

[0080] An image lesion identification result acquisition module 14 is configured to determine each pixel point of the leaf grayscale image according to the lesion threshold, determine the pixel points less than the lesion threshold as lesions, and obtain an image lesion identification result.

[0081] A lesion proportion acquisition module 15 is configured to calculate the lesion proportion of the image lesion identification result based on a statistical method.

[0082] Further, the system further comprises:

[0083] The color space description information acquisition module is configured to acquire RGB color space description information of the grape leaf original image.

[0084] The gray image conversion formula construction module is configured to construct a gray image conversion formula: Gray=R*0.299+G*0.587+B*0.114, where Gray is a pixel gray value, and R, G and B are red, green and blue color coordinates, respectively.

[0085] The leaf gray image generation module is configured to calculate and convert the RGB color space description information based on the gray image conversion formula to generate the leaf gray image.

[0086] Further, the system further comprises:

[0087] The leaf edge recognition result acquisition module is configured to obtain a Gaussian filter, perform edge recognition on the leaf gray image by using the Gaussian filter, and acquire a leaf edge recognition result.

[0088] The leaf edge optimized image acquisition module is configured to optimize the leaf edge recognition result by using a non-maximum suppression method to obtain a leaf edge optimized image.

[0089] The leaf lesion boundary information acquisition module is configured to filter the leaf edge optimized image based on a double-threshold method to obtain leaf lesion boundary information.

[0090] Further, the system further comprises:

[0091] The target suppression point acquisition module is configured to acquire a target suppression point and a connected neighborhood point of the leaf edge recognition result.

[0092] The gradient direction acquisition module is configured to calculate the target suppression point based on an angle image function to obtain a gradient direction.

[0093] The gray value size comparison module is configured to obtain a gray value size comparison result of the target suppression point and the connected neighborhood point.

[0094] The gradient intersection point acquisition module is configured to obtain a first gradient intersection point and a second gradient intersection point of the gradient direction when the gray value size comparison result is that the gray value of the target suppression point is the largest.

[0095] The blade edge optimization image obtaining module is configured to, if the gray value of the target suppression point is greater than the gray value of the first gradient intersection point and the second gradient intersection point, the target suppression point is a maximum value point, pixels of the maximum value point are set to 1, and pixels of non-maximum value points are set to 0, to obtain the blade edge optimization image.

[0096] Further, the system further comprises:

[0097] The edge threshold value selecting module is configured to select an edge low threshold value and an edge high threshold value.

[0098] The blade lesion boundary information obtaining module is configured to set pixels less than the edge low threshold value in the blade edge optimization image to 0 and set pixels greater than the edge high threshold value to 1 to obtain the blade lesion boundary information.

[0099] Further, the system further comprises:

[0100] The order statistic obtaining module is configured to obtain an order statistic of the blade lesion boundary information.

[0101] The boundary sample quantile obtaining module is configured to set a boundary sample quantile.

[0102] The lesion threshold value determining module is configured to divide and extract the order statistic based on the boundary sample quantile to determine the lesion threshold value.

[0103] Further, the system further comprises:

[0104] The image lesion identification verification information obtaining module is configured to perform random identification verification on the grape leaf original image by using a Monte Carlo method to obtain image lesion identification verification information.

[0105] The image lesion calibration information obtaining module is configured to determine the grape leaf original image by using an artificial calibration method to obtain image lesion calibration information.

[0106] The image lesion identification precision obtaining module is configured to compare the image lesion identification verification information and the image lesion calibration information to obtain image lesion identification precision.

[0107] It should be noted that the above-mentioned embodiment sequences of the present application are merely for description only, but not for representing the advantages and disadvantages of the embodiments. And the above-mentioned embodiments of the present specification are described. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than the order in which they are recited and still achieve desirable results. In addition, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous or necessary.

[0108] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0109] The specification and drawings are merely exemplary of the present application, and are considered to cover any and all modifications, variations, combinations or equivalents that fall within the scope of the present application. Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is intended to include these modifications and variations.

Claims

1. A method for adaptive recognition of grape downy mildew based on boundary sample quantile, characterized in that, The method comprises: Obtaining a grape leaf original image, and mapping and converting the grape leaf original image into a leaf gray image through a gray mapping method; Performing boundary recognition on the leaf gray image to obtain leaf lesion boundary information; Determining a lesion threshold of the leaf lesion boundary information based on a sample quantile extraction method; Determining each pixel point of the leaf gray image according to the lesion threshold, determining a pixel point less than the lesion threshold as a lesion, and obtaining an image lesion recognition result; Calculating a lesion ratio of the image lesion recognition result based on a statistical method; The leaf lesion boundary information comprises: Performing edge recognition on the leaf gray image by using a Gaussian filter to obtain a leaf edge recognition result; Optimizing the leaf edge recognition result by using a non-maximum suppression method to obtain a leaf edge optimization image; Screening the leaf edge optimization image based on a double-threshold method to obtain the leaf lesion boundary information; The sample quantile extraction method for determining the lesion threshold of the leaf lesion boundary information comprises: Obtaining an order statistic of the leaf lesion boundary information; Setting a boundary sample quantile; Dividing and extracting the order statistic based on the boundary sample quantile to determine the lesion threshold.

2. The method of claim 1, wherein, The method for mapping and converting the grape leaf original image into the leaf gray image through the gray mapping method comprises: Obtaining RGB color space description information of the grape leaf original image; The conversion formula of gray image is constructed as follows: , wherein Gray is the pixel gray value, R, G and B are red, green and blue color coordinates, respectively. Calculating and converting the RGB color space description information based on a gray image conversion formula to generate the leaf gray image.

3. The method of claim 1, wherein, The method for obtaining the leaf edge optimization image comprises: Obtaining a target suppression point and a connected neighborhood point of the leaf edge recognition result; Calculating the target suppression point based on an angle image function to obtain a gradient direction; Obtaining a gray value size comparison result of the target suppression point and the connected neighborhood point; When the gray value size comparison result is a maximum gray value of the target suppression point, first and second gradient intersection points of the gradient direction are obtained; If the gray value of the target suppression point is greater than the gray values of the first and second gradient intersection points, the target suppression point is a maximum value point, pixels of the maximum value point are set to 1, and pixels of a non-maximum value point are set to 0 to obtain the leaf edge optimization image.

4. The method of claim 1, wherein, The method for screening the leaf edge optimization image based on the double-threshold method comprises: Selecting an edge low threshold and an edge high threshold; Setting pixels less than the edge low threshold in the leaf edge optimization image to 0 and setting pixels greater than the edge high threshold to 1 to obtain the leaf lesion boundary information.

5. The method of claim 1, wherein, The method comprises: Performing random recognition verification on the grape leaf original image by using a Monte Carlo method to obtain image lesion recognition verification information; Determining the grape leaf original image by using an artificial calibration method to obtain image lesion calibration information; Comparing the image lesion recognition verification information and the image lesion calibration information to obtain image lesion recognition accuracy.

6. An adaptive recognition system for grape downy mildew based on boundary sample quantiles, characterized in that, The system comprises: A leaf gray scale map acquisition module is configured to acquire a grape leaf original image and convert the grape leaf original image into a leaf gray scale map through a gray scale mapping method. A leaf disease spot boundary information acquisition module is configured to perform boundary identification on the leaf gray scale map and acquire leaf disease spot boundary information. A leaf disease spot boundary threshold value determination module is configured to determine a disease spot threshold value of the leaf disease spot boundary information based on a sample quantile extraction method. An image disease spot identification result acquisition module is configured to determine pixels less than the disease spot threshold value as disease spots according to the disease spot threshold value and acquire an image disease spot identification result. A disease spot proportion acquisition module is configured to calculate a disease spot proportion of the image disease spot identification result based on a statistical method. The leaf disease spot boundary information acquisition module is configured to perform: Edge identification on the leaf gray scale map through a Gaussian filter to acquire a leaf edge identification result; Optimization of the leaf edge identification result through a non-maximum suppression method to acquire a leaf edge optimized image; Screening of the leaf edge optimized image based on a double threshold method to acquire the leaf disease spot boundary information. The leaf disease spot boundary threshold value determination module is configured to perform: Acquisition of order statistics of the leaf disease spot boundary information; Setting of an obtained boundary sample quantile; Division and extraction of the order statistics based on the boundary sample quantile to determine the disease spot threshold value.

Images