Idesia polycarpa disease and insect pest monitoring method and system

Abstract

The invention relates to the technical field of disease and insect pest monitoring, solves the technical problems of poor segmentation performance, excessive segmentation and edge detail loss in the prior art, and particularly relates to an idesia disease and insect pest monitoring method and system, which comprises the following steps: collecting an idesia remote sensing image, and preprocessing the remote sensing image to obtain a preprocessed image; according to the idesia range image segmentation method, through the improved fuzzy clustering method, the problems that in the prior art, adjacent pixels are divided into the same class, the segmentation performance is poor, excessive segmentation exists, edge details are lost and the like can be solved, and the segmentation efficiency is improved. The problem that adjacent pixels are divided into the same kind can be avoided, the segmentation performance can be improved, the problem of excessive segmentation can be reduced, edge details are kept, and the accuracy and the classification efficiency of the classification method are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pest and disease monitoring, and particularly to a monitoring method and system for pests and diseases of Idesia polycarpa. Background Art

[0002] Idesia polycarpa is a deciduous tree widely distributed in the Yangtze River Basin and southern regions of China. It has strong adaptability and high ecological, economic, and energy values. Idesia polycarpa can not only be used for ecological greening, landscape improvement, soil conservation, and water source conservation, but also has a high oil content and a high proportion of unsaturated fatty acids in its fruits, making it a high-quality woody oil and biofuel raw material. In recent years, it has received extensive attention. With the development of the national grain and oil security strategic layout and the bioenergy industry, the planting area of Idesia polycarpa has been continuously expanding, and its pest and disease problems have become increasingly prominent. These problems not only affect the growth of Idesia polycarpa, resulting in a reduction in fruit yield, but also reduce the fruit quality. The main diseases of Idesia polycarpa include root rot, anthracnose, powdery mildew, rust, and sooty mold. In the existing classification methods, the A-IT2FCM method is commonly used to classify and analyze images. Although this method is simple, it only considers the relationship between local neighborhood pixels. When the noise in the image has a large impact, it is easy to divide adjacent pixels into the same class, resulting in poor segmentation performance. At the same time, there are also problems such as over-segmentation and loss of edge details. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the present invention provides a monitoring method and system for pests and diseases of Idesia polycarpa, solves the technical problems of poor segmentation performance, over-segmentation, and loss of edge details in the prior art, and achieves the purpose of improving segmentation performance and avoiding over-segmentation and loss of edge details.

[0004] To solve the above technical problems, the present invention provides the following technical solutions: A monitoring method for pests and diseases of Idesia polycarpa, the method comprising the following steps: S1. Collect remote sensing images of Idesia polycarpa, and preprocess the remote sensing images to obtain multiple noiseless preprocessed images; S2. Perform correction and stitching processing on each preprocessed image to obtain a range image including Idesia polycarpa plants; S3. Calculate the invariant feature value of any pixel point in the range image, and establish a feature matrix based on the invariant feature value ; S4. Calculate the optimized feature value and the healthy feature value of the healthy image of Idesia polycarpa according to the feature matrix, and screen to obtain a pest and disease area set based on the healthy feature value ; S5. Determine the initial center in the pest and disease area concentratedly by the random method And calculate the non-local value for reducing local noise pixels through non-local pixels , based on the non-local value Calculate the first type of weight And the second type of weight ; S6. Calculate the reduced membership value for deblurring the pest and disease area set according to the first type of weight And the second type of weight , and calculate the new center for classifying the pest and disease area set according to the reduced membership value , and calculate the new center for classifying the pest and disease area set according to the reduced membership value ; ; S7. Use the standard image containing the level of Idesia polycarpa pests and diseases as the comparison standard, and divide the pest and disease set into mild, moderate, and severe pests and diseases of different levels

[0005] Furthermore, in step S1, the specific implementation steps are as follows: S11. Split the remote sensing image into an R image, a G image, and a B image using the RGB channels, and respectively obtain the RGB pixel values of the pixel points at the same position in the R image, the G image, and the B image ; , and ; S12. Calculate the grayscale pixel value of the pixel point , and according to the RGB pixel values , and the calculation formula is: ; ; Among them, represents the th grayscale pixel value; S13. Calculate the denoising value of the pixel point according to the grayscale pixel value , and the calculation formula is: ; Among them, represents the standard deviation of the grayscale pixel value ; S14. Repeat step S12 and step S13 until the calculation of the denoising values of all pixel points is completed, and the processed pixel points are synthesized into a preprocessed image

[0006] Furthermore, in step S2, the specific implementation steps are as follows: S21. Establish a partitioning window with a side length of and a step size of on the preprocessed image. Obtain two adjacent rectified images and in the preprocessed image through the partitioning window; S22. Respectively obtain the pixel points and at the same position in the rectified images and , and calculate the deviation angle . The calculation formula is: ; where represents the number of pixel points and ; S23. Obtain the central pixel point in the preprocessed image. Calculate the corrected pixel point according to the deviation angle . The calculation formula is: ; where and represent the pixel values of the pixel points in the th rectified image; S24. Convert the corrected pixel point into the corrected coordinate through the pixel coordinate transformation method, and calculate the ground coordinate value of the pixel point in the preprocessed image. The calculation formula is: ; where represents the rotation matrix of the camera's external parameters, represents the coordinate value of the camera, respectively represent the external parameters of the camera, represents the focal length of the camera, represents the height of the camera; S25. Calculate the image fusion values and and based on the ground coordinate values in multiple preprocessed images. The calculation formula is: where and respectively represent the weight coefficients, and respectively represent the abscissa values of the first and second preprocessed images, and respectively represent the ordinate values of the first and second preprocessed images; S26. Repeat step S25 to fuse multiple preprocessed images into a range image.

[0007] Furthermore, in step S3, the specific implementation steps are as follows: S31. Obtain the pixel value of any pixel point in the range image as the central pixel point, and obtain the radius of the central pixel point within neighborhood pixel points, and calculate the reference value for obtaining the binary result , and the calculation formula is: ; where represents the sign function, represents the gray value of the central pixel point, represents the gray value of the neighborhood pixel point; S32. Change the radius of the central pixel point to , repeat step S31 to obtain multiple reference values , and calculate the invariant eigenvalue according to the multiple reference values , and the calculation formula is: ; where represents circularly shifting the binary result of the reference value to the right by times; S33. Obtain the gray level and of the range image through the gray quantization method, and calculate the gray value and according to the gray level , and the calculation formula is: ; where represents the pixel value of the pixel point in the range image, and respectively represent the offset amounts of the pixel point at different angles, and represent the length and width of the range image; S34. Calculate the contrast value , the feature entropy , and the correlation value of the range image respectively according to the gray value , and the calculation formula is: ; where represents the quantity of and represents the average value of the grey value ; and respectively represent and the standard deviation of; S35. Combine the comparison value , the feature entropy and the correlation value into a multi - vector ; S36. Calculate the specific filtering value based on the pixel value of any pixel point in the range image. The calculation formula is: ; wherein, respectively represent the direction number and scale number of the specific filtering value , represents the standard deviation of the grey value of the range image, represents the wavelength of the pixel value , represents the phase shift of the pixel value ; S37. Calculate the filtering feature based on the specific filtering value . The calculation formula is: ; wherein, represents the convolution calculation between the grey value of the original image and the specific filtering value ; S38. Combine the invariant feature value , the multi - vector and the filtering feature into the feature matrix of the range image.

[0008] Furthermore, in step S4, the specific implementation steps are as follows: S41. Perform importance screening on the feature matrix of the range image through the random forest model and obtain multiple optimized feature values , and obtain the healthy feature value from the healthy Idesia polycarpa image through steps S3 and the random forest model; S42. Calculate the division value based on the optimized feature value . The calculation formula is: ; wherein, Represents the optimized eigenvalue of the gradient value; S43. According to the division value the range image is segmented to obtain multiple regions to be detected , and according to the division value of the region to be detected a set of segmented regions is obtained , and the expression is: ; Among them, represents the th set of segmented regions; S44. Calculate the fluctuation value according to the health eigenvalue , and the calculation formula is: ; Among them, represents the health eigenvalue of the health image of the th Idesia polycarpa; S45. Obtain the difference threshold through detrending analysis , and based on the difference threshold the pest and disease areas are screened out from the set of segmented regions ; If , it means that the optimized eigenvalue is abnormal and is marked as a pest and disease area; If , it means that the optimized eigenvalue is normal and is marked as a healthy area; S46. Combine multiple pest and disease areas into a set of pest and disease areas.

[0009] Furthermore, in step S5, the specific implementation steps are as follows: S51. Randomly select a pixel point in the set of pest and disease areas as the initial center , and define the clustering radius of the initial center as , and randomly and non-repeatedly select two pixel windows and and within the range of , and calculate the window difference ; Among them, represents the smoothing parameter that controls the exponential function decay, represents the logarithm of the pixel windows and , and ; S52. According to the window difference ​​Computing non-local values , the calculation formula is: ; in, Represents a pixel window The central pixel value of S53, get pixel window Gray value of inner pixel , according to the gray value Calculate local variation value , the calculation formula is: ; in, Represents a pixel window Gray value of inner pixel The average of Represents grayscale value and average The number of groups; S54, based on local variation value Calculate association weight , the calculation formula is: ; in, Indicates association weights; S55. According to the association weight Calculate the first regularization factor and the second regularization factor , the calculation formula is: ; in, and Represents the random constants and non-local values ​​of the pixels in the pest area set The random constant, Represents a non-local value The mean of S56. According to the first regularization factor and the second regularization factor Calculate the first class weights and the second type of weight , the calculation formula is: ; in, represents a constant greater than zero, Represents a very small number greater than zero.

[0010] Furthermore, in step S6, the specific implementation steps are as follows: S61. According to the first type of weight and the second type of weight Calculate the first fuzzy factor and the second fuzzy factor , and the calculation formula is: ; Wherein, represents the average spatial Euclidean distance between pixels, and respectively represent the initial lower limit and the initial upper limit, and represent the fuzzy exponent, and ， ， represents the spatial distance relative to the initial center ; S62. According to the first fuzzy factor and the second fuzzy factor calculate the updated upper and lower limit updated membership degrees and , and the calculation formula is: ; Wherein, represents the spatial distance of the update point from the initial center , and represent the fuzzy factors of the pixel point with the gray value of and the initial center , and represent the fuzzy factors of the update point and the initial center , represents the calculation quantity; S63. According to the upper and lower limit updated membership degrees and construct the updated membership degree range , and randomly determine the initial updated membership value in the updated membership degree range ; S64. According to the initial updated membership value calculate the update factor , and the calculation formula is: ; Wherein, and represent random numbers, and , , Represents the cardinality of the set of pest and disease areas; S65. According to the update factor Obtain the reduced membership value ; S66. According to the reduced membership value Calculate the updated new center , and the calculation formula is: ; Among them, Represents the gray value of the pixel point The number of, Represents the New center of the generation; S67. Calculate the iteration threshold by the error square method , and judge whether the iteration stops according to the iteration threshold ; If , then the iteration ends, and according to the new center Classify the set of pest and disease areas to obtain the pest and disease set; If , then let And return to step S61.

[0011] Furthermore, the expression of the reduced membership value Is: ; Among them, Represents the reduction coefficient.

[0012] This technical solution also provides a system for the above-mentioned monitoring method of Idesia polycarpa pests and diseases. The system includes: A preprocessing module for collecting remote sensing images of Idesia polycarpa and preprocessing the remote sensing images to obtain multiple noise-free preprocessed images; A correction and stitching module for performing correction and stitching processing on each preprocessed image to obtain a range image containing Idesia polycarpa plants; A texture module for calculating the invariant feature value Of any pixel point In the range image, and based on the invariant feature value Establish a feature matrix; A preliminary screening module for calculating the optimized feature value And the health feature value Of the healthy image of Idesia polycarpa, and screening to obtain the set of pest and disease areas based on the health feature value ; A pre-module for determining the initial center And calculate the non-local value used to reduce the local noise pixel by the non-local pixel , based on non-local values Calculate the first class weights and the second type of weight ; Fuzzy module, used to calculate the weight of the first category and the second type of weight Compute reduced membership values ​​for defuzzifying a set of pest and disease areas , and according to the reduced membership value Compute new centers for classifying sets of pest and disease areas , get the pest and disease collection; The classification module is used to use the standard image containing the levels of plant diseases and pests of Castanopsis chinensis as the comparison standard to divide the disease and pest set into different levels of mild, moderate and severe diseases and pests.

[0013] By means of the above technical scheme, the present invention provides a method and system for monitoring diseases and insect pests of Castanopsis sylvestris, which have at least the following beneficial effects: 1. The present invention can avoid the problems of offset angle and coordinate position deviation after image shooting by correcting and stitching the pre-processed images. Due to the wide viewing angle of the camera, multiple images need to be stitched. After correcting the stitching, overlap or duplication in the stitching process can be avoided, which not only improves the accuracy of image processing, but also improves the processing efficiency of subsequent steps.

[0014] 2. The present invention extracts texture features of preprocessed images, which can not only reduce the problem of image over-segmentation through improved steps, improve the accuracy of segmentation and the robustness of image processing, and improve the accuracy of texture processing, but also enable rapid classification of pest and disease images and healthy images, thereby improving the accuracy, robustness and work efficiency of image processing.

[0015] 3. The present invention uses an improved fuzzy clustering method to avoid the problem of dividing adjacent pixels into the same category in the prior art, which will lead to poor segmentation performance, and there will also be problems such as over-segmentation and loss of edge details. It can not only avoid the problem of dividing adjacent pixels into the same category, but also improve the segmentation performance, reduce the problem of over-segmentation, and ensure the retention of edge details, thereby improving the accuracy and efficiency of the classification method. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 This is a flow chart of a method for monitoring diseases and insect pests of Castanopsis sylvestris of the present invention; Figure 2 This is the structural block diagram of a monitoring system for the diseases and pests of Idesia polycarpa Maxim. of the present invention. Specific embodiments

[0017] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Thereby, the implementation process of how this application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0018] Due to the poor segmentation performance of the prior art and the technical problems of over-segmentation and loss of edge details, this embodiment proposes a method and system for monitoring the diseases and pests of Idesia polycarpa Maxim., which can improve the segmentation performance and avoid over-segmentation and loss of edge details. As Figure 1 shown, the method includes the following steps: S1. Collect the remote sensing images of Idesia polycarpa Maxim. and preprocess the remote sensing images to obtain multiple noise-free preprocessed images; in the actual process of taking images, due to the influence of air flow and the occlusion of high-altitude fog, the taken images are prone to problems such as image offset or image occlusion. To solve this problem, it is necessary to preprocess the images. The specific implementation steps are as follows: S11. Split the remote sensing image into an R image, a G image, and a B image using the RGB channels, and respectively obtain the RGB pixel values of the pixel points at the same position in the R image, the G image, and the B image , and ; S12. Calculate the grayscale pixel value of the pixel point according to the RGB pixel values , , and . The calculation formula is: ; where represents the th grayscale pixel value; S13. Calculate the denoising value of the pixel point according to the grayscale pixel value . The calculation formula is: ; where represents the standard deviation of the grayscale pixel value ; S14. Repeat step S12 and step S13 to obtain the denoising values ​The calculation is completed, and the processed pixel points are synthesized into a preprocessed image. Through the correction and stitching process of the preprocessed image, problems such as deviation angles and coordinate position deviations after image shooting can be avoided. And due to the wide camera view, multiple images need to be stitched. After correction and stitching, the situation of overlap or duplication during the stitching process can be avoided, which not only improves the accuracy of image processing but also enhances the processing efficiency of subsequent steps.

[0019] S2. Perform correction and stitching processing on each preprocessed image to obtain a range image containing the Idesia polycarpa Maxim. plants; the shaking of the drone makes the angle of the image prone to deviation, resulting in a problem of relative inclination between the preprocessed image and the actual image. At the same time, the range of the preprocessed image is very small. Not only does the image need to be corrected, but multiple preprocessed images also need to be stitched without overlap. To solve this problem, the specific implementation steps are as follows: S21. Establish a division window with a side length of and a step size of on the preprocessed image. Through the division window, two adjacent rectified images and in the preprocessed image are obtained; through the division window, two adjacent rectified images and inside the window are obtained.

[0020] S22. Respectively obtain the pixel points and at the same position in the rectified images and , and calculate the deviation angle . The calculation formula is: ; Among them, represents the number of pixel points and ; S23. Obtain the central pixel point in the preprocessed image, and calculate the corrected pixel point according to the deviation angle . The calculation formula is: ; Among them, and represent the pixel values of the pixel points in the th rectified image; S24. Convert the corrected pixel point into a corrected coordinate through the pixel coordinate conversion method, and calculate the ground coordinate value of the pixel point in the preprocessed image. The calculation formula is: ; Among them, represents the rotation matrix of the external camera parameters, represents the coordinate value of the camera, respectively represent the external camera parameters, represents the focal length of the camera, represents the height of the camera; The pixel coordinate conversion method is a method that converts the pixels in an image into world coordinates based on the coordinates calibrated by the camera. The coordinates calibrated by the camera can use the Zhang Zhengyou calibration method. Therefore, the pixel coordinate conversion method is a commonly used coordinate conversion method and will not be elaborated here.

[0021] S25. Calculate the image fusion value and and , and the calculation formula is: ; Among them, and respectively represent the weight coefficients, and respectively represent the abscissa values of the first and second preprocessed images, and respectively represent the ordinate values of the first and second preprocessed images; The weight coefficients and are obtained through the analytic hierarchy process. The analytic hierarchy process is a commonly used method for obtaining weight coefficients and will not be elaborated here.

[0022] S26. Repeat step S25 to fuse multiple preprocessed images into a range image. Through the correction and stitching process of the preprocessed images, problems such as offset angles and coordinate position deviations after image shooting can be avoided. And due to the camera's viewing angle breadth, multiple images need to be stitched. After correction and stitching, situations of overlap or duplication during the stitching process can be avoided, which not only improves the accuracy of image processing but also improves the processing efficiency of subsequent steps.

[0023] S3. Calculate the invariant feature value of any pixel point in the range image, and establish a feature matrix based on the invariant feature value ; The features of the Idesia polycarpa images in different health states in the range image will be different. If no preliminary screening is performed, it is easy to increase the processing steps of subsequent steps and increase the computational complexity. In order to preliminarily screen out healthy Idesia polycarpa, it is necessary to determine the characteristics of Idesia polycarpa with diseases and pests. The specific implementation steps are as follows:

[0024] S31. Obtain the pixel value Take it as the central pixel point and obtain the radius of the central pixel point within the number of neighborhood pixel points, and calculate the reference value for obtaining the binary result , and the calculation formula is: ; wherein, represents the sign function, represents the gray value of the central pixel point, represents the gray value of the neighborhood pixel point; S32. Change the radius of the central pixel point to , repeat step S31 to obtain multiple reference values , and calculate the invariant eigenvalue according to the multiple reference values , and the calculation formula is: ; wherein, represents the binary result of the reference value circularly shifted to the right by times; S33. Obtain the gray level of the range image through the gray quantization method and , and calculate the gray value and according to the gray level , and the calculation formula is: ; wherein, represents the pixel value of the pixel point in the range image, and respectively represent the offsets of the pixel point at different angles, and represent the length and width of the range image; S34. Calculate the contrast value , the feature entropy and the correlation value of the range image respectively according to the gray value , and the calculation formula is: ; wherein, represents the number of , represents the average value of the gray value , and respectively represent and standard deviations; S35. Take the contrast value , Feature Entropy and related values are combined into a multi-vector ; S36. Calculate a specific filtering value based on the pixel value of any pixel point in the range image , and the calculation formula is: ; Among them, respectively represent the number of directions and the number of scales of the specific filtering value , represents the standard deviation of the gray value of the range image, represents the wavelength of the pixel value , represents the phase shift of the pixel value ; S37. Calculate the filtering feature based on the specific filtering value , and the calculation formula is: ; Among them, represents the convolution calculation between the gray value of the original image and the specific filtering value ; S38. Combine the invariant feature value , the multi-vector and the filtering feature into the feature matrix of the range image. The feature matrix contains multiple features of the range image. By extracting the texture features of the preprocessed image, not only can the problem of over-segmentation of the image be reduced through the improved steps, the accuracy of segmentation and the robustness of image processing be improved, the accuracy of texture processing be enhanced, so that the pest and disease images and healthy images can be quickly classified, but also the accuracy, robustness and working efficiency of image processing be improved.

[0025] S4. Calculate the optimized feature value and the healthy feature value of the healthy Idesia polycarpa image based on the feature matrix. Based on the healthy feature value , the pest and disease area set is screened; after obtaining the feature matrix, healthy Idesia polycarpa and pest and disease Idesia polycarpa also need to be screened on the basis of the feature matrix. The specific implementation steps are as follows: S41. Perform importance screening on the feature matrix of the range image through the random forest model and obtain multiple optimized feature values , and obtain the healthy feature value from the healthy Idesia polycarpa image through steps S3 and the random forest model; the random forest model is a commonly used model for screening features and will not be elaborated here. The healthy feature value It is the image feature of healthy Idesia polycarpa. Since it is the same as the method and step S3, it will not be repeated here.

[0026] S42. Calculate the division value according to the optimized eigenvalue The calculation formula is: ; ; where represents the gradient value of the optimized eigenvalue ; S43. Segment the range image according to the division value to obtain multiple regions to be detected . According to the division value of the region to be detected , obtain the segmentation region set , and the expression is: ; where represents the th segmentation region set. Since the area of the range image is very large and the preprocessed image cannot analyze the whole Idesia polycarpa, on the basis of the range image, Idesia polycarpa is segmented, and then the health of each Idesia polycarpa is studied.

[0027] S44. Calculate the fluctuation value according to the health eigenvalue . The calculation formula is: ; where represents the health eigenvalue of the healthy image of the th Idesia polycarpa; S45. Obtain the difference threshold through the detrending analysis method, and screen out the pest and disease areas in the segmentation region set based on the difference threshold ; If , it means that the optimized eigenvalue is abnormal and is marked as a pest and disease area; If , it means that the optimized eigenvalue is normal and is marked as a healthy area; S46. Combine multiple pest and disease areas into a pest and disease area set. This step can initially screen out the Idesia polycarpa with pests and diseases. By extracting the texture features of the preprocessed image, not only can the problem of image over-segmentation be reduced through the improved steps, the accuracy of segmentation and the robustness of image processing can be improved, the accuracy of texture processing can be improved, the pest and disease images and healthy images can be quickly classified, and the accuracy, robustness and working efficiency of image processing can be improved.

[0028] ​

[0028] S5. Determine the initial center in the pest and disease area set by random method And calculate the non-local value for reducing local noise pixels through non-local pixels , based on the non-local value Calculate the first type of weight and the second type of weight ; The pest and disease area sets are all Idesia polycarpa with pests and diseases. In order to further classify the Idesia polycarpa plants with pests and diseases, it is also necessary to further divide the Idesia polycarpa with pests and diseases in detail according to characteristics. The specific implementation steps are as follows: S51. Randomly select a pixel point in the pest and disease area set as the initial center , define the clustering radius of the initial center as , and randomly and non-repeatedly select two pixel windows and in the range of , and calculate the window difference . The calculation formula is: ; Among them, represents the smoothing parameter that controls the exponential function decay, represents the logarithm of the pixel windows and , and ; S52. Calculate the non-local value according to the window difference . The calculation formula is: ; Among them, represents the central pixel value of the pixel window . In order to solve the problem that the local pixel block contains too many noise pixels, the auxiliary image can be reconstructed by calculating the pixels outside the local pixel block, and more image information can be utilized to improve the accuracy and stability of image segmentation, that is, calculate the non-local value ; S53. Obtain the gray value of the pixel points in the pixel window , and calculate the local variation value according to the gray value . The calculation formula is: ; Among them, represents the average value of the gray values of the pixel points in the pixel window , represents the gray value and the average value of the number of groups; S54. Calculate the association weight according to the local variation value with the calculation formula: ; wherein, represents the th association weight; S55. Calculate the first regularization factor and the second regularization factor with the calculation formula: ; wherein, and respectively represent the random constant and the non-local value of the pixels in the pest and disease area set of the random constant, represents the non-local value of the mean value; S56. Calculate the first type of weight and the second type of weight with the calculation formula: ; ; wherein, represents a constant greater than zero, represents an extremely small number greater than zero. Through the improved fuzzy clustering method, it can avoid the problem in the prior art of dividing adjacent pixels into the same class, which will lead to poor segmentation performance, and there will also be problems such as over-segmentation and loss of edge details. It can not only avoid the problem of dividing adjacent pixels into the same class, but also improve the segmentation performance, reduce the problem of over-segmentation, and ensure the retention of edge details, improving the accuracy and classification efficiency of the classification method.

[0029] S6. Calculate the reduced membership value for defuzzifying the pest and disease area set according to the first type of weight and the second type of weight , and calculate the new center for classifying the pest and disease area set according to the reduced membership value , to obtain the pest and disease set; on the basis of step S5, it is also necessary to classify the pest and disease levels of the pest and disease Chinese parasol trees. The specific implementation steps are as follows: S61. Calculate the first fuzzy factor according to the first type of weight and the second type of weight ​​​​ and the second fuzzy factor , the calculation formula is: ; wherein, represents the average spatial Euclidean distance between pixels, and respectively represent the initial lower limit and the initial upper limit, and represent the fuzzy exponent, and ， ， represents the spatial distance relative to the initial center ; S62. Calculate the updated upper and lower limit update membership degrees and the second fuzzy factor , the calculation formula is: and , the calculation formula is: ; wherein, represents the spatial distance between the update point and the initial center , and represent the fuzzy factors of the pixel points with the gray value of and the initial center , and represent the fuzzy factors of the update point and the initial center , represents the calculation quantity; S63. Construct the updated membership degree range and , and randomly determine the initial updated membership value in the updated membership degree range ; ; S64. Calculate the update factor according to the initial updated membership value , the calculation formula is: ; wherein, and represent random numbers, and , , represents the cardinality of the pest and disease area set; S65. According to the update factor Obtain the reduced membership value , and the expression is: ; Among them, represents the reduction coefficient. In order to calculate the clear clustering center of the model, it is necessary to defuzzify the classification of the pest and disease area set, and use the defuzzification method to calculate the reduced membership value ; S66. Calculate the updated new center according to the reduced membership value , and the calculation formula is: ; Among them, represents the number of pixel point gray values , represents the new center of the th generation; S67. Calculate the iteration threshold by the error square method, and judge whether the iteration stops according to the iteration threshold ; If , the iteration ends, and the pest and disease area set is classified according to the new center to obtain the pest and disease set; If , then let and return to step S61. By using the improved fuzzy clustering method, the problem of dividing adjacent pixels into the same class in the prior art can be avoided, which will lead to poor segmentation performance, and there will also be problems such as over-segmentation and loss of edge details. It can not only avoid the problem of dividing adjacent pixels into the same class, but also improve the segmentation performance, reduce the problem of over-segmentation, and ensure the retention of edge details, thus improving the accuracy and classification efficiency of the classification method.

[0030] S7. Use the standard images containing the levels of Idesia polycarpa pests and diseases as the comparison standard, divide the pest and disease set into different levels of mild, moderate, and severe pests and diseases. Through the classification in step S6, according to the characteristics of mild, moderate, and severe pests and diseases in historical images, the pest and disease set can be classified. The feature method is a method for extracting image features. For example, feature extraction based on gray values. When the Idesia polycarpa tree is still in the initial stage of damage that is not easily detectable by the naked eye, the color it presents in the infrared image will be different from that of a normal Idesia polycarpa tree. Because the Idesia polycarpa tree has a relatively high spectral reflectance in the near-infrared wavelength range, after being damaged by pests and diseases, the water content of the leaves of Idesia polycarpa decreases, the chlorophyll decreases, and the spectral reflectance of the leaves in the infrared wavelength range will decrease significantly. As a result, the tone of the damaged Idesia polycarpa tree in the infrared image is darker than that of normal crops. As the damage degree of Idesia polycarpa pests and diseases gradually deepens, after the chlorophyll in its leaves disappears completely, the image tone will become darker and even show a cyan tone. Therefore, the damaged Idesia polycarpa trees can be screened according to the tone difference generated by image comparison, and this method can be used to extract features for further differentiation. The feature method is a commonly used feature extraction method and will not be elaborated here.

[0031] Due to the poor segmentation performance of the existing technology and the technical problems of over-segmentation and loss of edge details, this embodiment also proposes a monitoring system for Idesia polycarpa pests and diseases, which can improve the segmentation performance and avoid over-segmentation and loss of edge details, as Figure 2 shown, the monitoring system includes a preprocessing module, a correction and stitching module, a texture module, a preliminary screening module, a pre-module, a fuzzy module, and a classification module.

[0032] The preprocessing module is used to collect remote sensing images of Idesia polycarpa and preprocess the remote sensing images to obtain multiple noiseless preprocessed images; the correction and stitching module is used to perform correction and stitching processing on each preprocessed image to obtain a range image containing Idesia polycarpa plants; the texture module is used to calculate the invariant feature value of any pixel point in the range image and establish a feature matrix based on the invariant feature value ; the preliminary screening module is used to calculate the optimized feature value and the healthy feature value of the healthy image of Idesia polycarpa based on the feature matrix, and screen to obtain a pest and disease area set based on the healthy feature value ; the pre-module is used to determine the initial center in the pest and disease area set by the random method and calculate the non-local value for reducing local noise pixels through non-local pixels, and calculate the first type of weight and the second type of weight based on the non-local value ; Fuzzy module, used to calculate the weight of the first category and the second type of weight Compute reduced membership values ​​for defuzzifying a set of pest and disease areas , and according to the reduced membership value Compute new centers for classifying sets of pest and disease areas ; A classification module is used to use a standard image containing the levels of plant diseases and pests of Castanopsis chinensis as a comparison standard to divide the disease and pest set into different levels of mild, moderate and severe diseases and pests.

[0033] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, so the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0034] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the above embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0035] The above implementation methods have been described in detail. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for monitoring diseases and insect pests of Castanopsis sylvestris, characterized in that: The method comprises the following steps: S1, collecting remote sensing images of Castanopsis sylvestris, and preprocessing the remote sensing images to obtain multiple noise-free preprocessed images; S2, performing correction and splicing processing on each pre-processed image to obtain a range image containing the Castanopsis affine plant; S3, calculate any pixel point in the range image The invariant eigenvalues ​​of , and based on the invariant eigenvalues Establish feature matrix; S4. Calculate and optimize eigenvalues ​​based on the feature matrix Health feature values ​​of the healthy image of the tung tree , based on health characteristic values Screening to obtain a set of pest and disease areas; S5. Determine the initial center in the pest and disease area by random method And calculate the non-local value used to reduce the local noise pixel by the non-local pixel , based on non-local values Calculate the first class weights and the second type of weight ; S6. According to the first type of weight and the second type of weight Compute reduced membership values ​​for defuzzifying a set of pest and disease areas , and according to the reduced membership value Compute new centers for classifying sets of pest and disease areas , get the pest and disease collection; S7. Use standard images containing the levels of Castanopsis affine diseases and pests as a comparison standard to divide the disease and pest set into different levels of mild, moderate and severe diseases and pests.

2. The monitoring method according to claim 1, characterized in that: In step S1, the specific implementation steps are as follows: S11, split the remote sensing image into R image, G image and B image using RGB channels, and obtain the pixel points at the same position in the R image, G image and B image respectively RGB pixel value , and ; S12, according to RGB pixel value , and Counting pixels Grayscale pixel value , the calculation formula is: ； in, Indicates Grayscale pixel values; S13, according to the grayscale pixel value Counting pixels The denoising value , the calculation formula is: ； in, Represents grayscale pixel value The standard deviation of S14, repeat step S12 and step S13 until the denoising values ​​of all pixels are completed The processed pixels are synthesized into a preprocessed image.

3. The monitoring method according to claim 1, characterized in that: In step S2, the specific implementation steps are as follows: S21, establish a side length on the preprocessed image And the step length is The divided window is used to obtain two adjacent corrected images in the preprocessed image. and ; S22, respectively obtain the deflection-corrected images and Pixels at the same position in and , and calculate the deviation angle , the calculation formula is: ； in, Represents pixel and the number of S23, obtaining the central pixel point in the preprocessed image , according to the deviation angle Calculate corrected pixels , the calculation formula is: ； in, and Indicates The pixel value of a pixel in the corrected image; S24, correcting the pixel point by pixel coordinate conversion method Convert to corrected coordinates , calculate the ground coordinate values ​​of the pixels in the preprocessed image , the calculation formula is: ； in, represents the rotation matrix of the camera extrinsic parameters, Represents the coordinate value of the camera, They represent the external parameters of the camera respectively, represents the focal length of the camera, Indicates the height of the camera; S25, based on the ground coordinate values ​​in the multiple pre-processed images Calculate image fusion value and , the calculation formula is: ； in, and They represent weight coefficients respectively, and Respectively represent the horizontal coordinate values ​​of the first and second pre-processed images, and Respectively represent the ordinate values ​​of the first and second pre-processed images; S26. Repeat step S25 to fuse the multiple pre-processed images into a range image.

4. The monitoring method according to claim 1, characterized in that: In step S3, the specific implementation steps are as follows: S31. Obtain the pixel value of any pixel in the range image As the center pixel point, and get the radius of the center pixel point within Field pixels, calculate the reference value used to obtain the binary result , the calculation formula is: ； in, represents the symbolic function, Represents the gray value of the center pixel, Represents the grayscale value of the neighborhood pixel; S32, change the radius of the central pixel to , repeat step S31 to obtain multiple reference values , based on multiple reference values Compute invariant eigenvalues , the calculation formula is: ； in, Indicates that the reference value Circularly shift the binary result right Second-rate; S33, obtaining the gray level of the range image by gray quantization method and , according to the gray level and Calculate the ash value , the calculation formula is: ； in, Represents the pixel value of the pixel in the range image, and Respectively represent the offset of the pixel at different angles, and Indicates the length and width of the range image; S34, according to the ash value Calculate the contrast value of the range image separately , characteristic entropy and related values , the calculation formula is: ； in, express The number of Gray value The average of and Respectively and The standard deviation of S35, compare the value , characteristic entropy and related values Combine into multiple vectors ; S36, according to the pixel value of any pixel point in the range image Calculates specific filter values , the calculation formula is: ； in, Respectively represent specific filter values The number of directions and scales, Represents the standard deviation of the grayscale value of the range image, Represents pixel value The wavelength of Represents pixel value The phase shift of S37, according to a specific filter value Calculate filter characteristics , the calculation formula is: ； in, Represents the grayscale value of the original image and the specific filter value Convolution calculation between ; S38, the unchanged eigenvalue , multiple vectors and filtering characteristics The combination becomes the feature matrix of the range image.

5. The monitoring method according to claim 1, characterized in that: In step S4, the specific implementation steps are as follows: S41. Use the random forest model to screen the importance of the feature matrix of the range image and obtain multiple optimized eigenvalues , and obtain the health feature value from the healthy image of Castanopsis serrata through step S3 and the random forest model ; S42, according to the optimized characteristic value Calculate partition value , the calculation formula is: ； in, Represents the optimized eigenvalue The gradient value of S43, according to the division value Segment the range image to obtain multiple areas to be detected , according to the area to be detected The partition value Get the segmentation region set , the expression is: ； in, Indicates a set of partitioned regions; S44. According to health characteristic value Calculate the volatility , the calculation formula is: ； in, Indicates The health feature value of the health image of the tree; S45. Obtaining the differential threshold by detrending analysis , based on the difference threshold In the segmentation area set Screen out pest and disease areas; like , it means that the optimized eigenvalue is abnormal and marked as a pest and disease area; like , it means that the optimized eigenvalue is normal and marked as a healthy area; S46. Grouping multiple pest and disease areas into a pest and disease area set.

6. The monitoring method according to claim 1, characterized in that: In step S5, the specific implementation steps are as follows: S51. Randomly select a pixel point in the pest area as the initial center , define the initial center The cluster radius is ,exist Randomly select two pixel windows without duplication within the range and , and calculate the window difference , the calculation formula is: ； in, represents the smoothing parameter that controls the decay of the exponential function, Represents a pixel window and The logarithm of ; S52, according to the window difference Computing non-local values , the calculation formula is: ； in, Represents a pixel window The central pixel value of S53, get pixel window Gray value of inner pixel , according to the gray value Calculate local variation value , the calculation formula is: ； in, Represents a pixel window Gray value of inner pixel The average of Represents grayscale value and average The number of groups; S54, based on local variation value Calculate association weight , the calculation formula is: ； in, Indicates association weights; S55. According to the association weight Calculate the first regularization factor and the second regularization factor , the calculation formula is: ； in, and Represents the random constants and non-local values ​​of the pixels in the pest area set The random constant, Represents a non-local value The mean of S56. According to the first regularization factor and the second regularization factor Calculate the first class weights and the second type of weight , the calculation formula is: ； in, represents a constant greater than zero, Represents a very small number greater than zero.

7. The monitoring method according to claim 1, characterized in that: In step S6, the specific implementation steps are as follows: S61. According to the first type of weight and the second type of weight Calculate the first fuzzy factor and the second fuzzy factor , the calculation formula is: ； in, represents the average spatial Euclidean distance between pixels, and denote the initial lower limit and initial upper limit respectively, and represents the fuzzy index, and ， ， Relative to the initial center spatial distance; S62, according to the first fuzzy factor and the second fuzzy factor Calculate the updated upper and lower limits to update the membership and , the calculation formula is: ； in, Indicates update point Distance from initial center The spatial distance and Indicates the gray value is The pixel point and initial center The fuzzy factor, and Indicates update point and the initial center The fuzzy factor, Indicates the number of calculations; S63. Update the membership degree according to the upper and lower limits and Construct update membership range , in the updated membership range Randomly determine the initial updated membership value ; S64, based on the initial update membership value Calculate the update factor , the calculation formula is: ； in, and represents a random number, and , , represents the cardinality of the pest and disease area set; S65, according to the update factor Get the reduced membership value ; S66, according to the reduced membership value Calculate the updated new center , the calculation formula is: ； in, Represents the gray value of the pixel The number of Indicates The new center of the generation; S67. Calculate the iteration threshold by error square method , according to the iteration threshold Determine whether the iteration stops; like , the iteration ends, and according to the new center Classify the pest and disease area set to obtain the pest and disease set; like , then let And return to step S61.

8. The monitoring method according to claim 7, characterized in that: The reduced membership value The expression is: ； in, Represents the reduction factor.

9. A system for monitoring diseases and insect pests of Castanopsis sylvestris as claimed in any one of claims 1 to 8, characterized in that: The system includes: A preprocessing module is used to collect remote sensing images of Castanopsis sylvestris and preprocess the remote sensing images to obtain multiple noise-free preprocessed images; A correction and splicing module is used to perform correction and splicing processing on each pre-processed image to obtain a range image containing the Castanopsis affine plant; Texture module, used to calculate any pixel in the range image The invariant eigenvalues ​​of , and based on the invariant eigenvalues Establish feature matrix; Preliminary screening module, used to calculate and optimize eigenvalues ​​based on the feature matrix Health feature values ​​of the healthy image of the tung tree , based on health characteristic values Screening to obtain a set of pest and disease areas; Pre-module for determining the initial center of the pest area by random method And calculate the non-local value used to reduce the local noise pixel by the non-local pixel , based on non-local values Calculate the first class weights and the second type of weight ; Fuzzy module, used to calculate the weight of the first category and the second type of weight Compute reduced membership values ​​for defuzzifying a set of pest and disease areas , and according to the reduced membership value Compute new centers for classifying sets of pest and disease areas , get the pest and disease collection; The classification module is used to use the standard image containing the levels of plant diseases and pests of Castanopsis chinensis as the comparison standard to divide the disease and pest set into different levels of mild, moderate and severe diseases and pests.

Images