Intelligent Recognition Method for Forest Tree Diseases and Insect Pests Based on UAV Remote Sensing

Abstract

The present invention relates to the technical field of image processing, and particularly to an intelligent recognition method for forest pests and diseases based on UAV remote sensing, including: analyzing the position distribution of pixel points in each suspected pest and disease connected domain in the LAB color space coordinate system to obtain the green sensitivity of each suspected pest and disease connected domain; dividing all suspected pest and disease connected domains into high green sensitivity connected domains and low green sensitivity connected domains based on the green sensitivity; obtaining the enhancement coefficient of the remote sensing gray-scale image of the plantation to be detected according to the gray-scale difference situation of pixel points between the high green sensitivity connected domain and the low green sensitivity connected domain, as well as the edge feature parameters; enhancing the remote sensing gray-scale image of the plantation to be detected according to the enhancement coefficient to obtain the enhanced remote sensing image of the plantation to be detected; and performing intelligent recognition of forest pests and diseases based on the enhanced remote sensing image of the plantation to be detected. The present invention improves the recognition degree of the forest areas with pests and diseases in the plantation.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to an intelligent recognition method for forest pests and diseases based on UAV remote sensing. Background Art

[0002] Since the trees suffering from pests and diseases in the forest are usually difficult to be discovered in time and effectively, it leads to the aggravation of forest pests and diseases, causing serious losses to the forest and immeasurable damage to the ecological environment. Therefore, the monitoring of forest pests and diseases is extremely important; for the common pine nematode disease of forest pests and diseases, it mostly occurs in the planted forests with a single planting structure. The monitoring method of forest pests and diseases for UAV remote sensing images mainly relies on color filtering for detection, that is, using UAVs to photograph the forest, and then enhancing the obtained images. The difference in colors between normal trees and pest-infected trees on the enhanced remote sensing images obtained by shooting is analyzed for discrimination; when performing image enhancement, the linear gray-scale transformation algorithm is often used for processing. The traditional linear gray-scale transformation algorithm performs global linear enhancement on the image, so it can only increase the overall brightness of the image and cannot perform local enhancement on the areas of pest-infected trees, resulting in insufficient distinguishability between the areas of pest-infected forest trees and normal forest trees, and making the identified areas of pest-infected forest trees in the planted forest inaccurate. Summary of the Invention

[0003] In order to solve the above problems, the present invention provides an intelligent recognition method for forest pests and diseases based on UAV remote sensing. The method includes:

[0004] Obtain the remote sensing gray-scale image of the planted forest to be detected;

[0005] Through the connected component analysis of the remote sensing gray-scale image of the planted forest to be detected, obtain several circular connected components in the remote sensing gray-scale image of the planted forest to be detected; through the analysis of the edge concave-convex characteristics of each circular connected component, obtain the edge feature parameters of each circular connected component; through the edge feature parameters, screen all the circular connected components to obtain all the suspected pest and disease connected components;

[0006] Convert each suspected pest and disease connected component into the LAB color space, analyze the position distribution of the pixel points in each suspected pest and disease connected component in the LAB color space coordinate system, obtain the green sensitivity of each suspected pest and disease connected component; based on the green sensitivity, divide all the suspected pest and disease connected components into high green sensitivity connected components and low green sensitivity connected components; according to the gray-scale difference situation of the pixel points between the high green sensitivity connected components and the low green sensitivity connected components, as well as the edge feature parameters, obtain the enhancement coefficient of the remote sensing gray-scale image of the planted forest to be detected;

[0007] Enhance the remote sensing grayscale image of the plantation to be detected according to the enhancement coefficient to obtain the enhanced remote sensing image of the plantation to be detected; perform intelligent identification of forest tree pests and diseases based on the enhanced remote sensing image of the plantation to be detected.

[0008] Preferably, the method for obtaining several circular connected domains in the remote sensing grayscale image of the plantation to be detected by performing connected domain analysis on the remote sensing grayscale image of the plantation to be detected includes the following specific steps:

[0009] Obtain the gradient image of the remote sensing grayscale image of the plantation to be detected, perform the watershed segmentation algorithm on the gradient image to obtain multiple regions of the remote sensing grayscale image of the plantation to be detected, and take each region as a circular connected domain, thereby obtaining several circular connected domains in the remote sensing grayscale image of the plantation to be detected.

[0010] Preferably, the method for obtaining the edge feature parameters of each circular connected domain by analyzing the edge concavity and convexity features of each circular connected domain includes the following specific steps:

[0011] For any circular connected domain in the remote sensing grayscale image of the plantation to be detected, perform edge detection on the circular connected domain to obtain all edge pixel points in the circular connected domain;

[0012] Obtain the center position of the circular connected domain;

[0013] According to the position distribution between the center position of the circular connected domain and the edge pixel points, obtain all convex point angles of the circular connected domain;

[0014] Take the normalized value of the mean of the angle values of all convex point angles in the circular connected domain as the edge feature parameter of the circular connected domain.

[0015] Preferably, the method for obtaining the center position of the circular connected domain includes the following specific steps:

[0016] Arbitrarily combine all edge pixel points in the circular connected domain to obtain several edge point combinations of the circular connected domain; for any edge point combination, record the Euclidean distance between the two edge pixel points in the edge point combination as the first distance of the edge point combination; among all edge point combinations of the circular connected domain, record the edge point combination with the largest first distance as the target edge point combination of the circular connected domain; take the center point of the straight line connection between the two edge pixel points in the target edge point combination as the center position of the circular connected domain.

[0017] Preferably, the method for obtaining all convex point angles of the circular connected domain according to the position distribution between the center position of the circular connected domain and the edge pixel points includes the following specific steps:

[0018] For any edge pixel point in the circular connected domain, the Euclidean distance between the edge pixel point and the center position of the circular connected domain is denoted as the target distance of the edge pixel point; both the left adjacent edge pixel point and the right adjacent edge pixel point of the edge pixel point are used as the comparison pixel points of the edge pixel point.

[0019] Among the edge pixel point and its comparison edge pixel points, if the target distance of the edge pixel point is the maximum value, the edge pixel point is denoted as the convex point of the circular connected domain; if the target distance of the edge pixel point is the minimum value, the edge pixel point is denoted as the concave point of the circular connected domain.

[0020] For any convex point of the circular connected domain, the included angle formed by the convex point and its two adjacent concave points with the convex point as the vertex is denoted as the convex point included angle of the circular connected domain.

[0021] Preferably, the specific method for screening all circular connected domains through edge feature parameters to obtain all suspected pest and disease connected domains includes:

[0022] Preset a feature threshold parameter , for any circular connected domain in the remote sensing grayscale image of the to-be-detected plantation forest, if the edge feature parameter of the circular connected domain is less than the feature threshold parameter , the circular connected domain is denoted as a suspected pest and disease connected domain.

[0023] Preferably, the specific method for converting each suspected pest and disease connected domain into the LAB color space, analyzing the position distribution of pixel points in the LAB color space coordinate system of each suspected pest and disease connected domain, and obtaining the green sensitivity of each suspected pest and disease connected domain includes:

[0024] For any pixel point in any suspected pest and disease connected domain, the absolute value of the difference between the chromaticity value of positive green on the a-axis of the LAB color space coordinate system and the chromaticity value of the pixel point on the a-axis of the LAB color space coordinate system is denoted as the green difference value of the pixel point; the normalized value of the sum of the green difference values of all pixel points in the suspected pest and disease connected domain is used as the green sensitivity of the suspected pest and disease connected domain.

[0025] Preferably, the specific method for dividing all suspected pest and disease connected domains into high green sensitivity connected domains and low green sensitivity connected domains based on green sensitivity includes:

[0026] Preset a sensitivity threshold parameter , for any suspected pest and disease connected region, if the green sensitivity of the suspected pest and disease connected region is less than the sensitive threshold parameter , mark the suspected pest and disease connected region as a low green sensitivity connected region; if the green sensitivity of the suspected pest and disease connected region is greater than or equal to the sensitive threshold parameter , mark the suspected pest and disease connected region as a high green sensitivity connected region.

[0027] Preferably, the method for obtaining the enhancement coefficient of the remote sensing gray image of the plantation to be detected according to the gray difference situation of the pixel points between the high green sensitivity connected region and the low green sensitivity connected region, and the edge feature parameter, includes the following specific method:

[0028] Denote the average value of the gray values of all pixel points in all high green sensitivity connected regions in the remote sensing gray image of the plantation to be detected as the first average value; denote the average value of the edge feature parameters of all high green sensitivity connected regions in the remote sensing gray image of the plantation to be detected as the second average value; denote the product of the first average value and the second average value as the first product; denote the average value of the gray values of all pixel points in all low green sensitivity connected regions in the remote sensing gray image of the plantation to be detected as the third average value; denote the average value of the edge feature parameters of all low green sensitivity connected regions in the remote sensing gray image of the plantation to be detected as the fourth average value; denote the product of the third average value and the fourth average value as the second product; use the ratio of the first product to the second product as the enhancement coefficient of the remote sensing gray image of the plantation to be detected.

[0029] Preferably, the method for intelligent identification of forest pests and diseases based on the enhanced remote sensing image of the plantation to be detected includes the following specific method:

[0030] Input the enhanced remote sensing image of the plantation to be detected into the trained neural network to obtain the forest area with pests and diseases in the enhanced remote sensing image of the plantation to be detected.

[0031] The beneficial effects of the technical solution of the present invention are as follows: By analyzing the position distribution of pixel points in each suspected pest and disease connected region in the LAB color space coordinate system, the green sensitivity of each suspected pest and disease connected region is obtained; based on the green sensitivity, all suspected pest and disease connected regions are divided into high green sensitivity connected regions and low green sensitivity connected regions; according to the gray scale difference of pixel points between the high green sensitivity connected region and the low green sensitivity connected region, as well as the edge feature parameters, the enhancement coefficient of the remote sensing gray scale image of the plantation to be detected is obtained; the remote sensing gray scale image of the plantation to be detected is enhanced according to the enhancement coefficient to obtain the enhanced remote sensing image of the plantation to be detected; intelligent identification of forest pests and diseases is carried out based on the enhanced remote sensing image of the plantation to be detected; it avoids the traditional linear gray scale transformation algorithm from only enhancing the image as a whole; the enhancement coefficient is used to increase the distinction between the pest-infected forest area and the normal forest area, thereby improving the recognition degree of the pest-infected forest area in the plantation. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0033] Figure 1 It is a flowchart of the steps of the intelligent identification method for forest pests and diseases based on UAV remote sensing of the present invention;

[0034] Figure 2 It is a flowchart of the characteristic relationship of the intelligent identification method for forest pests and diseases based on UAV remote sensing of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0035] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following will, in combination with the drawings and preferred embodiments, describe in detail the specific implementation manner, structure, characteristics and effects of the intelligent identification method for forest pests and diseases based on UAV remote sensing proposed by the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0037] The following will specifically describe the specific solution of the intelligent identification method for forest pests and diseases based on UAV remote sensing provided by the present invention in conjunction with the drawings.

[0038] Please refer to Figure 1 , which shows the flowchart of the steps of the intelligent recognition method for forest pests and diseases based on UAV remote sensing provided by an embodiment of the present invention. The method includes the following steps:

[0039] Step S001: Obtain the remote sensing grayscale image of the plantation to be detected.

[0040] It should be noted that the pests and diseases detected in this embodiment are the common pine nematode diseases in forest pests and diseases, which mostly occur in plantations with a single planting structure. The planting pattern of plantations is regular, and in the initial stage of pests and diseases, their color changes are not obvious. Therefore, it is necessary to enhance the trees in the remote sensing image of the plantation to be detected that may have pests and diseases, so that it is easier to judge the trees with pests and diseases from the enhanced remote sensing image.

[0041] Specifically, first, it is necessary to collect the remote sensing grayscale image of the plantation to be detected. The specific process is as follows:

[0042] Use a UAV equipped with a high-pixel camera to obtain the remote sensing image of the plantation to be detected, and perform median filtering denoising and grayscale conversion operations on the remote sensing image of the plantation to be detected to obtain the remote sensing grayscale image of the plantation to be detected.

[0043] Among them, median filtering and grayscale conversion operations are prior arts, and will not be elaborated here in this embodiment.

[0044] So far, the remote sensing grayscale image of the plantation to be detected is obtained through the above method.

[0045] Step S002: By performing connected component analysis on the remote sensing grayscale image of the plantation to be detected, obtain several circular connected components in the remote sensing grayscale image of the plantation to be detected; by analyzing the edge features of each circular connected component, obtain the edge feature parameters of each circular connected component; screen all circular connected components through the edge feature parameters to obtain all suspected pest and disease connected components.

[0046] It should be noted that due to the planting rules of planted forests, in order to ensure the uniform growth of trees, there are strict planting standard specifications for the spacing between trees. The distance between each tree is relatively large, and there is less debris in the forest. The trees are planted at the same time, and their growth conditions are basically similar. Therefore, after edge detection, circular connected regions that are evenly distributed are obtained. Also, because the edges of trees without pests and diseases have good ductility and grow evenly, the edges of their connected regions are relatively smooth; while the edges of trees with pests and diseases have poor ductility and uneven growth, so the edges of their connected regions are uneven, that is, there are many protrusions and depressions, and the smoothness is relatively low. Therefore, by analyzing the edge features of circular connected regions in the remote sensing grayscale image, edge feature parameters can be obtained; based on the edge feature parameters, the circular connected regions are screened to obtain suspected pest and disease connected regions.

[0047] Preferably, in some implementation manners of the embodiments of the present invention, the specific method for obtaining a plurality of circular connected regions in the remote sensing grayscale image of the planted forest to be detected by performing connected region analysis on the remote sensing grayscale image of the planted forest to be detected is as follows:

[0048] Obtain the gradient image of the remote sensing grayscale image of the planted forest to be detected, perform the watershed segmentation algorithm on the gradient image to obtain multiple regions of the remote sensing grayscale image of the planted forest to be detected, and take each region as a circular connected region, so as to obtain a plurality of circular connected regions in the remote sensing grayscale image of the planted forest to be detected.

[0049] Among them, the watershed segmentation algorithm is a prior art, and it will not be elaborated here in this embodiment.

[0050] Preferably, in some implementation manners of the embodiments of the present invention, since the edges of trees with pests and diseases have poor ductility and uneven growth, the edges of their circular connected regions are uneven, that is, there are many protrusions and depressions, and the smoothness is relatively low. Therefore, the specific method for obtaining the edge feature parameters of each circular connected region by performing edge concave-convex feature analysis on each circular connected region is as follows:

[0051] For any circular connected region in the remote sensing grayscale image of the planted forest to be detected, use the Canny edge detection algorithm to perform edge detection on the circular connected region to obtain all edge pixel points in the circular connected region; combine all edge pixel points in the circular connected region in any pairwise combination to obtain a plurality of edge point combinations of the circular connected region; for any edge point combination, record the Euclidean distance between the two edge pixel points in the edge point combination as the first distance of the edge point combination; among all edge point combinations of the circular connected region, record the edge point combination with the largest first distance as the target edge point combination of the circular connected region; take the center point of the straight line connection between the two edge pixel points in the target edge point combination as the center position of the circular connected region.

[0052] Among them, the Canny edge detection algorithm is a prior art and will not be elaborated here in this embodiment.

[0053] For any edge pixel point in the circular-like connected domain, the Euclidean distance between the edge pixel point and the center position of the circular-like connected domain is denoted as the target distance of the edge pixel point; both the left adjacent edge pixel point and the right adjacent edge pixel point of the edge pixel point are used as the comparison pixel points of the edge pixel point.

[0054] Among the edge pixel point and its comparison edge pixel points, if the target distance of the edge pixel point is the maximum value, the edge pixel point is denoted as the convex point of the circular-like connected domain; if the target distance of the edge pixel point is the minimum value, the edge pixel point is denoted as the concave point of the circular-like connected domain.

[0055] For any convex point of the circular-like connected domain, the included angle formed by the convex point and its two adjacent concave points with the convex point as the vertex is denoted as the convex point included angle of the circular-like connected domain.

[0056] The normalized value of the mean of the angle values of all convex point included angles in the circular-like connected domain is used as the edge feature parameter of the circular-like connected domain.

[0057] The specific formula is:

[0058]

[0059] In the formula, represents the edge feature parameter of the th circular-like connected domain; represents the number of all convex point included angles in the th circular-like connected domain; represents the angle value of the th convex point included angle in the th circular-like connected domain; represents a preset hyperparameter, and in this embodiment, is preset, which is used to prevent the denominator from being 0; represents the normalization function.

[0060] It should be noted that since the edge of the circular-like connected domain of normal trees has a relatively high smoothness, it means that the mean of the angle values of the convex point included angles will be larger, that is, the edge feature parameter is larger; while the edge of the circular-like connected domain of trees with pests and diseases has a relatively low smoothness, then the mean of the angle values of the convex point included angles will be smaller, that is, the edge feature parameter is smaller.

[0061] Preferably, in some implementation manners of the embodiments of the present invention, the specific method for screening all circular connected regions through edge feature parameters to obtain all suspected pest and disease connected regions is as follows:

[0062] Preset a feature threshold parameter , where this embodiment takes as an example for description, and this embodiment does not make specific limitations, where is determined according to the specific implementation situation;

[0063] For any circular connected region in the remote sensing grayscale image of the to-be-detected plantation, if the edge feature parameter of the circular connected region is less than the feature threshold parameter , mark the circular connected region as a suspected pest and disease connected region.

[0064] So far, all suspected pest and disease connected regions in the remote sensing grayscale image of the to-be-detected plantation are obtained through the above method.

[0065] Step S003: Convert each suspected pest and disease connected region into the LAB color space, analyze the position distribution of the pixel points in each suspected pest and disease connected region in the LAB color space coordinate system, and obtain the green sensitivity of each suspected pest and disease connected region; divide all suspected pest and disease connected regions into high green sensitivity connected regions and low green sensitivity connected regions based on the green sensitivity; obtain the enhancement coefficient of the remote sensing grayscale image of the to-be-detected plantation according to the gray difference situation between the pixel points in the high green sensitivity connected region and the low green sensitivity connected region, as well as the edge feature parameter.

[0066] It should be noted that in the LAB color space coordinate system, the value range of a is from -128 to +127, and the color change it represents is from green to magenta; since the normally growing trees appear green, and green is on the negative half-axis of the a (chromaticity) axis in the LAB color space coordinate system; while the trees with pests and diseases appear red, brown, or purple, that is, on the positive half-axis of the a (chromaticity) axis in the color space; therefore, the suspected pest and disease connected regions can be converted into the LAB color space, analyze the position distribution of the pixel points in the suspected pest and disease connected regions in the LAB color space, and obtain the green sensitivity of the suspected pest and disease connected regions; divide all suspected pest and disease connected regions into high green sensitivity connected regions and low green sensitivity connected regions based on the green sensitivity.

[0067] Preferably, in some implementation manners of the embodiments of the present invention, the specific method for converting each suspected pest and disease connected region into the LAB color space, analyzing the position distribution of the pixel points in each suspected pest and disease connected region in the LAB color space coordinate system, and obtaining the green sensitivity of each suspected pest and disease connected region is as follows:

[0068] For any pixel point in any suspected pest and disease connected region, the absolute value of the difference between the chromaticity value of positive green on the a-axis of the LAB color space coordinate system and the chromaticity value of the pixel point on the a-axis of the LAB color space coordinate system is denoted as the green difference value of the pixel point; the normalized value of the sum of the green difference values of all pixel points in the suspected pest and disease connected region is used as the green sensitivity of the suspected pest and disease connected region.

[0069] The specific formula is:

[0070]

[0071] In the formula, represents the green sensitivity of the th suspected pest and disease connected region; represents the number of all pixel points in the th suspected pest and disease connected region; represents the chromaticity value of positive green on the a-axis of the LAB color space coordinate system; represents the th pixel point in the th suspected pest and disease connected region on the a-axis of the LAB color space coordinate system; represents taking the absolute value; represents the exponential function with the natural constant as the base. In the embodiment, the model is used to present the inverse proportional relationship and normalization process, is the input of the model. The implementer can select the inverse proportional function and the normalization function according to the actual situation.

[0072] It should be noted that the larger the value, the farther the chromaticity distance between the th pixel point and positive green on the a-axis of the LAB color space coordinate system, that is, the lower the green sensitivity of the th suspected pest and disease connected region, and the higher the possibility that the th suspected pest and disease connected region is a pest and disease tree.

[0073] Preferably, in some implementation manners of the embodiment of the present invention, the specific method for dividing all suspected pest and disease connected regions into high green sensitivity connected regions and low green sensitivity connected regions based on green sensitivity is:

[0074] Preset a sensitive threshold parameter , where in this embodiment, is taken as an example for description. This embodiment does not make specific limitations, and is determined according to the specific implementation situation;

[0075] For any suspected pest and disease connected region, if the green sensitivity of the suspected pest and disease connected region is less than the sensitive threshold parameter , the suspected pest and disease connected region is recorded as a low green sensitivity connected region; if the green sensitivity of the suspected pest and disease connected region is greater than or equal to the sensitive threshold parameter , the suspected pest and disease connected region is recorded as a high green sensitivity connected region.

[0076] So far, all high green sensitivity connected regions and low green sensitivity connected regions in the remote sensing grayscale image of the artificial forest to be detected are obtained.

[0077] It should be noted that the enhancement effect of the linear grayscale transformation algorithm can be quantified by the degree of distinction between the high green sensitivity connected region and the low green sensitivity connected region, that is, the greater the degree of distinction between the two, the better the enhancement effect; in order to prevent over-enhancement of the image, it is necessary to weaken through the edge feature parameters of the connected region, so that the enhanced low green sensitivity connected region preserves better detail.

[0078] Preferably, in some implementation manners of the embodiments of the present invention, the specific method for obtaining the enhancement coefficient of the remote sensing grayscale image of the artificial forest to be detected according to the grayscale difference situation of the pixel points between the high green sensitivity connected region and the low green sensitivity connected region, and the edge feature parameters is as follows:

[0079] The mean value of the grayscale values of all pixel points in all high green sensitivity connected regions in the remote sensing grayscale image of the artificial forest to be detected is denoted as the first mean value; the mean value of the edge feature parameters of all high green sensitivity connected regions in the remote sensing grayscale image of the artificial forest to be detected is denoted as the second mean value; the product of the first mean value and the second mean value is denoted as the first product; the mean value of the grayscale values of all pixel points in all low green sensitivity connected regions in the remote sensing grayscale image of the artificial forest to be detected is denoted as the third mean value; the mean value of the edge feature parameters of all low green sensitivity connected regions in the remote sensing grayscale image of the artificial forest to be detected is denoted as the fourth mean value; the product of the third mean value and the fourth mean value is denoted as the second product; the ratio of the first product to the second product is used as the enhancement coefficient of the remote sensing grayscale image of the artificial forest to be detected;

[0080] The specific formula is:

[0081]

[0082] In the formula, represents the enhancement coefficient of the remote sensing grayscale image of the artificial forest to be detected; represents the mean value of the grayscale values of all pixel points in all high green sensitivity connected regions in the remote sensing grayscale image of the artificial forest to be detected; represents the mean of the edge feature parameters of all highly green-sensitive connected regions in the remote sensing grayscale image of the plantation to be detected; represents the mean of the grayscale values of all pixel points within all low green-sensitive connected regions in the remote sensing grayscale image of the plantation to be detected; represents the mean of the edge feature parameters of all low green-sensitive connected regions in the remote sensing grayscale image of the plantation to be detected.

[0083] It should be noted that The larger the ratio, the greater the difference in grayscale values between the low green-sensitive connected regions and the high green-sensitive connected regions in the image, that is, the greater the difference between the areas of pest-infected trees and normal trees, and the greater the degree of differentiation; furthermore, the ratio between the edge feature parameters of the low green-sensitive connected regions and the high green-sensitive connected regions is used as the confidence level, so that better details of the pest-infected tree areas are preserved.

[0084] Thus far, the enhancement coefficient of the remote sensing grayscale image of the plantation to be detected is obtained through the above method.

[0085] Step S004: Enhance the remote sensing grayscale image of the plantation to be detected according to the enhancement coefficient to obtain the enhanced remote sensing image of the plantation to be detected; perform intelligent identification of forest pests and diseases based on the enhanced remote sensing image of the plantation to be detected.

[0086] Preferably, input the enhancement coefficient of the remote sensing grayscale image of the plantation to be detected into the linear grayscale transformation algorithm to enhance the remote sensing grayscale image of the plantation to be detected, and obtain the enhanced remote sensing image of the plantation to be detected.

[0087] Among them, the linear grayscale transformation algorithm is a prior art, and this embodiment will not elaborate on it here.

[0088] Input the enhanced remote sensing image of the plantation to be detected into the trained neural network to obtain the areas of forest trees with pests and diseases in the enhanced remote sensing image of the plantation to be detected; among them, the neural network used in this embodiment is YOLOv3, and the method for obtaining the dataset for training this neural network is as follows:

[0089] Collect a large number of remote sensing grayscale images of the plantation to be detected, and manually use bounding boxes to mark the areas of forest trees with pests and diseases in each remote sensing grayscale image of the plantation to be detected, and record this marking result as the label of each remote sensing grayscale image of the plantation to be detected; collect a large number of remote sensing grayscale images of the plantation to be detected and their corresponding labels to form a dataset; use this dataset to train this neural network, and the loss function used during the training process is the mean square error loss function; among them, the specific training process is well-known content of the neural network, and this embodiment will not elaborate on the specific training process.

[0090] Please refer toFigure 2 , which shows a characteristic relationship flowchart of an intelligent identification method for forest pests and diseases based on UAV remote sensing.

[0091] So far, this embodiment is completed.

[0092] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent identification method for forest pests and diseases based on UAV remote sensing, characterized in that: The method comprises the following steps: Obtaining a remote sensing grayscale image of the artificial forest to be detected; By performing connected domain analysis on the remote sensing grayscale image of the artificial forest to be detected, several circle-like connected domains in the remote sensing grayscale image of the artificial forest to be detected are obtained; by performing edge concave-convex feature analysis on each circle-like connected domain, edge feature parameters of each circle-like connected domain are obtained; all circle-like connected domains are screened by edge feature parameters to obtain all suspected pest and disease connected domains; Each suspected pest and disease connected domain is converted into the LAB color space, and the position distribution of the pixel points in each suspected pest and disease connected domain in the LAB color space coordinate system is analyzed to obtain the green sensitivity of each suspected pest and disease connected domain, including the following specific methods: for any pixel point in any suspected pest and disease connected domain, the absolute value of the difference between the chromaticity value of positive green on the a-axis of the LAB color space coordinate system and the chromaticity value of the pixel point on the a-axis of the LAB color space coordinate system is recorded as the green difference value of the pixel point; the normalized value of the cumulative sum of the green difference values ​​of all the pixel points in the suspected pest and disease connected domain is used as the green sensitivity of the suspected pest and disease connected domain; Based on green sensitivity, all suspected pest and disease connected domains are divided into high green sensitivity connected domains and low green sensitivity connected domains; according to the grayscale difference of pixels between the high green sensitivity connected domain and the low green sensitivity connected domain, as well as edge feature parameters, the enhancement coefficient of the remote sensing grayscale image of the artificial forest to be detected is obtained, including the specific method of: recording the mean of the grayscale values ​​of all pixels in all high green sensitivity connected domains in the remote sensing grayscale image of the artificial forest to be detected as the first mean; recording the grayscale value of all pixels in the high green sensitivity connected domain in the remote sensing grayscale image of the artificial forest to be detected as the first mean; recording the grayscale value of all pixels in the high green sensitivity connected domain in the remote sensing grayscale image of the artificial forest to be detected as the first mean. The mean of the edge feature parameters of the domain is recorded as the second mean; the product of the first mean and the second mean is recorded as the first product; the mean of the gray values ​​of all pixels in all low green sensitivity connected domains in the remote sensing gray image of the artificial forest to be detected is recorded as the third mean; the mean of the edge feature parameters of all low green sensitivity connected domains in the remote sensing gray image of the artificial forest to be detected is recorded as the fourth mean; the product of the third mean and the fourth mean is recorded as the second product; the ratio of the first product to the second product is used as the enhancement coefficient of the remote sensing gray image of the artificial forest to be detected; The remote sensing grayscale image of the artificial forest to be detected is enhanced according to the enhancement coefficient to obtain an enhanced remote sensing image of the artificial forest to be detected; and intelligent identification of forest diseases and insect pests is performed based on the enhanced remote sensing image of the artificial forest to be detected.

2. The method for intelligent identification of forest pests and diseases based on UAV remote sensing according to claim 1 is characterized in that: The method of performing connected domain analysis on the remote sensing grayscale image of the artificial forest to be detected to obtain a plurality of circle-like connected domains in the remote sensing grayscale image of the artificial forest to be detected includes the following specific methods: A gradient image of a remote sensing grayscale image of an artificial forest to be detected is obtained, and a watershed segmentation algorithm is performed on the gradient image to obtain multiple regions of the remote sensing grayscale image of the artificial forest to be detected. Each region is used as a quasi-circle connected domain, and then several quasi-circle connected domains in the remote sensing grayscale image of the artificial forest to be detected are obtained.

3. The method for intelligent identification of forest pests and diseases based on UAV remote sensing according to claim 1 is characterized in that: The edge feature parameters of each quasi-circle connected domain are obtained by performing edge concave-convex feature analysis on each quasi-circle connected domain, and the specific method includes: For any quasi-circle connected domain in the remote sensing grayscale image of the artificial forest to be detected, edge detection is performed on the quasi-circle connected domain to obtain all edge pixel points in the quasi-circle connected domain; Obtaining the center position of the circle-like connected domain; According to the position distribution between the center position of the quasi-circle connected domain and the edge pixel points, all salient point angles of the quasi-circle connected domain are obtained; The normalized value of the average of the angle values ​​of all salient points in the quasi-circle connected domain is used as the edge feature parameter of the quasi-circle connected domain.

4. The method for intelligent identification of forest pests and diseases based on UAV remote sensing according to claim 3 is characterized in that: The specific method of obtaining the center position of the quasi-circle connected domain includes: All edge pixel points in the quasi-circle connected domain are arbitrarily combined in pairs to obtain several edge point combinations of the quasi-circle connected domain; for any edge point combination, the Euclidean distance between two edge pixel points in the edge point combination is recorded as the first distance of the edge point combination; among all edge point combinations in the quasi-circle connected domain, the edge point combination with the largest first distance is recorded as the target edge point combination of the quasi-circle connected domain; the center point of the straight line between two edge pixel points in the target edge point combination is used as the center position of the quasi-circle connected domain.

5. The method for intelligent identification of forest pests and diseases based on UAV remote sensing according to claim 3 is characterized in that: The method of obtaining all salient point angles of the quasi-circle connected domain according to the position distribution between the center position of the quasi-circle connected domain and the edge pixel points includes: For any edge pixel point in the quasi-circle connected domain, the Euclidean distance between the edge pixel point and the center position of the quasi-circle connected domain is recorded as the target distance of the edge pixel point; the left adjacent edge pixel point and the right adjacent edge pixel point of the edge pixel point are both used as comparison pixel points of the edge pixel point; Among the edge pixel points and the compared edge pixel points, if the target distance of the edge pixel point is the maximum value, the edge pixel point is recorded as a convex point of the quasi-circular connected domain; if the target distance of the edge pixel point is the minimum value, the edge pixel point is recorded as a concave point of the quasi-circular connected domain; For any convex point of the quasi-circular connected domain, an angle formed by the convex point and two adjacent concave points with the convex point as the vertex is recorded as the convex point angle of the quasi-circular connected domain.

6. The method for intelligent identification of forest pests and diseases based on UAV remote sensing according to claim 1 is characterized in that: The specific method of screening all the circle-like connected domains by edge feature parameters to obtain all the suspected pests and diseases connected domains includes: Preset a feature threshold parameter For any circle-like connected domain in the remote sensing grayscale image of the artificial forest to be detected, if the edge feature parameter of the circle-like connected domain is less than the feature threshold parameter , and record the circle-like connected domain as the suspected pest and disease connected domain.

7. The method for intelligent identification of forest pests and diseases based on UAV remote sensing according to claim 1 is characterized in that: The method of dividing all suspected pest and disease connected domains into high green sensitivity connected domains and low green sensitivity connected domains based on green sensitivity includes: Preset a sensitive threshold parameter For any suspected pest connected domain, if the green sensitivity of the suspected pest connected domain is less than the sensitivity threshold parameter , the suspected pest connected domain is recorded as a low green sensitivity connected domain; if the green sensitivity of the suspected pest connected domain is greater than or equal to the sensitivity threshold parameter , the suspected pest and disease connected domain is recorded as a high green sensitivity connected domain.

8. The method for intelligent identification of forest pests and diseases based on UAV remote sensing according to claim 1 is characterized in that: The specific method of intelligently identifying forest pests and diseases based on the enhanced remote sensing image of the artificial forest to be detected includes: The enhanced remote sensing image of the artificial forest to be detected is input into the trained neural network to obtain the forest area with pests and diseases in the enhanced remote sensing image of the artificial forest to be detected.

Images