Soybean mosaic disease detection method, device and equipment based on unmanned aerial vehicle image

Abstract

The invention provides a soybean mosaic disease detection method, device and equipment based on an unmanned aerial vehicle image, and relates to the technical field of image processing.The method comprises the steps that a visible light image and a multispectral image of a soybean planting area are collected according to a preset route and flight height through an unmanned aerial vehicle carrying a high-definition optical camera and a multispectral camera; the multispectral image comprises a plurality of wavebands; preprocessing the visible light image and the multispectral image; extracting color features from the preprocessed visible light image and extracting spectral features from the preprocessed multispectral image; based on the color features and the spectral features, soybean plants in the soybean planting area are classified through the optimized SVM model, and a disease infection classification result of the soybean planting area is output. According to the invention, the accuracy of disease identification is improved, and the misjudgment rate is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a detection method, device and equipment for soybean mosaic virus based on unmanned aerial vehicle (UAV) images. Background Art

[0002] Soybean Mosaic Virus (SMV) is a common viral disease in soybean production, which is caused by Soybean Mosaic Virus and is mainly transmitted by aphids or seed-borne virus. It is particularly important to identify and detect soybean mosaic virus.

[0003] In existing research, visible light band images are used to identify soybean mosaic virus, and diseased plants are identified by a simple image threshold segmentation method. For example, a fixed range of Red-Green-Blue (RGB) channel values is set, and pixels in the image that fall within this range are determined as diseased plant areas. However, due to the interference of factors such as natural light changes and soil background differences, this method of single band and simple threshold setting has a very high misjudgment rate and cannot adapt to the detection of soybean mosaic virus in complex field environments. Summary of the Invention

[0004] The present invention provides a detection method, device and equipment for soybean mosaic virus based on UAV images, which are used to solve the defect that the misjudgment rate in the prior art is extremely high and cannot adapt to the detection of soybean mosaic virus in complex field environments, and reduce the misjudgment rate of soybean mosaic virus detection.

[0005] In a first aspect, the present invention provides a detection method for soybean mosaic virus based on UAV images, and the method includes the following steps: Collect visible light images and multispectral images of a soybean planting area by a UAV equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude; the multispectral images include multiple bands; Perform preprocessing on the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; Extract color features from the preprocessed visible light images and extract spectral features from the preprocessed multispectral images; Based on the color features and the spectral features, use an optimized Support Vector Machine (SVM) model to classify soybean plants in the soybean planting area, and output a diseased classification result of the soybean planting area; the diseased classification result includes potential diseased spot areas or healthy areas; the objective function of the optimized SVM model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range where the color features and the spectral features of healthy plants are located.

[0006] A method for detecting soybean mosaic virus based on UAV images provided by the present invention. Before collecting visible light images and multispectral images of the soybean planting area by a UAV equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude, it further includes: Obtain the length and width of the soybean planting area; Based on the length and the width, determine the camera field of view angle and the flight altitude of the UAV; Based on the camera field of view angle and the flight altitude, determine the adjacent flight path spacing of the preset flight path so that the acquired images can fully cover the soybean planting area without obvious splicing gaps.

[0007] A method for detecting soybean mosaic virus based on UAV images provided by the present invention. The preprocessing of the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images includes: Use the median filtering algorithm to denoise the visible light images and the multispectral images to obtain the denoised visible light images and the denoised multispectral images; Based on ground control points, perform geometric correction on the denoised visible light images and the denoised multispectral images through an affine transformation model to obtain geometrically corrected visible light images and geometrically corrected multispectral images; the ground control points are marked points with known coordinates preset in the soybean planting area; Based on the geometrically corrected visible light images and geometrically corrected multispectral images, convert the image radiance values into surface true reflectance according to the radiative transfer model to obtain the preprocessed visible light images and the preprocessed multispectral images.

[0008] A method for detecting soybean mosaic virus based on UAV images provided by the present invention. The use of the median filtering algorithm to denoise the visible light images and the multispectral images to obtain the denoised visible light images and the denoised multispectral images includes: For each pixel point in the visible light images and the multispectral images, determine the pixel values of all neighborhood windows of each pixel point; Determine the median corresponding to the pixel values of all neighborhood windows of each pixel point; Based on each median, determine the denoised pixel value corresponding to each pixel point; Based on the denoised pixel values corresponding to each pixel point, determine the denoised visible light images and the denoised multispectral images.

[0009] A method for detecting soybean mosaic virus based on UAV images provided by the present invention, extracting color features from the preprocessed visible light images and extracting spectral features from the preprocessed multispectral images, includes: Based on the preprocessed visible light images, determine the mean and variance of the red, green, and blue (RGB) channels of the soybean planting area; Determine the RGB channel mean and the variance as the color features; Based on the preprocessed multispectral images, determine the ratio vegetation index (RVI), normalized difference vegetation index (NDVI), and difference vegetation index (DVI) of the soybean planting area; Determine the ratio vegetation index (RVI), normalized difference vegetation index (NDVI), and difference vegetation index (DVI) as the spectral features.

[0010] A method for detecting soybean mosaic virus based on UAV images provided by the present invention, classifying soybean plants in the soybean planting area by using an optimized support vector machine (SVM) model based on the color features and the spectral features, and outputting the disease classification results of the soybean planting area, includes: Based on the color features and the spectral features, determine a target feature vector; the target feature vector corresponds to multiple pixels; Input the target feature vector into the optimized support vector machine (SVM) model to obtain the objective function values corresponding to multiple pixel points; For any one of the pixel points, if the objective function value corresponding to the pixel point is greater than zero, determine the health status of the soybean plant corresponding to the pixel point as diseased; if the objective function value corresponding to the pixel point is less than zero, determine the health status of the soybean plant corresponding to the pixel point as healthy.

[0011] A method for detecting soybean mosaic virus based on UAV images provided by the present invention, the optimized support vector machine (SVM) model is constructed through the following steps, includes: Obtain a plurality of sample data; the sample data are labeled healthy samples and diseased samples; Based on the sample feature vectors and labels of the sample data, construct constraint conditions; the sample feature vectors are determined based on the color features of the sample data and the spectral features of the sample data; the constraint conditions are used to ensure that the samples are correctly classified and the margin meets specific requirements; Based on the sample feature vectors, labels of the sample data, and the constraint conditions, solve the objective function by the Lagrange multiplier method to obtain the hyperplane parameters; Based on the hyperplane parameters, determine the optimized support vector machine (SVM) model.

[0012] A method for detecting soybean mosaic virus based on UAV images provided by the present invention further includes: Performing clustering analysis and morphological processing on the potential lesion areas in the diseased classification results to optimize the lesion boundaries and label the diseased areas; Generating a disease detection report based on the number of diseased plants and the proportion of diseased area in the diseased areas.

[0013] In a second aspect, the present invention further provides a device for detecting soybean mosaic virus based on UAV images, which includes the following modules: An acquisition module, configured to collect visible light images and multispectral images of a soybean planting area through a UAV equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude; the multispectral images include multiple bands; A preprocessing module, configured to preprocess the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; A feature extraction module, configured to extract color features from the preprocessed visible light images and extract spectral features from the preprocessed multispectral images; A classification and detection module, configured to classify soybean plants in the soybean planting area based on the color features and the spectral features by using an optimized support vector machine (SVM) model, and output the diseased classification results of the soybean planting area; the diseased classification results include potential lesion areas or healthy areas; the objective function of the optimized support vector machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range where the color features and the spectral features of healthy plants are located.

[0014] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method for detecting soybean mosaic virus based on UAV images as described in any one of the above is implemented.

[0015] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for detecting soybean mosaic virus based on UAV images as described in any one of the above is implemented.

[0016] In a fifth aspect, the present invention further provides a computer program product, including a computer program. When the computer program is executed by a processor, the method for detecting soybean mosaic virus based on UAV images as described in any one of the above is implemented.

[0017] The soybean mosaic disease detection method, device and equipment based on UAV images provided by the present invention first collect visible light images and multispectral images of the soybean planting area through a UAV equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude. The multispectral images include multiple bands. Then, preprocess the visible light images and multispectral images to obtain the preprocessed visible light images and preprocessed multispectral images. Further, extract color features from the preprocessed visible light images and spectral features from the preprocessed multispectral images. Furthermore, based on the color features and spectral features, use an optimized support vector machine (SVM) model to classify the soybean plants in the soybean planting area and output the disease classification result of the soybean planting area. The disease classification result includes potential diseased areas or healthy areas, where the objective function of the optimized SVM model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range where the color features and spectral features of healthy plants are located.

[0018] In the present invention, UAV images are first obtained, including visible light images and multispectral images. The multispectral images include multiple bands. Then, through image preprocessing and feature extraction, color features and spectral features are obtained. Furthermore, based on the color features and spectral features, an optimized SVM model is used for detection to classify the soybean plants and obtain the disease classification result. The visible light and multispectral images collected by the UAV are comprehensively utilized, and the color features and spectral features contained therein are deeply mined and analyzed. Then, the optimized SVM model is used for detection, which improves the accuracy of disease identification and reduces the misjudgment rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in 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 drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0020] Figure 1 is one of the flow schematic diagrams of the soybean mosaic disease detection method based on UAV images provided by the present invention.

[0021] Figure 2 is the second flow schematic diagram of the soybean mosaic disease detection method based on UAV images provided by the present invention.

[0022] Figure 3 is the structural schematic diagram of the soybean mosaic disease detection device based on UAV images provided by the present invention.

[0023] Figure 4 is the structural schematic diagram of the electronic device provided by the present invention. Detailed implementation manners

[0024] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0025] To more clearly understand the various embodiments provided by the present invention, the technical content related to the present invention is first introduced as follows: Traditional detection methods for soybean mosaic virus mostly rely on manual field inspections. Planting personnel regularly walk in the field, visually observe the appearance of soybean plants with the naked eye, pay attention to whether the leaves show typical symptoms such as yellow-green mosaic, wrinkling, and deformity, and whether the overall growth trend of the plants is withered. For suspected diseased plants, a closer and more detailed inspection is carried out, such as checking whether there are disease spots, bacterial plaques, etc. on the back of the leaves. Manual field inspections consume a large amount of human and time costs. Especially in large-scale planting scenarios, it is very difficult to detect in a timely and comprehensive manner. Moreover, manual judgment is highly subjective and is easily affected by factors such as experience and fatigue, resulting in frequent misjudgments and missed judgments.

[0026] In addition, some studies use fixed-point monitoring equipment, set up fixed monitoring stations in the field, equipped with optical sensors, and regularly collect spectral information of specific parts of soybean plants. By analyzing these local spectral data, the health status of the plants is judged. However, the monitoring range is limited and it is difficult to cover large-scale planting areas. Although the fixed-point monitoring equipment can obtain spectral data with a certain accuracy, due to its fixed-position characteristics, the monitoring area is limited and it cannot reflect the overall disease situation of the entire planting area. There are easily monitoring blanks in the edge areas and complex terrain areas of the planting area, and it cannot provide comprehensive and accurate information for disease prevention and control.

[0027] Some studies have tried to introduce some general machine learning algorithms, such as decision trees, neural networks, etc., and use limited field-collected sample data for training. However, due to the lack of optimized design for the characteristics of soybean mosaic virus, the model is blind in feature selection and parameter adjustment, has insufficient processing ability for multi-source and massive UAV image data, has poor generalization performance, and the application effects are unstable under different regions and different planting conditions.

[0028] Based on the deficiencies existing in the prior art, the present invention aims to overcome the defects of the prior art and provides a method for detecting soybean mosaic virus based on UAV images.

[0029] The following combines with Figures 1-4Describe a method, device, and equipment for detecting soybean mosaic virus based on unmanned aerial vehicle (UAV) images according to the present invention.

[0030] Figure 1 It is one of the schematic flowcharts of the method for detecting soybean mosaic virus based on UAV images provided by the present invention. As Figure 1 shown, the method includes the following: Step 101: Use a UAV equipped with a high-definition optical camera and a multispectral camera to collect visible light images and multispectral images of the soybean planting area according to a preset flight path and altitude; the multispectral images include multiple bands; Specifically, the images obtained in this embodiment include multi-band images. During the critical growth stages of soybeans, such as the flowering stage and the pod-setting stage, the symptoms of soybean mosaic virus are more obvious at this time, and a UAV equipped with a high-definition optical camera and a multispectral camera is used for shooting.

[0031] For example, use a UAV equipped with a high-definition optical camera and a multispectral camera to collect visible light images and multispectral images of the soybean planting area according to a preset flight path and altitude.

[0032] The flight path and altitude of the UAV can be reasonably planned in advance according to the terrain and area of the planting area to ensure that the obtained images can fully cover the planting area without obvious splicing gaps, so as to obtain high-resolution visible light images and multispectral images.

[0033] Step 102: Preprocess the visible light images and multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; Specifically, after obtaining the visible light images and multispectral images, an optimized support vector machine (SVM) model can be used for recognition based on the visible light images and multispectral images. Before recognition, the visible light images and multispectral images can be preprocessed first. The purpose of preprocessing is to improve the image quality for subsequent analysis and processing. The preprocessing operations in the present invention include denoising, geometric correction, and radiometric correction.

[0034] Among them, denoising refers to removing the random noise introduced into the image due to sensor noise, transmission interference, or environmental factors, thereby improving the signal-to-noise ratio (SNR) of the image. The signal-to-noise ratio is the core index for measuring the signal quality, representing the ratio of the effective signal intensity to the background noise intensity.

[0035] Geometric correction refers to correcting the geometric deformation in the image caused by sensor attitude, terrain undulation, or projection method, so that the image has real geographical space coordinates. The correction steps include geometric modeling, control point matching, and resampling.

[0036] Radiometric calibration refers to eliminating radiometric distortion caused by inconsistent sensor responses, atmospheric scattering, or illumination changes, and restoring the true reflectance or radiance value of ground objects. Common methods include relative radiometric calibration and absolute radiometric calibration.

[0037] Step 103: Extract color features from the preprocessed visible light image and spectral features from the preprocessed multispectral image. Specifically, color features are extracted based on the preprocessed visible light image to obtain a feature vector, and then it is input into the optimized Support Vector Machine (SVM) model for disease identification and classification.

[0038] The visible light image is used to extract color features, and the color features include the mean and variance of the three channels of Red-Green-Blue (RGB).

[0039] The multispectral image is used to extract spectral features. By calculating and analyzing the changes in these spectral features, it is possible to determine whether the plants are diseased and further locate potential diseased area regions.

[0040] The spectral features include Ratio Vegetation Index (RVI), Normalized Difference Vegetation Index (NDVI), and Difference Vegetation Index (DVI). The Ratio Vegetation Index (RVI) is a commonly used vegetation index that quantifies the growth status of vegetation by comparing the reflectance of the Near Infrared (NIR) band and the Red (R) band. The Normalized Difference Vegetation Index (NDVI) is an important indicator reflecting the health and growth status of vegetation. It quantifies vegetation by calculating the difference between the reflectance of the near-infrared band and the red band. The Difference Vegetation Index (DVI) is an index composed of satellite detection data of different bands, which can reflect the growth status of plants.

[0041] Step 104: Based on the color features and spectral features, use the optimized Support Vector Machine (SVM) model to classify soybean plants in the soybean planting area, and output the disease classification result of the soybean planting area; the disease classification result includes potential diseased area regions or healthy regions; the objective function of the optimized Support Vector Machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range where the color features and spectral features of healthy plants are located.

[0042] Specifically, after extracting the color features and spectral features, a target feature vector is determined based on the color features and spectral features. Furthermore, the target feature vector is input into an optimized support vector machines model to classify the soybean plants in the soybean planting area, and the disease classification result is input. The disease classification result includes potential diseased areas or healthy areas.

[0043] Among them, the objective function of the optimized support vector machine SVM model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range where the color features and spectral features of healthy plants are located. For example, the comprehensive discrimination criterion is: When the average value of the green channel in the RGB channels of the leaf exceeds the threshold range , and at the same time the value of the RVI of the plant is lower than the healthy threshold RVI min , it is determined that the plant is diseased.

[0044] The method provided in this embodiment first uses a drone equipped with a high-definition optical camera and a multispectral camera to collect visible light images and multispectral images of the soybean planting area according to a preset flight path and flight altitude. The multispectral images include multiple bands. Then, the visible light images and multispectral images are preprocessed to obtain the preprocessed visible light images and preprocessed multispectral images. Further, color features are extracted from the preprocessed visible light images, and spectral features are extracted from the preprocessed multispectral images. Furthermore, based on the color features and spectral features, an optimized support vector machine SVM model is used to classify the soybean plants in the soybean planting area, and the disease classification result of the soybean planting area is output. The disease classification result includes potential diseased areas or healthy areas. Among them, the objective function of the optimized support vector machine SVM model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range where the color features and spectral features of healthy plants are located.

[0045] In the present invention, first, drone images are acquired, including visible light images and multispectral images. The multispectral images include multiple bands. Then, through image preprocessing and feature extraction, color features and spectral features are obtained. Furthermore, based on the color features and spectral features, an optimized support vector machine SVM model is used for detection, and the soybean plants are classified to obtain the disease classification result. The visible light and multispectral images collected by the drone are comprehensively utilized, the color features and spectral features contained therein are deeply mined and analyzed, and then the optimized SVM model is used for detection, which improves the accuracy of disease recognition and reduces the misjudgment rate.

[0046] It should be noted that each embodiment of this application can be freely combined, the order can be swapped, or each can be executed alone, and does not need to rely on or depend on a fixed execution order.

[0047] A method for detecting soybean mosaic virus based on UAV images provided by the present invention, before collecting visible light images and multispectral images of the soybean planting area by a UAV equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude, further includes: Obtain the length and width of the soybean planting area; Based on the length and width, determine the camera field of view angle and flight altitude of the UAV; Based on the camera field of view angle and flight altitude, determine the adjacent flight path spacing of the preset flight path, so that the obtained images can fully cover the soybean planting area without obvious splicing gaps.

[0048] Specifically, in some embodiments, before step 101, it further includes: First, in order to reasonably plan the flight path and altitude according to the terrain and area of the planting area, ensure that the obtained images can fully cover the planting area without obvious splicing gaps, first obtain the length L and width W of the soybean planting area, and based on the length and width, determine the camera field of view angle of the UAV and flight altitude H.

[0049] Furthermore, based on the camera field of view angle and flight altitude, determine the adjacent flight path spacing of the preset flight path, so that the obtained images can fully cover the soybean planting area without obvious splicing gaps, ensure that there is no omission in shooting, and obtain high-resolution visible light images and multispectral images. The calculation formula for the adjacent flight path spacing d is:

[0050] wherein, represents the adjacent flight path spacing, H represents the flight altitude, represents the camera field of view angle of the UAV.

[0051] After determining the adjacent flight path spacing, the UAV collects images according to the adjacent flight path spacing and flight altitude, which can ensure that there is no omission in image shooting.

[0052] The method provided in this embodiment determines the camera field of view angle and flight altitude of the UAV based on the length and width of the soybean planting area; then, based on the camera field of view angle and flight altitude, determines the adjacent flight path spacing of the preset flight path, so that the obtained images can fully cover the soybean planting area without obvious splicing gaps, and can ensure that there is no omission in image shooting.

[0053] A method for detecting soybean mosaic virus based on UAV images provided by the present invention preprocesses the visible light images and multispectral images to obtain preprocessed visible light images and preprocessed multispectral images, including: Use the median filtering algorithm to denoise the visible light images and multispectral images to obtain denoised visible light images and denoised multispectral images; Based on ground control points, geometric correction is performed on the denoised visible light image and the denoised multispectral image through an affine transformation model to obtain the geometrically corrected visible light image and the geometrically corrected multispectral image; the ground control points are marked points with known coordinates preset in the soybean planting area. Based on the geometrically corrected visible light image and the geometrically corrected multispectral image, according to the radiation transfer model, the image radiance value is converted into the true surface reflectance to obtain the preprocessed visible light image and the preprocessed multispectral image.

[0054] Specifically, in some embodiments, the steps of image preprocessing in step 102 are as follows: First, the median filter algorithm is used to denoise the visible light image and the multispectral image to obtain the denoised visible light image and the denoised multispectral image. The noise interference is removed from the denoised image, making the denoised image clearer.

[0055] Furthermore, based on ground control points, geometric correction is performed on the denoised visible light image and the denoised multispectral image through an affine transformation model to obtain the geometrically corrected visible light image and the geometrically corrected multispectral image.

[0056] Among them, the ground control points are, for example, marked points with known coordinates preset in the soybean planting area.

[0057] For example, through the affine transformation model, according to the correspondence between the coordinates of the ground control points in the image and the actual coordinates, the geometric distortion caused by the flight attitude and terrain undulation is corrected, so that the ground objects in the image can accurately correspond to the actual geographical positions. The affine transformation model (Affine Transformation Model) is a geometric transformation method used to describe the combination of linear transformations such as translation, rotation, scaling, and shear (shearing) of objects in two-dimensional or three-dimensional space. Its core feature is to maintain straight lines and parallelism (parallel lines remain parallel after transformation), but not necessarily angles or lengths. The matrix form of two-dimensional affine transformation is as follows: For a point (x, y) in two-dimensional space, the coordinates (x′, y′) after affine transformation can be expressed as:

[0058] Among them, a, b, d, e control rotation, scaling, and shear; c, f control translation; the matrix has a total of six degrees of freedom (six independent parameters).

[0059] Among them, the specific parameter decomposition is as follows: Translation: Translation is directly controlled by c and f; rotation is achieved through a = cosθ, b = -sinθ, e = cosθ; scaling: the scaling in the x and y directions is respectively controlled by the diagonal elements a and e; shearing: the shearing deformation along the coordinate axes is controlled by the non-diagonal elements b and d.

[0060] Translation: Further, based on the geometrically corrected visible light image and the geometrically corrected multispectral image, according to the radiative transfer model, the image radiance value is converted into the true surface reflectance of the earth, and the preprocessed visible light image and the preprocessed multispectral image are obtained.

[0061] Translation: The radiance value is the "raw signal" received by the sensor and is greatly affected by the environment. It represents the electromagnetic radiation energy received by the sensor in a specific band, with the unit of W / (m²·sr·μm), that is, the radiation power per unit area, per unit solid angle, and per unit wavelength, which is also the radiation flux per square meter per steradian per micrometer. The radiation flux refers to the electromagnetic wave energy that vertically passes through a unit area within a unit time, and its unit is usually watts per square meter (W / m²).

[0062] Translation: The surface reflectance of the earth is the "essential attribute" of the ground object and needs to be extracted through atmospheric correction. It is the cornerstone of remote sensing quantitative analysis. The surface reflectance surface is the reflection ability of the ground object surface to the incident solar radiation and is a dimensionless ratio (0 - 1), defined as: surface = reflected radiation energy / incident radiation energy Translation: This radiation correction process needs to consider the influence of factors such as atmospheric scattering and absorption on the radiance. Through a series of calculations and corrections, the radiance standard is unified, and the image quality is improved for subsequent analysis and processing.

[0063] Translation: The method provided in this embodiment performs denoising, geometric correction, and radiation correction on the collected original image, improving the image quality. Then, feature extraction is performed on the preprocessed image, and leaf disease detection and recognition are carried out based on the features, improving the accuracy of detection.

[0064] Translation: According to a method for detecting soybean mosaic disease based on UAV images provided by the present invention, the median filtering algorithm is used to denoise the visible light image and the multispectral image, and the denoised visible light image and the denoised multispectral image are obtained, including: Translation: For each pixel point in the visible light image and the multispectral image, determine the pixel values of all neighborhood windows of each said pixel point; Translation: Determine the median corresponding to the pixel values of all neighborhood windows of each said pixel point; Translation: Based on each median, determine the denoised pixel value corresponding to each pixel point; Translation: Based on the denoised pixel values corresponding to each pixel point, determine the denoised visible light image and the denoised multispectral image.

[0065] Specifically, in some embodiments, the specific implementation process of denoising the visible light image and the multispectral image includes the following steps: First, for each pixel point in the visible light image and the multispectral image, determine the pixel values of all neighborhood windows of each pixel point.

[0066] It should be noted that any pixel point corresponds to a window with a fixed size, such as 3×3. During the denoising process, for any pixel point, the pixel values within its neighborhood window are selected. Usually, the neighborhood windows can be of equal size.

[0067] Furthermore, determine the median of the pixel values corresponding to all neighborhood windows of each pixel point. Based on each median, determine the denoised pixel value corresponding to each pixel point. Based on the denoised pixel values corresponding to each pixel point, determine the denoised visible light image and the denoised multispectral image. That is, take the median of all these pixel values as the denoised value of this point, so as to remove the noise interference in the image and make the image clearer.

[0068] The method provided in this embodiment uses the median filtering algorithm to denoise the pixel points in the image, thereby removing the noise interference in the image and making the overall image clearer.

[0069] According to a method for detecting soybean mosaic virus based on UAV images provided by the present invention, color features are extracted from the preprocessed visible light image, and spectral features are extracted from the preprocessed multispectral image, including: Based on the preprocessed visible light image, determine the mean and variance of the red, green, and blue RGB channels in the soybean planting area; Determine the mean and variance of the RGB channels as color features; Based on the preprocessed multispectral image, determine the ratio vegetation index RVI, normalized vegetation index NDVI, and difference vegetation index DVI in the soybean planting area; Determine the ratio vegetation index RVI, normalized vegetation index NDVI, and difference vegetation index DVI as spectral features.

[0070] Specifically, in some embodiments, the steps of feature extraction in step 103 include the following: First, based on the preprocessed visible light image, determine the mean and variance of the red, green, and blue RGB channels in the soybean planting area; Furthermore, determine the mean and variance of the RGB channels as color features; Furthermore, based on the preprocessed multispectral image, determine the ratio vegetation index RVI, normalized vegetation index NDVI, and difference vegetation index DVI in the soybean planting area; Further, the ratio vegetation index (RVI), the normalized difference vegetation index (NDVI), and the difference vegetation index (DVI) are determined as spectral features.

[0071] For the method provided in this embodiment, first, based on the preprocessed visible light image, the mean and variance of the red, green, and blue (RGB) channels of the soybean planting area are determined; based on the preprocessed multispectral image, the ratio vegetation index (RVI), the normalized difference vegetation index (NDVI), and the difference vegetation index (DVI) of the soybean planting area are determined. Furthermore, the mean and variance of the RGB channels are determined as color features, and the ratio vegetation index (RVI), the normalized difference vegetation index (NDVI), and the difference vegetation index (DVI) are determined as spectral features. By comprehensively using the visible light image and the multispectral image collected by the unmanned aerial vehicle, the color features and spectral features contained therein are deeply mined and analyzed. By accurately calculating the mean, variance of the RGB channels, and various vegetation indices, the interference of environmental factors on the detection results is minimized, and the accuracy of disease identification is improved.

[0072] According to a soybean mosaic disease detection method based on unmanned aerial vehicle images provided by the present invention, based on color features and spectral features, an optimized support vector machine (SVM) model is used to classify soybean plants in the soybean planting area, and a disease classification result of the soybean planting area is output, including: Based on color features and spectral features, a target feature vector is determined; the target feature vector corresponds to multiple pixels; The target feature vector is input into the optimized support vector machine (SVM) model to obtain the objective function values corresponding to multiple pixel points; For any pixel point, if the objective function value corresponding to the pixel point is greater than zero, the health status of the soybean plant corresponding to the pixel point is determined to be diseased; if the objective function value corresponding to the pixel point is less than zero, the health status of the soybean plant corresponding to the pixel point is determined to be healthy.

[0073] Specifically, in some embodiments, the process of disease identification and classification in step 104 can be implemented through the following steps: First, based on color features and spectral features, a target feature vector is determined, and the target feature vector corresponds to multiple pixels.

[0074] Under normal growth conditions, the content of pigments such as chlorophyll in healthy soybean leaves is relatively stable, presenting a relatively uniform green tone in the RGB color space. When the leaves are infected with mosaic disease, the virus will affect the physiological structure and pigment synthesis of the leaves, resulting in a typical mosaic appearance of yellow and green patches on the leaf surface. This color change will be reflected in the RGB color space, causing changes in the mean and variance of the RGB three channels. By calculating and analyzing these changes, it can be used as one of the bases for judging whether the leaves are diseased.

[0075] Suppose a leaf area to be detected contains n pixel points, and the red, green, and blue channel values of each pixel point in this area are denoted as R i , G i , B i , respectively. Then the mean calculation formula for the three RGB channels is as follows:

[0076]

[0077]

[0078] Among them, is the mean value of the red channel of the leaf area to be detected, is the mean value of the green channel of the leaf area to be detected, is the mean value of the blue channel of the leaf area to be detected, and n represents the total number of pixel points.

[0079] The variance calculation formula is:

[0080]

[0081]

[0082] Among them, is the variance of the red channel of the leaf area to be detected, is the variance of the green channel of the leaf area to be detected, is the variance of the green channel of the leaf area to be detected, is the mean value of the red channel of the leaf area to be detected, is the mean value of the green channel of the leaf area to be detected, is the mean value of the blue channel of the leaf area to be detected, n represents the total number of pixel points, and the red, green, and blue channel values of each pixel point in this area are denoted as Ri, Gi, Bi.

[0083] Before determining the leaf area to be detected, it is necessary to determine the channel threshold range of healthy leaves.

[0084] For example, by statistically comparing and analyzing the average values and variances of the pixel values of the three RGB bands of visible light healthy samples, select the value of the channel with the most obvious difference value (such as the average value or variance value of the green channel) as the evaluation value of the healthy sample.

[0085] Specifically, through the statistics of a large number of healthy samples, the green channel threshold range of healthy leaves is (k usually takes 1 - 3).

[0086] Subject to the normal distribution, k = 1 - 3 is the commonly used confidence interval range in statistics, which balances sensitivity (the ability to detect anomalies) and specificity (the ability to avoid false positives). The smaller the K (such as 1), the narrower the threshold range, making it easier to detect subtle anomalies (high sensitivity), but there may be false positives for healthy samples (low specificity); the larger the K (such as 3), the wider the threshold range, reducing the risk of missed detections (high specificity), but early anomalies may be overlooked (low sensitivity).

[0087] During the recognition process, as long as the calculated average value of the green channel of the plant or the variance value exceeds the said range (the judgment value range of healthy samples) , it is determined as abnormal, that is, the plant of the leaf to be detected is a diseased plant.

[0088] Furthermore, the spectral features are, for example, the ratio vegetation index (RVI), the normalized vegetation index (NDVI), and the difference vegetation index (DVI).

[0089] In the multi-spectral band, there are significant differences in the spectral reflection characteristics between healthy soybean plants and diseased plants. Due to the rich chlorophyll in the leaves of healthy plants, they have a strong absorption ability for red light and a prominent reflection ability for near-infrared light; while for diseased plants, due to damaged leaves, the chlorophyll content is significantly reduced, the photosynthesis efficiency is greatly reduced, and both the absorption of red light and the reflection ability of near-infrared light are significantly decreased. By calculating and analyzing the changes in these spectral features, it is possible to determine whether the plant is diseased and further locate potential diseased areas.

[0090] The calculation formula for the ratio vegetation index (RVI) is:[[]]

[0091] where represents the reflectance in the near-infrared band, represents the reflectance in the red light band.

[0092] The calculation formula for the normalized vegetation index (NDVI) is:[[]]

[0093] where represents the reflectance in the near-infrared band, represents the reflectance in the red light band.

[0094] The calculation formula for the difference vegetation index (DVI) is:[[]]

[0095] Based on experience or statistical analysis of the spectral data of healthy plants, determine the RVI threshold range of healthy plants. For example, the RVI of general healthy green vegetation is much greater than 1. Assume the healthy threshold range is , when the calculated value is lower than , it is considered that the spectral characteristics of the plant are abnormal.

[0096] Furthermore, based on the extracted color features and spectral features, determine the target feature vector.

[0097] Further, input the target feature vector into the optimized support vector machine (SVM) model to obtain the values of the objective function corresponding to multiple pixel points. The objective function is:

[0098] where , represent the hyperparameters of the support vector machine, where is the normal vector of the hyperplane, is the sample feature vector (including color features and spectral features), and b is the bias term. The corresponding hyperparameters will be determined during the model training process.

[0099] Further, for any pixel point, if the value of the objective function corresponding to the pixel point is greater than zero, determine the health status of the soybean plant corresponding to the pixel point as diseased; if the value of the objective function corresponding to the pixel point is less than zero, determine the health status of the soybean plant corresponding to the pixel point as healthy.

[0100] During the classification process, for a new sample, calculate .

[0101] If f(x)>0, then it is judged as a healthy plant; if f(x)<0, then it is judged as a diseased plant.

[0102] For the method provided in this embodiment, input the image of the soybean planting area to be detected into the trained model. The model will judge and classify the health status of the soybean plants corresponding to each pixel point in the image according to the learned features and classification rules, output the information of the diseased area and the healthy area, and obtain the diseased classification result. The optimized support vector machine (SVM) model can operate reliably in soybean fields in different regions and different planting environments, accurately detect soybean mosaic disease, and achieve good generalization performance.

[0103] According to a method for detecting soybean mosaic disease based on UAV images provided by the present invention, the optimized support vector machine (SVM) model is constructed through the following steps, including: Obtain multiple sample data; the sample data are labeled healthy samples and diseased samples; Construct constraints based on the sample feature vectors and labels of each sample data; the sample feature vectors are determined based on the color features of each sample data and the spectral features of each sample data; the constraints are used to ensure that the samples are correctly classified and the margin meets specific requirements; Based on the sample feature vectors, labels, and constraints of each sample data, solve the objective function by the Lagrange multiplier method to obtain the hyperplane parameters; Based on the hyperplane parameters, determine the optimized support vector machine (SVM) model.

[0104] Specifically, in some embodiments, the process of constructing the optimized SVM model includes: First, obtain multiple sample data, where the sample data are labeled healthy samples and diseased samples. Optionally, the sample data can be divided into a training set and a test set. The training set is used to train the model to learn the relationship between the feature patterns of healthy and diseased samples and their corresponding labels (healthy or diseased). The test set is used to evaluate the performance and generalization ability of the model to ensure that the model can accurately classify samples it has not seen before.

[0105] Further, construct constraints based on the sample feature vectors and labels of each sample data.

[0106] Among them, the sample feature vectors are determined based on the color features of each sample data and the spectral features of each sample data, and the constraints are used to ensure that the samples are correctly classified and the margin meets specific requirements. The constraints are, for example: For each sample in the training set ( is the sample label, represents healthy, represents diseased), it is required to satisfy the constraint . This constraint ensures that the samples are correctly classified and the margin meets specific requirements.

[0107] Further, based on the sample feature vectors, labels, and constraints of each sample data, solve the objective function by the Lagrange multiplier method to obtain the hyperplane parameters.

[0108] The construction and solution of the objective function are as follows: The goal of the support vector machine (SVM) model is to find an optimal hyperplane , where is the normal vector of the hyperplane, is the sample feature vector (including color features, spectral features, etc.), and b is the bias term. The goal is to maximize the margin between the two types of samples (healthy and diseased) on both sides of the hyperplane, which is equivalent to minimizing , which is a convex quadratic optimization problem.

[0109] The solution process uses the Lagrange multiplier method, introducing the Lagrange multiplier , constructing the Lagrangian function , then taking the partial derivatives of w and b and setting them to 0, and further solving for . Then, based on , w and b are obtained, thereby determining the hyperplane parameters w and b.

[0110] Furthermore, based on the hyperplane parameters, an optimized support vector machine (SVM) model can be determined.

[0111] The method provided in this embodiment, by reasonably dividing the training set and the test set, constructing the constraint conditions and the objective function based on the characteristics of soybean mosaic disease, and accurately solving the hyperplane parameters using the Lagrange multiplier method, enables the model to learn the essential feature differences between diseased plants and healthy plants, obtaining an optimized support vector machine (SVM) model with high generalization performance.

[0112] According to a method for detecting soybean mosaic disease based on UAV images provided by the present invention, the method further includes: Performing clustering analysis and morphological processing on the potential lesion areas in the disease classification result to optimize the lesion boundaries and label the diseased areas; Generating a disease detection report based on the number of diseased plants and the proportion of diseased area in the diseased area.

[0113] Specifically, in some embodiments, the method further includes post-processing the disease classification result, and the specific steps are as follows: First, perform clustering analysis and morphological processing on the potential lesion areas in the disease classification result to optimize the lesion boundaries and label the diseased areas.

[0114] For example, assume that the feature vector of the potential lesion area is:

[0115] where X represents the feature vector of a potential lesion area, consisting of multiple feature dimensions, represents the i-th feature, which can be a position feature, a color feature, or a spectral feature, etc., specifically including: Position feature: the coordinates of the pixel (such as row and column numbers).

[0116] Color feature: RGB value color space parameters, etc.

[0117] Spectral feature: the band reflectance in multi-spectral or hyperspectral images (such as near-infrared and red light bands).

[0118] Texture feature: the texture statistics of the local area (such as the contrast and entropy of the gray-level co-occurrence matrix, etc.).

[0119] Each potential lesion area in the image is transformed into a numerical description through the eigenvector X, facilitating mathematical analysis and clustering by the computer.

[0120] By calculating the clustering distance such as the Euclidean distance, the potential lesion areas with similar features are clustered together to achieve clustering analysis.

[0121]

[0122] Among them, and represent the eigenvectors of two different potential lesion areas respectively, and are the values of the two eigenvectors in the k-th feature dimension, represents the Euclidean distance between the two eigenvectors and , K represents the k-th feature dimension, and n represents the total number of feature dimensions.

[0123] The above formula is used to calculate the sum of the squares of the differences between two eigenvectors in all feature dimensions and then take the square root. The smaller the calculated Euclidean distance, the closer the two potential lesion areas are in the feature space and they may belong to the same category (such as the same disease type or health status).

[0124] Example: Suppose =(100, 0.5), =(120, 0.7), where X1 is the reflectance in the red light band and X2 is the texture contrast.

[0125]

[0126] This indicates that there are significant differences between the two regions in terms of red light reflectance and texture contrast, and they do not belong to the same category.

[0127] Also, for example, morphological processing (i.e., dilation or erosion operations) can be performed on the binary image of the potential lesion area (the lesion area is -1 and the non-lesion area is 0), such as opening and closing operations (opening operation is dilation first and then erosion, and closing operation is erosion first and then dilation). This can not only merge adjacent potential lesion areas or remove small noise areas and optimize the boundaries of the lesion areas, but also be used to optimize the boundaries of the lesion areas, etc.

[0128] The purpose of dilation is to expand the lesion area. For each position z in the image, if the translated overlaps with the original image A (i.e., at least one lesion pixel -1 is covered), then the position z will be marked as a lesion (-1).

[0129] Example: If B is a 3×3 rectangular structuring element, dilation will expand the diseased area by 1 pixel in all directions and merge adjacent small diseased spots.

[0130] The dilation operation is as follows:

[0131] The purpose of erosion is to shrink the diseased area or remove noise. For each position z in the image, the position z will be retained as a diseased spot (-1) only when the translated B completely covers the diseased area (i.e., all positions covered by B are -1), otherwise it will be set to the background (0).

[0132] Example: If B is a 3×3 rectangular structuring element, erosion will "shave off" 1 pixel at the edge of the diseased spot and remove isolated small noise points.

[0133] The erosion operation is as follows:

[0134] where A is the original binary image (the diseased area is -1 and the background is 0), B is the structuring element (a predefined shape template), represents the translated position, represents the reflection of the structuring element B (symmetric flipping about the origin), represents translating the reflected to the position z.

[0135] The role of the structuring element B: The shape of B (such as rectangular, circular) determines the directionality of dilation / erosion (for example, a cross-shaped structuring element only expands along the horizontal and vertical directions). The size of B determines the scope of influence of the operation (for example, a 5×5 structuring element will produce a more significant dilation or erosion than a 3×3 one).

[0136] Reflection The significance of: In dilation, the reflection operation ensures symmetry. For example, if B is an arrow pointing to the right, after reflection it becomes an arrow pointing to the left, and during dilation it will expand evenly on both sides.

[0137] The particularity of the diseased spot value: In traditional morphology, the foreground is 1 and the background is 0. Here, the diseased spot is -1. So during dilation, as long as there is at least one -1 in the covered area, the current position z will be set to -1. During erosion, all positions covered by B must be -1 for the current position z to be retained as -1.

[0138] Furthermore, a disease detection report is generated based on the number of diseased plants and the proportion of the diseased area in the diseased region.

[0139] For example, accurately label the identified diseased areas on the image, and count information such as the number of diseased plants and the proportion of diseased area to generate a detailed detection report, providing a basis for the prevention and management of soybean mosaic virus disease. Among them, the proportion of diseased area is calculated by the following formula:

[0140] Among them, the total number of pixels in the image is N, the number of pixels labeled as diseased is n, and the proportion of diseased area is .

[0141] The method provided in this embodiment can perform clustering analysis and morphological processing on the potential diseased area in the diseased classification result to optimize the diseased area boundary and label the diseased area. Furthermore, based on the number of diseased plants and the proportion of diseased area in the diseased area, a disease detection report can be generated, providing a basis for the prevention and management of soybean mosaic virus disease.

[0142] Figure 2 is the second flow diagram of the soybean mosaic virus disease detection method based on UAV images provided by the present invention. As Figure 2 shown, the method includes: First, UAV image acquisition, including visible light images and multispectral images.

[0143] Then, image processing, including image denoising, radiometric correction, and geometric correction.

[0144] Furthermore, feature extraction is performed, including color feature extraction and spectral feature extraction.

[0145] Furthermore, model construction and training are performed based on comprehensive discriminant criteria. The detection model is a support vector machine model. The model training process includes: data division, model training, constraint conditions, objective function, function solution, and sample classification.

[0146] Furthermore, disease identification and annotation. Input the image of the soybean planting area into the trained model to classify the diseased results of the plants.

[0147] After that, diseased area clustering, morphological processing, and accurate annotation are performed.

[0148] Next, the soybean mosaic virus disease detection device provided by the present invention will be described. The soybean mosaic virus disease detection device described below can be mutually corresponding and referred to with the soybean mosaic virus disease detection method described above.

[0149] Figure 3 is the structural diagram of the soybean mosaic virus disease detection device based on UAV images provided by the present invention. As Figure 3As shown, the soybean mosaic disease detection device 300 based on drone images includes an acquisition module 310, a preprocessing module 320, a feature extraction module 330, and a classification and detection module 340; where: The acquisition module 310 is configured to collect visible light images and multispectral images of the soybean planting area through a drone equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude; the multispectral images include multiple bands; The preprocessing module 320 is configured to preprocess the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; The feature extraction module 330 is configured to extract color features from the preprocessed visible light images and extract spectral features from the preprocessed multispectral images; The classification and detection module 340 is configured to classify the soybean plants in the soybean planting area based on the color features and the spectral features by using an optimized support vector machine (SVM) model, and output a disease classification result of the soybean planting area; the disease classification result includes a potential diseased spot area or a healthy area; the objective function of the optimized support vector machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to represent the threshold range where the color features and the spectral features of healthy plants are located.

[0150] For the device provided in this embodiment, first, the acquisition module 310 is configured to collect visible light images and multispectral images of the soybean planting area through a drone equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude, and the multispectral images include multiple bands; then, the preprocessing module 320 is configured to preprocess the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; further, the feature extraction module 330 is configured to extract color features from the preprocessed visible light images and extract spectral features from the preprocessed multispectral images; furthermore, the classification and detection module 340 is configured to classify the soybean plants in the soybean planting area based on the color features and the spectral features by using an optimized support vector machine (SVM) model, and output a disease classification result of the soybean planting area; the disease classification result includes a potential diseased spot area or a healthy area, where the objective function of the optimized support vector machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to represent the threshold range where the color features and the spectral features of healthy plants are located.

[0151] In the present invention, first, drone images are acquired, including visible light images and multispectral images. The multispectral images include multiple bands. Then, through image preprocessing and feature extraction, color features and spectral features are obtained. Furthermore, based on the color features and spectral features, an optimized support vector machine (SVM) model is used for detection, and the soybean plants are classified to obtain the disease classification results. The visible light and multispectral images collected by the drone are comprehensively utilized, the color features and spectral features contained therein are deeply mined and analyzed, and then the optimized SVM model is used for detection, improving the accuracy of disease recognition and reducing the false positive rate.

[0152] According to a soybean mosaic disease detection device 300 based on drone images provided by the present invention, the acquisition module 310 is further configured to: Obtain the length and width of the soybean planting area; Based on the length and the width, determine the camera field of view angle and the flight altitude of the drone; Based on the camera field of view angle and the flight altitude, determine the adjacent flight line spacing of the preset flight line, so that the acquired images can completely cover the soybean planting area without obvious stitching gaps.

[0153] According to a soybean mosaic disease detection device 300 based on drone images provided by the present invention, the preprocessing module 320 is specifically configured to: Adopt a median filtering algorithm to denoise the visible light image and the multispectral image, obtaining the denoised visible light image and the denoised multispectral image; Based on ground control points, perform geometric correction on the denoised visible light image and the denoised multispectral image through an affine transformation model, obtaining the geometrically corrected visible light image and the geometrically corrected multispectral image; the ground control points are marked points with known coordinates preset in the soybean planting area; Based on the geometrically corrected visible light image and the geometrically corrected multispectral image, convert the image radiance value into the true surface reflectance according to the radiative transfer model, obtaining the preprocessed visible light image and the preprocessed multispectral image.

[0154] According to a soybean mosaic disease detection device 300 based on drone images provided by the present invention, the preprocessing module 320 is further configured to: For each pixel point in the visible light image and the multispectral image, determine the pixel values of all neighborhood windows of each pixel point; Determine the median value corresponding to the pixel values of all neighborhood windows of each pixel point; Based on each median value, determine the denoised pixel value corresponding to each pixel point; Based on the denoised pixel values corresponding to each of the pixel points, determine the denoised visible light image and the denoised multispectral image.

[0155] According to a soybean mosaic disease detection device 300 based on drone images provided by the present invention, the feature extraction module 330 is specifically configured to: Based on the preprocessed visible light image, determine the mean and variance of the red, green, and blue (RGB) channels of the soybean planting area; Determine the color features based on the mean and variance of the RGB channels; Based on the preprocessed multispectral image, determine the ratio vegetation index (RVI), normalized difference vegetation index (NDVI), and difference vegetation index (DVI) of the soybean planting area; Determine the spectral features based on the ratio vegetation index (RVI), normalized difference vegetation index (NDVI), and difference vegetation index (DVI);

[0156] According to a soybean mosaic disease detection device 300 based on drone images provided by the present invention, the classification and detection module 340 is specifically configured to: Based on the color features and the spectral features, determine a target feature vector; the target feature vector corresponds to multiple pixels; Input the target feature vector into the optimized support vector machine (SVM) model to obtain the target function values corresponding to multiple pixel points; For any one of the pixel points, if the target function value corresponding to the pixel point is greater than zero, determine the health status of the soybean plant corresponding to the pixel point as diseased; if the target function value corresponding to the pixel point is less than zero, determine the health status of the soybean plant corresponding to the pixel point as healthy.

[0157] According to a soybean mosaic disease detection method based on drone images provided by the present invention, the device further includes a training module; The training module is used for: Obtain a plurality of sample data; the sample data are labeled healthy samples and diseased samples; Based on the sample feature vectors and labels of the sample data, construct constraint conditions; the sample feature vectors are determined based on the color features of the sample data and the spectral features of the sample data; the constraint conditions are used to ensure that the samples are correctly classified and the margin meets specific requirements; Based on the sample feature vectors, labels, and constraint conditions of the sample data, solve the target function by the Lagrange multiplier method to obtain hyperplane parameters; Based on the hyperplane parameters, determine the optimized support vector machine (SVM) model.

[0158] A soybean mosaic disease detection device 300 based on UAV images provided by the present invention, the device further includes a post-processing module; The post-processing module is used for: Performing clustering analysis and morphological processing on the potential lesion areas in the disease-infected classification results to optimize the lesion boundaries and label the disease-infected areas; Generating a disease detection report based on the number of disease-infected plants and the proportion of disease-infected area in the disease-infected areas.

[0159] Figure 4 Illustrated is a schematic physical structure diagram of an electronic device, as Figure 4 shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440. Among them, the processor 410, the communication interface 420, and the memory 430 complete mutual communication through the communication bus 440. The processor 410 can call the logical instructions in the memory 430 to execute a soybean mosaic disease detection method based on UAV images. The method includes: Collecting visible light images and multispectral images of a soybean planting area by a UAV equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude; the multispectral images include multiple bands; Performing preprocessing on the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; Extracting color features from the preprocessed visible light images and extracting spectral features from the preprocessed multispectral images; Based on the color features and the spectral features, classifying the soybean plants in the soybean planting area by using an optimized support vector machine (SVM) model, and outputting a disease-infected classification result of the soybean planting area; the disease-infected classification result includes potential lesion areas or healthy areas; the objective function of the optimized support vector machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range where the color features and the spectral features of healthy plants are located.

[0160] In addition, when the logical instructions in the above-mentioned memory 430 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0161] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the soybean mosaic disease detection method based on drone images provided by the above-mentioned various methods. The method includes: Using a drone equipped with a high-definition optical camera and a multispectral camera to collect visible light images and multispectral images of the soybean planting area according to a preset flight path and flight altitude; the multispectral images include multiple bands; Preprocessing the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; Extracting color features from the preprocessed visible light images and extracting spectral features from the preprocessed multispectral images; Based on the color features and the spectral features, using an optimized support vector machine (SVM) model to classify the soybean plants in the soybean planting area and output the disease classification result of the soybean planting area; the disease classification result includes potential disease spot areas or healthy areas; the objective function of the optimized support vector machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range in which the color features and the spectral features of healthy plants are located.

[0162] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the soybean mosaic disease detection method based on drone images provided by the above-mentioned various methods. The method includes: The visible light images and multispectral images of the soybean planting area are collected by a drone equipped with a high-definition optical camera and a multispectral camera according to a preset flight route and flight altitude; the multispectral images include multiple bands; Preprocess the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; Extract color features from the preprocessed visible light images and extract spectral features from the preprocessed multispectral images; Based on the color features and the spectral features, use an optimized support vector machine (SVM) model to classify the soybean plants in the soybean planting area, and output the disease classification result of the soybean planting area; the disease classification result includes potential disease spot areas or healthy areas; the objective function of the optimized support vector machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range in which the color features and the spectral features of healthy plants are located.

[0163] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0164] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A detection method for soybean mosaic virus based on drone images, characterized in that, Including: Using a drone equipped with a high-definition optical camera and a multispectral camera to collect visible light images and multispectral images of the soybean planting area according to a preset flight path and flight altitude; the multispectral images include multiple bands; Preprocessing the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images; Extracting color features from the preprocessed visible light images and extracting spectral features from the preprocessed multispectral images; Based on the color features and the spectral features, using an optimized support vector machine (SVM) model to classify the soybean plants in the soybean planting area and output the disease classification results of the soybean planting area; The disease classification results include potential disease spot areas or healthy areas; the objective function of the optimized support vector machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range where the color features and the spectral features of healthy plants are located.

2. The soybean mosaic disease detection method based on UAV images according to claim 1, wherein, Before using the drone equipped with a high-definition optical camera and a multispectral camera to collect visible light images and multispectral images of the soybean planting area according to a preset flight path and flight altitude, it further includes: Obtaining the length and width of the soybean planting area; Based on the length and the width, determining the camera field of view angle and the flight altitude of the drone; Based on the camera field of view angle and the flight altitude, determining the adjacent flight path spacing of the preset flight path so that the acquired images can completely cover the soybean planting area without obvious stitching gaps.

3. The detection method of soybean mosaic virus based on UAV images according to claim 1, characterized in that, The preprocessing the visible light images and the multispectral images to obtain preprocessed visible light images and preprocessed multispectral images includes: Using a median filtering algorithm to denoise the visible light images and the multispectral images to obtain denoised visible light images and denoised multispectral images; Based on ground control points, geometrically correcting the denoised visible light images and denoised multispectral images through an affine transformation model to obtain geometrically corrected visible light images and geometrically corrected multispectral images; the ground control points are marked points with known coordinates preset in the soybean planting area; Based on the geometrically corrected visible light images and geometrically corrected multispectral images, converting the image radiance values into surface true reflectance according to a radiative transfer model to obtain the preprocessed visible light images and the preprocessed multispectral images.

4. The method for detecting soybean mosaic virus based on UAV images according to claim 3, characterized in that The using a median filtering algorithm to denoise the visible light images and the multispectral images to obtain denoised visible light images and denoised multispectral images includes: For each pixel point in the visible light images and the multispectral images, determining the pixel values of all neighborhood windows of each pixel point; Determining the median corresponding to the pixel values of all neighborhood windows of each pixel point; Based on each median, determining the denoised pixel value corresponding to each pixel point; Based on the denoised pixel values corresponding to each pixel point, determining the denoised visible light images and denoised multispectral images.

5. The soybean mosaic virus detection method based on drone images according to claim 1, wherein Extracting color features from the preprocessed visible light image and spectral features from the preprocessed multispectral image includes: Based on the preprocessed visible light image, determining the mean and variance of the red, green, and blue (RGB) channels of the soybean planting area; Determining the RGB channel mean and the variance as the color features; Based on the preprocessed multispectral image, determining the ratio vegetation index (RVI), normalized difference vegetation index (NDVI), and difference vegetation index (DVI) of the soybean planting area; Determining the RVI, NDVI, and DVI as the spectral features.

6. The soybean mosaic disease detection method based on drone images according to claim 1, characterized in that Classifying soybean plants in the soybean planting area using an optimized support vector machine (SVM) model based on the color features and the spectral features, and outputting the disease classification result of the soybean planting area, including: Based on the color features and the spectral features, determining a target feature vector; the target feature vector corresponds to multiple pixels; Inputting the target feature vector into the optimized SVM model to obtain the objective function values corresponding to multiple pixel points; For any one of the pixel points, if the objective function value corresponding to the pixel point is greater than zero, determining the health status of the soybean plant corresponding to the pixel point as diseased; if the objective function value corresponding to the pixel point is less than zero, determining the health status of the soybean plant corresponding to the pixel point as healthy.

7. The method for detecting soybean mosaic virus based on UAV images according to claim 1, characterized in that, The optimized SVM model is constructed through the following steps, including: Obtaining a plurality of sample data; the sample data are labeled healthy samples and diseased samples; Based on the sample feature vectors and labels of the sample data, constructing constraint conditions; the sample feature vectors are determined based on the color features and spectral features of the sample data; the constraint conditions are used to ensure that the samples are correctly classified and the margin meets specific requirements; Based on the sample feature vectors, labels, and the constraint conditions of the sample data, solving the objective function by the Lagrange multiplier method to obtain hyperplane parameters; Based on the hyperplane parameters, determining the optimized SVM model.

8. The detection method of soybean mosaic virus based on UAV images according to any one of claims 1-7, characterized in that, The method further includes: Performing clustering analysis and morphological processing on the potential diseased area in the disease classification result to optimize the diseased area boundary and label the diseased area; Generating a disease detection report based on the number of diseased plants and the proportion of diseased area in the diseased area.

9. A detection device for soybean mosaic virus based on UAV images, characterized in that, Including: An acquisition module, configured to collect visible light images and multispectral images of a soybean planting area by a drone equipped with a high-definition optical camera and a multispectral camera according to a preset flight path and flight altitude; the multispectral image includes multiple bands; A preprocessing module, configured to preprocess the visible light image and the multispectral image to obtain a preprocessed visible light image and a preprocessed multispectral image; A feature extraction module, configured to extract color features from the preprocessed visible light image and spectral features from the preprocessed multispectral image; A classification detection module, which is used to classify soybean plants in the soybean planting area based on the color feature and the spectral feature by using an optimized support vector machine (SVM) model, and output the disease classification result of the soybean planting area; The disease classification result includes a potential disease spot area or a healthy area; the objective function of the optimized support vector machine (SVM) model is determined based on a comprehensive discrimination criterion, and the comprehensive discrimination criterion is used to characterize the threshold range in which the color feature and the spectral feature of healthy plants are located.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, When the processor executes the computer program, it implements the method for detecting soybean mosaic disease based on UAV images according to any one of claims 1 to 8.

Images