Wheat stripe rust multi-scale detection system and method based on unmanned aerial vehicle multi-spectral image

Abstract

The invention relates to the technical field of intelligent agricultural remote sensing monitoring, in particular to a wheat stripe rust multi-scale detection system and method based on unmanned aerial vehicle multi-spectral images. Comprising an attitude compensation image registration module, a sheath and leaf boundary enhancement module, a stripe rust incubation period stripe extraction module, a canopy stomatal density inversion and mapping module and a three-dimensional space infection simulation module. Wherein the attitude compensation image registration module is used for eliminating flight attitude interference; the leaf sheath and leaf junction enhancing module is used for enhancing the structural boundary characteristics of a leaf sheath and leaf junction area; the stripe rust incubation period stripe extraction module is used for extracting disease incubation period stripe features; and the canopy stomatal density inversion and mapping module is used for inverting canopy stomatal conductance parameters. According to the method, by fusing multispectral image feature extraction, thermal infrared stomatal conductance inversion and three-dimensional space propagation modeling, multi-scale accurate detection and visual expression of the wheat stripe rust from incubation period identification to a spatial diffusion process are realized.

Description

Technical Field

[0001] The present invention relates to the field of smart agricultural remote sensing monitoring technology, and in particular to a multi-scale detection system and method for wheat stripe rust based on unmanned aerial vehicle multispectral images. Background Art

[0002] Wheat stripe rust is a highly contagious fungal disease that poses a serious threat to global wheat production safety. It has the characteristics of a long incubation period, rapid spread, and strong spatial heterogeneity. In the early stages of the disease, stripe rust hyphae have begun to invade the tissue along the leaf sheath to the leaf blade, but because the lesions have not yet appeared, it is difficult to effectively identify them with the naked eye or traditional imaging methods. In recent years, unmanned aerial vehicle multispectral remote sensing technology has gradually been applied to the field of agricultural disease monitoring. By obtaining multi-band reflectance characteristics, thermal infrared information and spatial structure data from the crop surface, it is expected to achieve non-contact observation and early identification of physiological changes of the disease. However, due to the existence of posture disturbances in drone images, complex plant structure, and uncertain disease expansion paths, the stability of multi-scale disease feature extraction is poor, spatial positioning is inaccurate, and the spread trend is difficult to quantify.

[0003] Existing disease monitoring methods are mostly limited to static recognition based on two-dimensional images. They lack modeling that extends disease morphological characteristics to physiological indicators. This is particularly true during the incubation period, making it difficult to effectively quantify disease risk using indicators such as stomatal conductance and transpiration response. Furthermore, they are unable to simulate disease transmission processes incorporating three-dimensional spatial structures, hindering the development of refined disease prevention and control strategies. Therefore, there is an urgent need to develop a multi-scale detection system and method for wheat stripe rust based on drone multispectral imagery to provide scientific support for precise stripe rust monitoring and early intervention. Summary of the Invention

[0004] Based on the above objectives, the present invention provides a multi-scale detection system and method for wheat stripe rust based on UAV multispectral imagery.

[0005] A multi-scale wheat stripe rust detection system based on UAV multispectral imagery includes an attitude-compensated image registration module, a leaf sheath-leaf boundary enhancement module, a stripe rust latent period pattern extraction module, a canopy stomatal density inversion and mapping module, and a three-dimensional infection simulation module. Attitude Compensation Image Registration Module: This module receives the original multispectral image data collected by the drone, fuses the attitude information output by the drone's inertial measurement unit with a sub-pixel image registration algorithm based on phase correlation, and generates a registered image set that eliminates flight attitude interference. The leaf sheath-leaf boundary enhancement module is connected to the attitude compensation image registration module, which receives the registered image set, fuses the image gradient information of the yellow edge band and the red edge band, enhances the structural boundary features of the leaf sheath-leaf boundary area, and outputs an enhanced image. Stripe rust latent period pattern extraction module: This module is connected to the leaf sheath-leaf boundary enhancement module to receive the enhanced image, construct a directional neighborhood contrast enhancement operator based on the potential expansion path of stripe rust hyphae, extract the latent period pattern features of the disease, and output the latent period stripe rust feature map; Canopy Stomatal Density Inversion and Mapping Module: This module is connected to the Stripe Rust Incubation Period Mark Extraction Module and is used to receive the stripe rust characteristic map during the incubation period. It inverts the canopy stomatal conductance parameters based on the thermal infrared band image, overlays the generated stomatal open density spatial distribution map, and outputs the disease development probability map. Three-dimensional infection simulation module: connected to the canopy stomatal density inversion and mapping module, used to receive the disease development probability map, and combined with the three-dimensional point cloud data of wheat plants, to construct a three-dimensional disease distribution map including vertical height distribution, infection gradient evolution and horizontal transmission trend.

[0006] Optionally, the posture compensation image registration module includes a posture data solving unit, a posture disturbance pre-compensation unit and a sub-pixel image registration unit; wherein: Attitude data calculation unit: used to receive the raw data of acceleration, angular velocity and magnetic field intensity output by the UAV inertial measurement unit, and calculate the three-dimensional attitude angle at the current image capture time based on the extended Kalman filter method, including pitch angle, roll angle and yaw angle; Attitude disturbance pre-compensation unit: used to construct the attitude disturbance compensation matrix of the multispectral image in three-dimensional space based on the three-dimensional attitude angle parameters output by the attitude data solution unit, and perform rotation and projection transformation on the original image coordinates to generate a coarse registration image sequence; Sub-pixel image registration unit: It is used to refine the coarse registration image sequence output by the attitude disturbance pre-compensation unit. It uses a sub-pixel displacement estimation algorithm based on the frequency domain phase correlation method to calculate the sub-pixel translation vector between images and perform image resampling, thereby outputting a set of registration images that eliminates flight attitude interference.

[0007] Optionally, the sub-pixel image registration unit includes: Reference image selection subunit: used to select a frame with the best signal-to-noise ratio from the coarse registration image sequence as the global registration reference image, and use it as the comparison benchmark frame for subsequent registration operations; Frequency domain conversion subunit: used to perform two-dimensional Fourier transform on the reference image and the image to be registered, and the transformation result is and ,in is the frequency domain coordinate; Phase difference calculation subunit: used to calculate the phase difference map of the normalized cross power spectrum ; Sub-pixel displacement estimation subunit: Phase difference map Perform a two-dimensional inverse Fourier transform to obtain the cross-correlation function graph, and use quadratic interpolation at its peak position to calculate the sub-pixel displacement vector ; Image resampling subunit: Based on the estimated , a sub-pixel translation transformation is performed on the registration image, and the bicubic interpolation method is used to complete the image resampling, and the final output is a set of registration images that eliminates the interference of flight posture.

[0008] Optionally, the leaf sheath-leaf boundary enhancement module includes a band image extraction unit, an edge gradient calculation unit, a structure boundary fusion unit, and an image enhancement output unit; wherein: Band image extraction unit: used to extract yellow edge band image and red edge band image from the registration image set output by the attitude compensation image registration module; Edge gradient calculation unit: used to calculate the corresponding gradient amplitude of the yellow edge band and red edge band images respectively; Structural boundary fusion unit: uses a weighted fusion strategy to fuse the gradient maps of the two bands and construct a fused boundary map , whose expression is: ,in, and The gradient amplitudes of the yellow edge band and the red edge band are respectively: and are the fusion weight parameters of the yellow edge and red edge bands respectively; Image enhancement output unit: based on fusion boundary map The original image is enhanced in the gradient domain to highlight the edge response of the boundary between the leaf sheath and the leaf blade, and the enhanced image is output for subsequent stripe feature extraction.

[0009] Optionally, the stripe rust latent period pattern extraction module includes a direction hypothesis generation unit, a directional neighborhood construction unit, a contrast enhancement calculation unit, and a feature map output unit; wherein: Direction hypothesis generation unit: This unit is used to set multiple sets of preset direction angles based on the direction of the main veins of wheat leaves and their consistency with the typical propagation direction of stripe rust hyphae. This is used as the orientation basis for potential expansion paths, forming a direction set list for subsequent stripe detection. Directional neighborhood construction unit: Based on each direction angle provided by the direction hypothesis generation unit, a directional trapezoidal neighborhood window with the direction as the main axis is constructed around each pixel point in the image; Contrast enhancement calculation unit: used to calculate the brightness contrast difference between the current pixel and its neighboring background within a directional trapezoidal neighborhood window. It uses the principle of directional consistency to enhance the response intensity of latent lesions, highlight the grayscale feature differences in the latent stripe rust area, and eliminate noise interference not along the extension direction. Characteristic map output unit: used to perform maximum fusion on the enhanced response maps in all directions, extract the latent period stripe area with the strongest response in multiple directions, and generate the latent period stripe rust characteristic map.

[0010] Optionally, the contrast enhancement calculation unit includes: Local mean brightness calculation subunit: used to calculate the local mean brightness of the current pixel in the directional trapezoidal neighborhood window. As the center, extract the direction The neighborhood pixel set under , and calculate the average brightness value of the neighborhood ; Directional consistency score calculation subunit: used to extract the current pixel point The image gradient direction at the position and the preset direction angle Perform similarity calculation to obtain direction consistency score , the formula is: ,in, Indicates the image gradient direction of the current pixel; Enhanced response generation subunit: used to calculate the direction of the current pixel Enhanced response value under , the formula is: ,in, Indicates the grayscale value of the current pixel.

[0011] Optionally, the canopy stomatal density inversion and mapping module includes a temperature map extraction unit, a stomatal conductance inversion unit, a density map generation unit, and a disease probability output unit; wherein: Temperature map extraction unit: used to extract thermal infrared band images from the registered image set, perform radiation calibration, background thermal field correction and noise filtering on the images, and output a temperature map representing the thermal distribution of the crop surface; Stomatal conductance inversion unit: This unit receives temperature maps and latent stripe rust characteristic maps, models crop surface heat flux, and constructs an analytical model based on environmental meteorological parameters to invert leaf stomatal conductance values, thereby representing changes in transpiration levels and water vapor exchange capacity. Density map generation unit: used to calculate the stomatal open density value after obtaining the stomatal conductance value of the leaf surface, combined with the distribution characteristics of the latent period spots, and construct a spatial distribution map to form a stomatal open density map; Disease probability output unit: used to classify and evaluate the density level of each area based on the stomatal opening density map and the prior range of stomatal behavior of healthy leaves, and output the disease development probability map accordingly.

[0012] Optionally, the disease probability output unit includes: Health behavior modeling subunit: used to construct the normal range of stomatal opening density of healthy wheat leaves, including the minimum reference value With the maximum reference value , as a priori benchmark for subsequent classification evaluation; Density classification judgment subunit: used to receive the density value of each pixel in the stomata open density map , and based on its and and make regional classification judgments based on the relationship between them; like , determined to be abnormally closed type, indicating a potential stomatal inactivation area; like , it is judged as abnormally open type, indicating a transpiration disorder area; like , determined as the normal stomatal behavior area; Probability map generation subunit: used to assign disease probability values to each pixel in the image based on the classification judgment results The disease probability value of each pixel is spliced according to the original image coordinates to generate a complete disease development probability map. The expression is: .

[0013] Optionally, the three-dimensional space infection simulation module includes a point cloud structure reconstruction unit, a disease probability mapping unit, an infection and spread modeling unit, and a three-dimensional distribution map generation unit; wherein: Point cloud structure reconstruction unit: used to receive multi-angle aerial images or laser scanning data, reconstruct a three-dimensional point cloud model of the wheat canopy, and assign spatial coordinates to each point to form a spatial skeleton with plant structure levels; Disease probability mapping unit: used to spatially register the two-dimensional probability values in the disease development probability map with the positions of the corresponding canopy projection points in the three-dimensional point cloud according to the image pixel coordinates, and map the disease probability value of each two-dimensional pixel to the corresponding point cloud node; Infection and Diffusion Modeling Unit: Based on the mapped disease probability points, combined with the spatial adjacency and vertical hierarchical distribution information of each point in the 3D point cloud, it constructs disease propagation weight functions in the vertical and horizontal directions, and iteratively expands and simulates the spread trend of the disease from high-risk points to adjacent structures, while recording its diffusion path and probability gradient changes; Three-dimensional distribution map generation unit: Based on the infection modeling results, the disease risk level of each point in the three-dimensional space is output, and a three-dimensional disease distribution map is formed according to the spatial coordinates.

[0014] A multi-scale detection method for wheat stripe rust based on drone multispectral imagery is implemented by the aforementioned multi-scale detection system for wheat stripe rust based on drone multispectral imagery, comprising the following steps: S1: Use UAV to acquire multispectral image data, and combine it with the attitude information output by the inertial measurement unit to generate a registered image set that removes flight attitude errors through a phase-correlation sub-pixel registration algorithm; S2: Based on the registered image set, the gradient information of the yellow edge band and the red edge band is extracted, and the edge area of the junction between the leaf sheath and the leaf is weighted fused to output the junction enhanced image; S3: Directed neighborhood contrast enhancement is applied to the boundary-enhanced image, and an enhancement operator is constructed based on the potential diffusion direction of stripe rust hyphae to identify the latent period marking area and generate a stripe rust feature map. S4: Based on the characteristic map of stripe rust during the latent period, the stomatal conductance value of the leaf surface is inverted by combining thermal infrared band images and environmental meteorological parameters. At the same time, the stomatal opening level at each spatial position is quantitatively mapped to form a stomatal opening density map and output a disease development probability map; S5: Align the disease development probability map with the 3D point cloud data of wheat plants. Based on the disease probability and spatial adjacency of each point, construct vertical and horizontal propagation weight functions and perform iterative modeling to generate a 3D disease distribution map.

[0015] Beneficial effects of the present invention: This method combines multispectral image registration, sheath-leaf boundary enhancement, and latent stripe pattern extraction with thermal infrared stomatal conductance inversion to achieve linked detection of disease pathological characteristics during the latent period and crop physiological indicators. This overcomes the limitations of traditional two-dimensional image recognition, which struggles to quantify early disease risk. Multi-band gradient fusion and directional neighborhood contrast enhancement accurately locate latent stripe rust patterns and, combined with spatial distribution maps of stomatal density, effectively quantify the physiological stress response caused by the disease.

[0016] This invention uses a three-dimensional point cloud model to iterate the stripe rust transmission process, constructs vertical and horizontal transmission weight functions, and realizes dynamic simulation and risk prediction of the disease at the spatial level. By seamlessly integrating pathological texture information and physiological transpiration parameters, it not only improves the accuracy and robustness of disease monitoring, but also provides visual spatial support for wheat stripe rust prevention and control strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 Schematic diagram of a multi-scale detection system for wheat stripe rust according to an embodiment of the present invention; Figure 2 Schematic diagram of a multi-scale detection method for wheat stripe rust according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0020] It should be noted that references in the specification to "one embodiment," "an embodiment," "exemplary embodiments," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not necessarily every embodiment will include such specific features, structures, or characteristics. Furthermore, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0021] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0022] like Figure 1 As shown in the figure, a multi-scale detection system for wheat stripe rust based on UAV multispectral imagery includes an attitude-compensated image registration module, a leaf sheath-leaf boundary enhancement module, a stripe rust latent period pattern extraction module, a canopy stomatal density inversion and mapping module, and a three-dimensional space infection simulation module; wherein: Attitude Compensation Image Registration Module: This module receives the original multispectral image data collected by the drone, fuses the attitude information output by the drone's Inertial Measurement Unit (IMU) with a sub-pixel image registration algorithm based on phase correlation, and generates a registered image set that eliminates flight attitude interference. The leaf sheath-leaf boundary enhancement module is connected to the attitude compensation image registration module, which receives the registered image set, fuses the image gradient information of the yellow edge band and the red edge band, enhances the structural boundary features of the leaf sheath-leaf boundary area, and outputs an enhanced image. Stripe rust latent period pattern extraction module: This module is connected to the leaf sheath-leaf boundary enhancement module to receive the enhanced image, construct a directional neighborhood contrast enhancement operator based on the potential expansion path of stripe rust hyphae, extract the latent period pattern features of the disease, and output the latent period stripe rust feature map; Canopy Stomatal Density Inversion and Mapping Module: This module is connected to the Stripe Rust Incubation Period Mark Extraction Module and is used to receive the stripe rust characteristic map during the incubation period. It inverts the canopy stomatal conductance parameters based on the thermal infrared band image, overlays the generated stomatal open density spatial distribution map, and outputs the disease development probability map. Three-dimensional spatial infection simulation module: connected to the canopy stomatal density inversion and mapping module, used to receive the disease development probability map, and combined with the three-dimensional point cloud data of wheat plants, construct a three-dimensional disease distribution map including vertical height distribution, infection gradient evolution and horizontal transmission trend, to realize the simulation and prediction of wheat stripe rust in the spatial domain.

[0023] The attitude compensation image registration module includes an attitude data solution unit, an attitude disturbance pre-compensation unit and a sub-pixel image registration unit; wherein: Attitude data calculation unit: used to receive the raw data of acceleration, angular velocity and magnetic field intensity output by the UAV inertial measurement unit, and calculate the three-dimensional attitude angle at the current image capture time based on the extended Kalman filter method, including pitch angle, roll angle and yaw angle; Attitude disturbance pre-compensation unit: used to construct the attitude disturbance compensation matrix of the multispectral image in three-dimensional space based on the three-dimensional attitude angle parameters output by the attitude data solution unit, and perform rotation and projection transformation on the original image coordinates to generate a coarse registration image sequence; The specific steps are as follows: Attitude angle parameter extraction: call the three-axis attitude angle data output by the attitude data solution unit, which are recorded as pitch angle , roll angle and yaw angle ; Rotation matrix construction: based on pitch angle , roll angle and yaw angle , and construct the two-dimensional Euler angle rotation matrix in turn , calculate the overall rotation matrix , where the expression is: ,in, They are the rotation matrices around the X, Y, and Z axes respectively; all angle values are entered in radians; Image coordinate rotation: rotate the original image coordinate point Defined as a vector in three-dimensional space based on the imaging plane and applying the compensation matrix implement: , and then get the coordinates of the image point after rotation ; Projection transformation and image resampling: rotate the image coordinates Map it back to the imaging plane through normalized projection transformation to obtain the corresponding coordinate points on the two-dimensional image plane , and bilinear interpolation is used to resample the image to generate the target image frame in the coarse registration image sequence.

[0024] Sub-pixel image registration unit: used to refine the coarse registration image sequence output by the attitude disturbance pre-compensation unit, using a sub-pixel displacement estimation algorithm based on the frequency domain phase correlation method to calculate the sub-pixel translation vector between images and perform image resampling, thereby outputting a set of registered images that eliminate flight attitude interference; through the above unit, a staged correction process for the UAV's flight attitude interference is achieved, which not only ensures the geometric consistency of the registration image, but also improves the spatial alignment accuracy of the subsequent processing of the multispectral image, providing a high-quality image foundation for disease detection tasks.

[0025] The sub-pixel image registration unit includes: Reference image selection subunit: used to select a frame with the best signal-to-noise ratio from the coarse registration image sequence as the global registration reference image, and use it as the comparison benchmark frame for subsequent registration operations; Frequency domain conversion subunit: used to perform two-dimensional Fourier transform on the reference image and the image to be registered, and the transformation result is and ,in is the frequency domain coordinate; Phase difference calculation subunit: used to calculate the phase difference map of the normalized cross power spectrum , the formula is: ,in for The complex conjugate of Sub-pixel displacement estimation subunit: Phase difference map Perform a two-dimensional inverse Fourier transform to obtain the cross-correlation function graph, and use quadratic interpolation at its peak position to calculate the sub-pixel displacement vector ; Image resampling subunit: Based on the estimated , performs sub-pixel translation transformation on the image to be registered, and uses bicubic interpolation to complete image resampling, and finally outputs a registered image set that eliminates flight attitude interference; the above sub-units construct a sub-pixel displacement estimation process based on frequency domain phase difference, combined with the fine peak positioning mechanism of cross-power spectrum phase normalization and quadratic interpolation, to achieve high-precision translation vector calculation and registration between multispectral images, effectively compensate for the micro-displacement error caused by flight attitude changes, and improve the spatial consistency and analysis reliability of the final registered image set.

[0026] The leaf sheath-leaf boundary enhancement module includes a band image extraction unit, an edge gradient calculation unit, a structure boundary fusion unit, and an image enhancement output unit; wherein: Band image extraction unit: used to extract yellow edge band image and red edge band image from the registered image set output by the attitude compensation image registration module for subsequent edge feature analysis; Edge gradient calculation unit: used to calculate the corresponding gradient amplitude of the yellow edge band and red edge band images respectively. The gradient calculation formula is: ,in, Indicates the corresponding band image, is the direction of the image coordinate axis; Structural boundary fusion unit: uses a weighted fusion strategy to fuse the gradient maps of the two bands and construct a fused boundary map , whose expression is: ,in, and The gradient amplitudes of the yellow edge band and the red edge band are respectively: and are the fusion weight parameters of the yellow edge and red edge bands, respectively, satisfying ; Image enhancement output unit: based on fusion boundary map The original image is subjected to weighted enhancement processing in the gradient domain to highlight the edge response of the boundary area between the leaf sheath and the leaf blade, and the enhanced image is output for subsequent pattern feature extraction; the above-mentioned unit extracts the yellow-edge and red-edge band images and calculates the gradient information respectively, and combines the weighted fusion algorithm to highlight the edge strength at the junction, which can significantly enhance the structural characteristics of the boundary area between the leaf sheath and the leaf blade, effectively improving the image resolution basis for latent period pattern identification and the visibility of the disease expansion path.

[0027] The stripe rust latent period pattern extraction module includes a direction hypothesis generation unit, a directional neighborhood construction unit, a contrast enhancement calculation unit, and a feature map output unit; wherein: Direction hypothesis generation unit: This unit is used to set multiple sets of preset direction angles based on the direction of the main veins of wheat leaves and their consistency with the typical propagation direction of stripe rust hyphae. This is used as the orientation basis for potential expansion paths, forming a direction set list for subsequent stripe detection. Directional neighborhood construction unit: Based on each direction angle provided by the direction hypothesis generation unit, a directional trapezoidal neighborhood window with that direction as the main axis is constructed around each pixel in the image, ensuring that the enhancement operation focuses on the possible path of hyphae expansion and avoiding horizontal texture interference; The steps to construct a directional trapezoidal neighborhood window are as follows: Direction vector definition: Let the current direction angle be , construct the principal axis direction vector expressed in the form of a unit vector , this vector is used to determine the extension direction of the main axis of the neighborhood; Neighborhood window parameter setting: Define the length of the directional neighborhood window as , with a width of , the center of the window is the current processing pixel , the main axis direction is along Extension; the width of the trapezoidal window on both sides is set symmetrically , extending front and back length; Neighborhood boundary point calculation: based on direction vector and its normal vector , construct the neighborhood vertex coordinate set : The center of the front vertex is: ; The center of the back vertex is: ; The four sides of the trapezoid are: ; ; ; The above four points form a direction angle The trapezoidal neighborhood window of the principal axis has a geometric outline that accurately covers the potential expansion channels of hyphae under the current direction.

[0028] Contrast enhancement calculation unit: used to calculate the brightness contrast difference between the current pixel and its neighboring background within a directional trapezoidal neighborhood window. It uses the principle of directional consistency to enhance the response intensity of latent lesions, highlight the grayscale feature differences in the latent stripe rust area, and eliminate noise interference not along the extension direction. Feature map output unit: used to maximize the enhanced response maps in all directions, extract the latent period stripe area with the strongest response in multiple directions, generate a latent period stripe rust feature map with spatial continuity and directional specificity, and output the map to the downstream module for disease probability analysis; by setting a multi-directional hypothesis based on leaf vein structure and constructing a directional neighborhood contrast enhancement mechanism, the extraction of latent period stripe rust patterns is more consistent with its actual expansion behavior, thereby significantly improving the detection ability of hidden disease areas in the image, and providing a stable map input basis for stomatal density inversion and disease risk reasoning.

[0029] The contrast enhancement calculation unit includes: Local mean brightness calculation subunit: used to calculate the local mean brightness of the current pixel in the directional trapezoidal neighborhood window. As the center, extract the direction The neighborhood pixel set under , and calculate the average brightness value of the neighborhood , the formula is: ,in, Indicates that the image is at the coordinate point Gray value at ; Represents the total number of pixels in the neighborhood; Indicates the direction angle The trapezoidal neighborhood area constructed below; Directional consistency score calculation subunit: used to extract the current pixel point The image gradient direction at the position and the preset direction angle Perform similarity calculation to obtain direction consistency score , the formula is: ,in, Indicates the image gradient direction of the current pixel; is the direction angle; Enhanced response generation subunit: used to calculate the direction of the current pixel Enhanced response value under The response value is determined by the difference between the current pixel brightness and the neighborhood average brightness and the direction consistency score. The formula is: ,in, Represents the grayscale value of the current pixel; the above subunit can effectively improve the response intensity of the latent period streak area by accurately calculating the brightness difference between the current pixel and its directional neighborhood and fusing the consistency score of the image gradient direction and the potential expansion direction of the lesion.

[0030] The specific steps of the maximum fusion method in the feature map output unit are as follows: Directional response diagram collection: The set direction angle collection is ,in is the number of directions; for each direction angle, a direction response map is output by the contrast enhancement calculation subunit, which is recorded as ,in is the image pixel coordinate, For the The enhanced response value of the pixel in each direction; Pixel-level maximum fusion: In all direction response maps, for the same pixel coordinate The maximum value operation is performed on the response value at , and the response map value in the final feature map is generated. The formula is: ,in, Indicates the fusion of the latent period stripe rust characteristic map in pixels The response strength at Indicates the The value of the direction response map at the pixel; High response area extraction: setting the response intensity threshold ,Will The pixel points are marked as the latent period stripe area, forming a binary mask image, which is then combined with the original registered image to output the final latent period stripe rust feature map.

[0031] The canopy stomatal density inversion and mapping module includes a temperature map extraction unit, a stomatal conductance inversion unit, a density map generation unit, and a disease probability output unit; among which: Temperature map extraction unit: used to extract thermal infrared band images from the registered image set, perform radiometric calibration, background thermal field correction, and noise filtering on the images, and output a temperature map representing the thermal distribution of the crop surface, which is used as the input basis for stomatal conductance inversion. Stomatal conductance inversion unit: This unit receives temperature maps and latent stripe rust characteristic maps, models crop surface heat flux, and constructs an analytical model based on environmental meteorological parameters to invert leaf stomatal conductance values. This represents changes in transpiration levels and water vapor exchange capacity, indirectly reflecting the physiological stress caused by the disease. The specific steps for inverting the stomatal conductance value of the leaf surface are as follows: Input data standardization: The temperature map is standardized and radiometrically converted to ensure that each pixel value in the image represents the actual temperature of the crop surface. The measured meteorological parameters of the external environment, including atmospheric temperature, wind speed, air pressure, relative humidity, and net radiation flux, are also received simultaneously. Establish an energy balance equation: Taking each pixel as a unit, construct the energy conservation equation at the leaf scale: ,in, is the net radiation flux of the leaf surface; is the latent heat flux, which represents the energy lost by water vapor due to transpiration; is the sensible heat flux, which represents the heat conduction caused by the air temperature difference; is the leaf surface heat conduction; Functional relationship conversion between latent heat flux and stomatal conductance: By calculating the transpiration rate, the latent heat flux is converted into stomatal conductance. Specifically, the latent heat flux expression is: ; Then solve the stomatal conductance value through the following conversion formula , the formula is: ,in, is the air density; is the specific heat capacity at constant pressure; is the humidity constant; is the saturated water vapor pressure-temperature slope; is the crop leaf surface temperature at the pixel; is the air resistance; is the pore resistance, which is the target to be inverted.

[0032] Density map generation unit: used to calculate the stomatal open density value after obtaining the stomatal conductance value of the leaf surface, combined with the distribution characteristics of the latent period spots, and construct a spatial distribution map to form a stomatal open density map; The specific steps to form the stomatal open density map are as follows: Stomatal conductance map normalization: The two-dimensional conductance map output by the stomatal conductance inversion subunit is normalized so that its value range is unified to the interval [0,1]. The normalized conductance value of each pixel represents the relative strength of the leaf transpiration capacity of that pixel in the entire map. Extraction of stripe pattern response weights: Extract the response intensity value of each pixel from the latent stripe rust characteristic map and set a high response weight coefficient for all non-zero response areas , set low weights for zero response areas ; Density value calculation: For each pixel point, multiply its normalized conductance value by the corresponding weight value to generate the stomatal open density value; Spatial distribution map construction: The stomatal openness density value is mapped to a two-dimensional spatial coordinate system, and the original image pixel positions are filled in sequentially to generate a complete stomatal openness density map.

[0033] Disease probability output unit: used to classify and evaluate the density levels of each area based on the stomatal opening density map and the prior range of stomatal behavior of healthy leaves, and output a disease development probability map accordingly, quantifying the possibility of disease aggravation at each location, and passing it to the downstream three-dimensional simulation module; the above-mentioned unit combines thermal infrared image information with the disease map, inverts stomatal conductance using physical modeling, regenerates a spatialized stomatal opening density map and converts it into a disease probability map, realizing the fusion analysis of disease morphology and physiological indicators, and improving the quantitative and precision control capabilities of spatial disease modeling.

[0034] The disease probability output unit includes: Health behavior modeling subunit: used to construct the normal range of stomatal opening density of healthy wheat leaves, including the minimum reference value With the maximum reference value , as a priori benchmark for subsequent classification evaluation; Density classification judgment subunit: used to receive the density value of each pixel in the stomata open density map , and based on its and and make regional classification judgments based on the relationship between them; like , determined to be abnormally closed type, indicating a potential stomatal inactivation area; like , it is judged as abnormally open type, indicating a transpiration disorder area; like , determined as the normal stomatal behavior area; Probability map generation subunit: used to assign disease probability values to each pixel in the image based on the classification judgment results The disease probability value of each pixel is spliced according to the original image coordinates to generate a complete disease development probability map. The expression is: ; The above-mentioned subunits establish a priori model of stomatal density of healthy crops and perform quantitative classification and judgment on the current density map, which can realize the graded assessment of disease probability based on differences in physiological behavior. The generated disease probability map can quantitatively reflect the spatial transmission risk of the disease and provide a reliable risk base layer for three-dimensional simulation.

[0035] The three-dimensional space infection simulation module includes a point cloud structure reconstruction unit, a disease probability mapping unit, an infection diffusion modeling unit, and a three-dimensional distribution map generation unit; among which: Point cloud structure reconstruction unit: used to receive multi-angle aerial images or laser scanning data, reconstruct a three-dimensional point cloud model of the wheat canopy, and assign spatial coordinates to each point to form a spatial skeleton with plant structure levels, which is used to carry out subsequent mapping calculations of disease probability information; Disease probability mapping unit: used to spatially register the two-dimensional probability values in the disease development probability map with the positions of the corresponding canopy projection points in the three-dimensional point cloud according to the image pixel coordinates, map the disease probability value of each two-dimensional pixel to the corresponding point cloud node, and establish a probability-structure binding relationship; Infection and Diffusion Modeling Unit: Based on the mapped disease probability points, combined with the spatial adjacency and vertical hierarchical distribution information of each point in the 3D point cloud, it constructs disease propagation weight functions in the vertical and horizontal directions, and iteratively expands and simulates the spread trend of the disease from high-risk points to adjacent structures, while recording its diffusion path and probability gradient changes; The vertical propagation weight function expression is: ; Horizontal propagation weight function: ;in, Indicates a point The probability value of stripe rust ranges from 0 to 1, and the larger the value, the higher the disease risk; Represent points with dot The vertical coordinates in the 3D point cloud are expressed in dimensionless form after normalization so that the multiplication operation maintains the same dimension; Indicates a point with dot The distance in the horizontal plane is also normalized to ensure dimensional consistency in subsequent index calculations; denote the scale coefficients of vertical and horizontal propagation respectively; Respectively represent the attenuation coefficients of vertical and horizontal transmission, which are used to control the attenuation rate of disease transmission as the distance increases. The larger the value, the faster the transmission weight decreases as the distance increases. It is the vertical transmission weight value. The larger the value, the more points Point in the vertical direction The stronger the impact of disease spread; It is the horizontal transmission weight value, dimensionless, and the larger the value, the more points Point in the horizontal direction The greater the impact of disease spread.

[0036] Three-dimensional distribution map generation unit: Based on the results of infection modeling, the disease risk level of each point in three-dimensional space is output, and a three-dimensional disease distribution map is reconstructed according to the spatial coordinates. The layered infection situation, disease density gradient and horizontal propagation direction in the height dimension are marked as a visual output for disease monitoring and spatial prevention and control. The above unit combines the disease probability map with the three-dimensional canopy structure of wheat to construct an infection model with spatial topology and propagation dynamics, and realizes the dynamic evolution simulation of stripe rust from the high-risk area during the latent period to the entire plant structure, providing a structured and hierarchical risk distribution basis for precise intervention and three-dimensional prevention and control.

[0037] like Figure 2 As shown, a multi-scale detection method for wheat stripe rust based on drone multispectral images is implemented by the above-mentioned multi-scale detection system for wheat stripe rust based on drone multispectral images, comprising the following steps: S1: Use UAV to acquire multispectral image data, and combine it with the attitude information output by the inertial measurement unit to generate a registered image set that removes flight attitude errors through a phase-correlation sub-pixel registration algorithm; S2: Based on the registered image set, the gradient information of the yellow edge band and the red edge band is extracted, and the edge area of the junction between the leaf sheath and the leaf is weighted fused to output the junction enhanced image; S3: Directed neighborhood contrast enhancement is applied to the boundary-enhanced image, and an enhancement operator is constructed based on the potential diffusion direction of stripe rust hyphae to identify the latent period marking area and generate a stripe rust feature map. S4: Based on the characteristic map of stripe rust during the latent period, the stomatal conductance value of the leaf surface is inverted by combining thermal infrared band images and environmental meteorological parameters. At the same time, the stomatal opening level at each spatial position is quantitatively mapped to form a stomatal opening density map and output a disease development probability map; S5: Align the disease development probability map with the 3D point cloud data of wheat plants. Based on the disease probability and spatial adjacency of each point, construct vertical and horizontal propagation weight functions and perform iterative modeling to generate a 3D disease distribution map.

[0038] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0039] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A multi-scale detection system for wheat stripe rust based on multispectral images from unmanned aerial vehicles, characterized by: It includes an attitude compensation image registration module, a leaf sheath-leaf boundary enhancement module, a stripe rust latent period pattern extraction module, a canopy stomatal density inversion and mapping module, and a three-dimensional space infection simulation module; among which: Attitude Compensation Image Registration Module: This module receives the original multispectral image data collected by the drone, fuses the attitude information output by the drone's inertial measurement unit with a sub-pixel image registration algorithm based on phase correlation, and generates a registered image set that eliminates flight attitude interference. The leaf sheath-leaf boundary enhancement module is connected to the attitude compensation image registration module, which receives the registered image set, fuses the image gradient information of the yellow edge band and the red edge band, enhances the structural boundary features of the leaf sheath-leaf boundary area, and outputs an enhanced image. Stripe rust latent period pattern extraction module: This module is connected to the leaf sheath-leaf boundary enhancement module to receive the enhanced image, construct a directional neighborhood contrast enhancement operator based on the potential expansion path of stripe rust hyphae, extract the latent period pattern features of the disease, and output the latent period stripe rust feature map; Canopy Stomatal Density Inversion and Mapping Module: This module is connected to the Stripe Rust Incubation Period Mark Extraction Module and is used to receive the stripe rust characteristic map during the incubation period. It inverts the canopy stomatal conductance parameters based on the thermal infrared band image, overlays the generated stomatal open density spatial distribution map, and outputs the disease development probability map. Three-dimensional infection simulation module: connected to the canopy stomatal density inversion and mapping module, used to receive the disease development probability map, and combined with the three-dimensional point cloud data of wheat plants, to construct a three-dimensional disease distribution map including vertical height distribution, infection gradient evolution and horizontal transmission trend.

2. The multi-scale wheat stripe rust detection system based on drone multispectral images according to claim 1, characterized in that: The posture compensation image registration module includes a posture data solving unit, a posture disturbance pre-compensation unit and a sub-pixel image registration unit; wherein: Attitude data calculation unit: used to receive the raw data of acceleration, angular velocity and magnetic field intensity output by the UAV inertial measurement unit, and calculate the three-dimensional attitude angle at the current image capture time based on the extended Kalman filter method, including pitch angle, roll angle and yaw angle; Attitude disturbance pre-compensation unit: used to construct the attitude disturbance compensation matrix of the multispectral image in three-dimensional space based on the three-dimensional attitude angle parameters output by the attitude data solution unit, and perform rotation and projection transformation on the original image coordinates to generate a coarse registration image sequence; Sub-pixel image registration unit: It is used to refine the coarse registration image sequence output by the attitude disturbance pre-compensation unit. It uses a sub-pixel displacement estimation algorithm based on the frequency domain phase correlation method to calculate the sub-pixel translation vector between images and perform image resampling, thereby outputting a set of registration images that eliminates flight attitude interference.

3. The multi-scale wheat stripe rust detection system based on drone multispectral images according to claim 2, characterized in that: The sub-pixel image registration unit includes: Reference image selection subunit: used to select a frame with the best signal-to-noise ratio from the coarse registration image sequence as the global registration reference image, and use it as the comparison benchmark frame for subsequent registration operations; Frequency domain conversion subunit: used to perform two-dimensional Fourier transform on the reference image and the image to be registered, and the transformation result is and ,in is the frequency domain coordinate; Phase difference calculation subunit: used to calculate the phase difference map of the normalized cross power spectrum ; Sub-pixel displacement estimation subunit: Phase difference map Perform a two-dimensional inverse Fourier transform to obtain the cross-correlation function graph, and use quadratic interpolation at its peak position to calculate the sub-pixel displacement vector ; Image resampling subunit: Based on the estimated , a sub-pixel translation transformation is performed on the registration image, and the bicubic interpolation method is used to complete the image resampling, and the final output is a set of registration images that eliminates the interference of flight posture.

4. The multi-scale wheat stripe rust detection system based on drone multispectral images according to claim 1, characterized in that: The leaf sheath-leaf boundary enhancement module includes a band image extraction unit, an edge gradient calculation unit, a structure boundary fusion unit, and an image enhancement output unit; wherein: Band image extraction unit: used to extract yellow edge band image and red edge band image from the registration image set output by the attitude compensation image registration module; Edge gradient calculation unit: used to calculate the corresponding gradient amplitude of the yellow edge band and red edge band images respectively; Structural boundary fusion unit: uses a weighted fusion strategy to fuse the gradient maps of the two bands and construct a fused boundary map , whose expression is: ,in, and The gradient amplitudes of the yellow edge band and the red edge band are respectively: and are the fusion weight parameters of the yellow edge and red edge bands respectively; Image enhancement output unit: based on fusion boundary map The original image is enhanced in the gradient domain to highlight the edge response of the boundary between the leaf sheath and the leaf blade, and the enhanced image is output for subsequent stripe feature extraction.

5. The multi-scale wheat stripe rust detection system based on drone multispectral images according to claim 1, characterized in that: The stripe rust latent period pattern extraction module includes a direction hypothesis generation unit, a directional neighborhood construction unit, a contrast enhancement calculation unit and a feature map output unit; wherein: Direction hypothesis generation unit: This unit is used to set multiple sets of preset direction angles based on the direction of the main veins of wheat leaves and their consistency with the typical propagation direction of stripe rust hyphae. This is used as the orientation basis for potential expansion paths, forming a direction set list for subsequent stripe detection. Directional neighborhood construction unit: Based on each direction angle provided by the direction hypothesis generation unit, a directional trapezoidal neighborhood window with the direction as the main axis is constructed around each pixel point in the image; Contrast enhancement calculation unit: used to calculate the brightness contrast difference between the current pixel and its neighborhood background in a directional trapezoidal neighborhood window, using the principle of directional consistency to enhance the response intensity of latent lesions, highlight the grayscale feature differences in the latent stripe rust area, and eliminate noise interference not along the extension direction; Characteristic map output unit: used to perform maximum fusion on the enhanced response maps in all directions, extract the latent period stripe area with the strongest response in multiple directions, and generate the latent period stripe rust characteristic map.

6. The multi-scale wheat stripe rust detection system based on drone multispectral images according to claim 1, characterized in that: The contrast enhancement calculation unit includes: Local mean brightness calculation subunit: used to calculate the local mean brightness of the current pixel in the directional trapezoidal neighborhood window. As the center, extract the direction The neighborhood pixel set under , and calculate the average brightness value of the neighborhood ; Directional consistency score calculation subunit: used to extract the current pixel point The image gradient direction at the position and the preset direction angle Perform similarity calculation to obtain direction consistency score , the formula is: ,in, Indicates the image gradient direction of the current pixel; Enhanced response generation subunit: used to calculate the direction of the current pixel Enhanced response value under , the formula is: ,in, Indicates the grayscale value of the current pixel.

7. The multi-scale wheat stripe rust detection system based on drone multispectral images according to claim 1, characterized in that: The canopy stomatal density inversion and mapping module includes a temperature map extraction unit, a stomatal conductance inversion unit, a density map generation unit, and a disease probability output unit; wherein: Temperature map extraction unit: used to extract thermal infrared band images from the registered image set, perform radiation calibration, background thermal field correction and noise filtering on the images, and output a temperature map representing the thermal distribution of the crop surface; Stomatal conductance inversion unit: This unit receives temperature maps and latent stripe rust characteristic maps, models crop surface heat flux, and constructs an analytical model based on environmental meteorological parameters to invert leaf stomatal conductance values, thereby representing changes in transpiration levels and water vapor exchange capacity. Density map generation unit: used to calculate the stomatal open density value after obtaining the stomatal conductance value of the leaf surface, combined with the distribution characteristics of the latent period spots, and construct a spatial distribution map to form a stomatal open density map; Disease probability output unit: used to classify and evaluate the density level of each area based on the stomatal opening density map and the prior range of stomatal behavior of healthy leaves, and output the disease development probability map accordingly.

8. The multi-scale wheat stripe rust detection system based on drone multispectral images according to claim 7, characterized in that: The disease probability output unit includes: Health behavior modeling subunit: used to construct the normal range of stomatal opening density of healthy wheat leaves, including the minimum reference value With the maximum reference value , as a priori benchmark for subsequent classification evaluation; Density classification judgment subunit: used to receive the density value of each pixel in the stomata open density map , and based on its and and make regional classification judgments based on the relationship between them; like , determined to be abnormally closed type, indicating a potential stomatal inactivation area; like , it is judged as abnormally open type, indicating a transpiration disorder area; like , determined as the normal stomatal behavior area; Probability map generation subunit: used to assign disease probability values to each pixel in the image based on the classification judgment results The disease probability value of each pixel is spliced according to the original image coordinates to generate a complete disease development probability map. The expression is: 。 9. The multi-scale wheat stripe rust detection system based on drone multispectral images according to claim 1, characterized in that: The three-dimensional space infection simulation module includes a point cloud structure reconstruction unit, a disease probability mapping unit, an infection diffusion modeling unit and a three-dimensional distribution map generation unit; wherein: Point cloud structure reconstruction unit: used to receive multi-angle aerial images or laser scanning data, reconstruct a three-dimensional point cloud model of the wheat canopy, and assign spatial coordinates to each point to form a spatial skeleton with plant structure levels; Disease probability mapping unit: used to spatially register the two-dimensional probability values in the disease development probability map with the positions of the corresponding canopy projection points in the three-dimensional point cloud according to the image pixel coordinates, and map the disease probability value of each two-dimensional pixel to the corresponding point cloud node; Infection and Diffusion Modeling Unit: Based on the mapped disease probability points, combined with the spatial adjacency and vertical hierarchical distribution information of each point in the 3D point cloud, it constructs disease propagation weight functions in the vertical and horizontal directions, and iteratively expands and simulates the spread trend of the disease from high-risk points to adjacent structures, while recording its diffusion path and probability gradient changes; Three-dimensional distribution map generation unit: Based on the infection modeling results, the disease risk level of each point in the three-dimensional space is output, and a three-dimensional disease distribution map is formed according to the spatial coordinates.

10. A multi-scale detection method for wheat stripe rust based on drone multispectral images, implemented by the multi-scale detection system for wheat stripe rust based on drone multispectral images according to any one of claims 1 to 9, characterized in that: The following steps are involved: S1: Use UAV to acquire multispectral image data, and combine it with the attitude information output by the inertial measurement unit to generate a registered image set that removes flight attitude errors through a phase-correlation sub-pixel registration algorithm; S2: Based on the registered image set, the gradient information of the yellow edge band and the red edge band is extracted, and the edge area of the junction between the leaf sheath and the leaf is weighted fused to output the junction enhanced image; S3: Directed neighborhood contrast enhancement is applied to the boundary-enhanced image, and an enhancement operator is constructed based on the potential diffusion direction of stripe rust hyphae to identify the latent period marking area and generate a stripe rust feature map. S4: Based on the characteristic map of stripe rust during the latent period, the stomatal conductance value of the leaf surface is inverted by combining thermal infrared band images and environmental meteorological parameters. At the same time, the stomatal opening level at each spatial position is quantitatively mapped to form a stomatal opening density map and output a disease development probability map; S5: Align the disease development probability map with the 3D point cloud data of wheat plants. Based on the disease probability and spatial adjacency of each point, construct vertical and horizontal propagation weight functions and perform iterative modeling to generate a 3D disease distribution map.

Images