A method and device for identifying diseases of perilla leaves

Abstract

The present invention discloses a method and device for identifying diseases of perilla leaves, and belongs to the technical field of disease identification. The present invention sets a plurality of detection stations of workbenches, continuously collects and processes radiographic images, fluorescent images, and thermal radiation images of perilla leaf plants at different collection periods at the same detection station, and effectively identifies early diseases of perilla leaf plants. Furthermore, by cropping the first continuous image, attention is placed on abnormal features, thereby accelerating the detection speed. Feature extraction is performed on the first abnormal image, which facilitates the image classifier to learn more subtle features, accelerates the speed of disease identification, and improves the sensitivity of disease identification. The signal acquisition device of the present invention uses broadband polarized light, which can identify the reflection distortion area of ​​the reflection image, and adjusts the detection station in combination with the position of the reflection distortion area to ensure that each pixel of the perilla leaf plant is fully displayed. The present invention adjusts the fluorescence data according to the thermal image data, which can improve the accuracy of the disease data.

Description

Technical Field

[0001] The invention relates to the technical field of plant disease identification, and in particular to a method and device for identifying a perilla leaf disease. Background Art

[0002] Perilla leaf diseases are characterized by diversity and continuity. In the early stage of the disease, the appearance of the plant does not change significantly, and the disease characteristics are mostly manifested at the molecular organic matter level. As the disease worsens, the disease characteristics can only be manifested through changes in the color, texture, and morphology of the leaves. A Chinese patent application with publication number CN117994729A discloses a method and system for monitoring the growth of garden plants. The method obtains multiple plant growth image data sets of a target garden, and a lesion abnormality recognition channel performs lesion abnormality analysis on the multiple plant growth image data sets to obtain abnormal lesion recognition results, and abnormal lesion areas are identified based on the abnormal lesion recognition results. In the early stage of plant infection, the plant growth image data set cannot be sensitively recognized by the lesion abnormality recognition channel.

[0003] The Chinese patent application with publication number CN118443604A discloses a tobacco disease monitoring method based on airborne hyperspectral reflectance and chlorophyll fluorescence. The method establishes a tobacco disease test area, collects tobacco disease samples, establishes a disease sample library, performs airborne hyperspectral collection in the tobacco disease test area, performs hyperspectral preprocessing, constructs a tobacco disease monitoring model based on the preprocessed hyperspectrum, and performs tobacco area disease monitoring based on the disease monitoring model. The above patent identifies plant diseases through hyperspectral reflectance and chlorophyll fluorescence, and constructs a fluorescence ratio index for early identification of tobacco diseases. Since not all types of diseases will immediately cause significant changes in the fluorescence characteristics of plants, the existing technology needs to be further improved. Summary of the invention

[0004] In order to solve the defects of the above-mentioned prior art, the present invention proposes a method and device for identifying diseases of perilla leaves. The present invention continuously collects reflection images, fluorescence images, and thermal radiation images of each detection station of perilla leaf plants, and processes images of different collection cycles of the same detection station respectively, so as to effectively identify diseases of perilla leaf plants at different periods and of different types. Furthermore, the present invention identifies the reflection distortion area of ​​the reflection image and adjusts the detection station to ensure that each pixel of the perilla leaf plant is fully displayed, and at the same time adjusts the fluorescence data according to the thermal image data, so as to improve the accuracy of the disease data.

[0005] The technical solution of the present invention is achieved in this way:

[0006] A method for identifying diseases of perilla leaves comprises the following steps:

[0007] Step 1: The signal acquisition device is installed on a workbench, the perilla leaf plants are arranged in the sample area of ​​the workbench, and multiple detection stations of the workbench are preset;

[0008] Step 2: The signal acquisition device enters different detection stations in sequence, and the signal acquisition device collects polarized light data, first fluorescence data, and thermal imaging data of the perilla leaf plant, and generates second fluorescence data after adjusting the first fluorescence data according to the thermal imaging data;

[0009] Step 3: Generate a reflection image based on the polarized light data, generate a fluorescence image based on the second fluorescence data, generate a thermal radiation image based on the thermal imaging data, and identify a reflection distortion area of ​​the reflection image in the polarized light data;

[0010] Step 4: Repeat steps 2 and 3 to obtain a reflection image, a fluorescence image, and a thermal radiation image of another detection station. If at least one coordinate point of the perilla leaf plant is within the reflection distortion area in any reflection image, proceed to step 5; otherwise, proceed to step 6.

[0011] Step 5: If all inspection stations have been completed, proceed to step 6, otherwise return to step 4;

[0012] Step 6: After fitting the reflection distortion area, the reflection image is updated, and multiple sets of reflection images, fluorescence images, and thermal radiation images are obtained in different acquisition cycles. The reflection images, fluorescence images, and thermal radiation images of the same detection position generate the first continuous atlas, the second continuous atlas, and the third continuous atlas;

[0013] Step 7: Identify the first continuous atlas, the second continuous atlas, and the third continuous atlas of the same detection position, obtain the first feature data, the second feature data, and the third feature data, and then determine the type of disease of the perilla leaf plant.

[0014] In the present invention, in step 2, the signal acquisition device includes a light generator and a light receiver. At each detection station, the light generator generates broadband polarized light and ultraviolet excitation light in sequence, and the light receiver collects reflected polarization signals, fluorescence signals, and thermal radiation signals in sequence to generate polarization light data, first fluorescence data, and thermal imaging data.

[0015] In the present invention, in step 2, the temperature deviation ΔC of each pixel is calculated based on the thermal imaging data. xy , the fluorescence intensity b of the pixel in the second fluorescence data xy =b xy 'exp[-E / (KΔC xy )],b xy ' is the fluorescence intensity of the pixel in the first fluorescence data, E is the fluorescence activation energy of the perilla leaf, K is the Boltzmann constant, x is the horizontal coordinate of the pixel, and y is the vertical coordinate of the pixel.

[0016] In the present invention, in step 3, three sets of polarization data are extracted from the polarized light data, and the pixel polarization degree of each pixel is calculated. The pixel points whose pixel polarization degree is greater than the polarization threshold constitute the reflection distortion area.

[0017] In the present invention, in step 4, the pixel points corresponding to the coordinate points of the perilla leaf plant in each reflection image are searched according to the image conversion relationship to determine whether the pixel points corresponding to the coordinate points are in the reflection distortion area.

[0018] In the present invention, in step 6, the first continuous atlas U 1 ={A 11 ,...,A 1t ,...,A 1T}, T is the total acquisition period, A 1t is the reflection image of the acquisition period t, the second continuous atlas U 2 ={A 21 ,...,A 2t ,...,A 2T}, A 2t The third continuous atlas U is the fluorescence image collected during the period t. 3 ={A 31 ,...,A 3t ,...,A 3T}, A 3t is the thermal radiation image of the acquisition period t.

[0019] In the present invention, in step 7, the first continuous atlas, the second continuous atlas, and the third continuous atlas of the same inspection station are substituted into the image segmentor to obtain the first abnormal atlas, the second abnormal atlas, and the third abnormal atlas, and the first feature data is obtained according to the first abnormal atlas, the second feature data is obtained according to the second abnormal atlas, and the third feature data is obtained according to the third abnormal atlas.

[0020] In the present invention, a first difference between a reflectance image in a first continuous atlas and a one-dimensional reflectance spectrum of a sample perilla leaf is calculated, and a first abnormal atlas is segmented from the first continuous atlas based on the first difference, wherein the first abnormal atlas consists of a plurality of first abnormal images, the second abnormal atlas consists of a plurality of second abnormal images, and the third abnormal atlas consists of a plurality of third abnormal images.

[0021] In the present invention, the first abnormal atlas is clustered to obtain the first characteristic data, the second abnormal atlas is clustered to obtain the second characteristic data, and the third abnormal atlas is clustered to obtain the third characteristic data.

[0022] An identification device for realizing the method for identifying diseases of perilla leaves, comprising:

[0023] A workbench having a plurality of testing stations, wherein the perilla leaf plants are arranged in a sample area of ​​the workbench;

[0024] A signal acquisition device, the signal acquisition device is used to collect polarized light data, first fluorescence data and thermal image data of the perilla leaf plant;

[0025] A first controller, the first controller is used to move the signal acquisition device to different detection positions;

[0026] A second controller, the second controller is used to switch the signal acquisition device to a reflected light detection state, a fluorescence detection state, and a thermal radiation detection state;

[0027] An image processor, the image processor is used to generate reflection images, fluorescence images, and thermal radiation images of different acquisition cycles and different detection positions;

[0028] An image segmenter, the image segmenter is used to identify the first continuous atlas set, the second continuous atlas set, and the third continuous atlas set;

[0029] A data analyzer, the data analyzer is used to obtain the first characteristic data, the second characteristic data, and the third characteristic data;

[0030] An image classifier is used to determine the types of diseases on perilla leaf plants.

[0031] The method and device for identifying diseases of perilla leaves of the present invention have the following beneficial effects: the workbench of the present invention has multiple detection stations, and the signal acquisition device continuously collects radiographic images, fluorescent images, and thermal radiation images of each detection station of the perilla leaf plant, which can reflect different types of diseases. The images of different acquisition cycles of the same detection station are processed separately to avoid the various changes of the plants in the early stage of disease infection, which lead to the inability to identify the collected data, and effectively identify the early diseases of the perilla leaf plant. Further, the first abnormal image is obtained by cutting the first continuous image through the first difference degree and setting the first difference threshold, and the attention of plant disease identification is placed on the abnormal features to speed up the detection speed. Feature extraction is performed on the first abnormal image, which facilitates the image classifier to learn more subtle features in the first abnormal image set, speeds up the speed of disease identification on the first feature data, and improves the sensitivity of disease identification. The signal acquisition device of the present invention uses circularly polarized light, can identify the reflection distortion area of ​​the reflection image, and adjusts the detection station in combination with the position of the reflection distortion area to ensure that each pixel of the perilla leaf plant is fully displayed. The present invention adjusts the fluorescence data according to the thermal imaging data, and can improve the accuracy of the disease data. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The flowchart of the method for identifying diseases of perilla leaves of the present invention is as follows;

[0033] Figure 2 It is a schematic diagram of multiple inspection stations of a workbench of the present invention;

[0034] Figure 3 The optical path diagram of the signal acquisition device of the present invention;

[0035] Figure 4 It is a schematic diagram of the reflection distortion area of ​​the reflection spectrum diagram of the present invention;

[0036] Figure 5 It is a schematic diagram of a first abnormal graph of the reflectance spectrum graph of the present invention;

[0037] Figure 6 is a schematic diagram of a one-dimensional reflectance spectrum of a sample of Perilla leaf of the present invention;

[0038] Figure 7 A schematic diagram of a process of obtaining first characteristic data according to a first abnormal atlas in the present invention;

[0039] Figure 8 The present invention is a block diagram of an identification device for implementing the method for identifying diseases of perilla leaves. DETAILED DESCRIPTION

[0040] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0041] In the early stage of infection, the disease characteristics are not obvious, and the disease cannot be identified based on the appearance image of the plant. At the same time, because different types of diseases have different effects on plants, it is not possible to sensitively, quickly, comprehensively, and completely identify plant diseases based on a single, single-directional, and certain sampling period of polarized light data or first fluorescence data or thermal image data. Embodiment 1

[0042] like Figures 1 to 5 As shown, the method for identifying diseases of perilla leaves of the present invention comprises the following steps.

[0043] Step 1: The signal acquisition device is installed on the workbench, the perilla leaf plants are arranged in the sample area of ​​the workbench, and multiple detection stations of the workbench are preset. Figure 2 As shown, the workbench is provided with four detection stations, and the dotted rectangle is a spare detection station. When the data collected by all the preset detection stations still have defects, the spare detection station is used to ensure that the signal acquisition device can collect complete data of the perilla leaf plant. Each detection station is the same distance from the perilla leaf plant.

[0044] Step 2: The signal acquisition device enters different detection stations in turn, collects polarized light data, first fluorescence data and thermal imaging data of the perilla leaf plant, and generates second fluorescence data after adjusting the first fluorescence data according to the thermal imaging data. Figure 3 As shown, the signal acquisition device includes a light generator and a light receiver, the angle between the light generator and the light receiver is 30°, the light generator is a surface light source, including a plurality of evenly arranged point light sources, and at each detection station, the light generator generates broadband polarized light and ultraviolet excitation light in sequence to illuminate the perilla leaf plants in the sample area, and the light receiver collects reflected polarization signals, fluorescence signals and thermal radiation signals in sequence to generate polarization light data, first fluorescence data and thermal imaging data.

[0045] Calculate the temperature deviation ΔC of each pixel based on the thermal image data xy Steps: First, convert the pixel value of each pixel in the thermal image data into a temperature value, and then calculate the average temperature of the thermal image data Finally, the temperature deviation of each pixel is calculated, and the temperature value C xy =d 1 ×c xy +d 2 , , temperature deviation ΔC xy =C xy - , d 1 is the first constant, d 2 is the second constant, the first constant and the second constant are both constants obtained by calibrating the optical receiver, c xy is the pixel value of the thermal image data, H is the number of pixels in the thermal image data, C h is the hth pixel in the thermal image data.

[0046] The second fluorescence data is generated by adjusting the first fluorescence data according to the temperature deviation. The fluorescence intensity b of the pixel in the second fluorescence data xy =b xy 'exp[-E / (KΔC xy )],b xy ' is the fluorescence intensity of the pixel in the first fluorescence data, E is the fluorescence activation energy of the perilla leaf, K is the Boltzmann constant, x is the horizontal coordinate of the pixel, and y is the vertical coordinate of the pixel.

[0047] Step 3: Generate a reflection image based on the polarized light data, generate a fluorescence image based on the second fluorescence data, generate a thermal radiation image based on the thermal imaging data, and identify the reflection distortion area of ​​the reflection image in the polarized light data. Extract three sets of polarization data from the polarized light data, calculate the pixel polarization degree of each pixel point, and the pixels whose pixel polarization degree is greater than the polarization threshold constitute the reflection distortion area. Extract three sets of polarization data from the polarized light data, and the three sets of polarization data are right-handed circularly polarized light intensity I 1 , left-handed circularly polarized light intensity I 2 , linear polarized light intensity I 3 , pixel polarization , S 0 =I 1 +I 2 +I 3 , S 1 =2I 1 -S 0 , S 2 =2I 2 -S 0 , S 3 =2I 3 -S 0 -S 1 -S 2 ,The area where the pixel polarization degree in the reflection image is greater than the reference value is the reflection distortion area.

[0048] Step 4: Repeat steps 2 and 3 to obtain a reflection image, a fluorescence image, and a thermal radiation image of another detection station. If at least one coordinate point of the perilla leaf plant is within the reflection distortion region in any reflection image, proceed to step 5, otherwise proceed to step 6. The present invention searches for the pixel points corresponding to the coordinate points of the perilla leaf plant in each reflection image based on the image conversion relationship, and determines whether the pixel points corresponding to the coordinate points are within the reflection distortion region. The image conversion relationship is, for example, the coordinate relationship of a two-dimensional image relative to a world coordinate system. This conversion is mainly completed through two steps: conversion from the world coordinate system to the optical receiver coordinate system, and conversion from the optical receiver coordinate system to the two-dimensional image. In a more specific embodiment, the image conversion relationship is obtained based on the external parameter matrix and the internal parameter matrix of the optical receiver. The external parameter matrix is ​​used to describe the position and posture of the optical receiver relative to the world coordinate system, and the internal parameter matrix is ​​used to describe the relationship between the coordinates of the optical receiver and the coordinates of the pixel points.

[0049] Step 5: If all the detection stations have been completed, go to step 6, otherwise return to step 4. The data collected by multiple detection stations ensure that all coordinate points of the perilla leaf plant appear in the non-reflection distortion area of ​​any reflection image, ensuring the integrity and accuracy of the collected data.

[0050] Step 6: Update the reflection image after fitting the reflection distortion area, obtain multiple sets of reflection images, fluorescence images, and thermal radiation images in different acquisition cycles, and generate the first continuous atlas, the second continuous atlas, and the third continuous atlas for the reflection images, fluorescence images, and thermal radiation images of the same detection position. Figure 4 As shown, the reflection distortion area is filled by the reflection images of different detection stations to update the reflection image. 1 ={A 11 ,...,A 1t ,...,A 1T}, T is the total acquisition period, A 1t is the reflection image of the acquisition period t, the second continuous atlas U 2 ={A 21 ,...,A 2t ,...,A 2T}, A 2t The third continuous atlas U is the fluorescence image collected during the period t. 3 ={A 31 ,...,A 3t ,...,A 3T}, A 3t is the thermal radiation image of the acquisition cycle t. The acquisition cycle is reasonably set according to the current growth stage of the perilla leaf plant, the disease infection time and the obvious degree of disease symptoms, so as to avoid the death of the perilla leaf plant due to too long acquisition cycle and unclear disease symptoms due to too short acquisition cycle. The reflection image, fluorescence image and thermal radiation image of each detection station of the perilla leaf plant are continuously collected through multiple acquisition cycles to avoid the various changes of the perilla leaf plant in the early stage of disease infection are not obvious, resulting in the inability to identify the collected data, thereby improving the accuracy of disease identification.

[0051] Step 7: Identify the first continuous atlas, the second continuous atlas, and the third continuous atlas of the same detection position, obtain the first feature data, the second feature data, and the third feature data, and then determine the type of disease of the perilla leaf plant. Substitute the first continuous atlas, the second continuous atlas, and the third continuous atlas of the same detection position into the image segmenter to obtain the first abnormal atlas, the second abnormal atlas, and the third abnormal atlas. The first feature data is obtained based on the first abnormal atlas, the second feature data is obtained based on the second abnormal atlas, and the third feature data is obtained based on the third abnormal atlas. Figure 5 As shown, the reflected image includes a perilla leaf plant image and a background image, the perilla leaf plant image includes a first abnormal image, the black spots in the image are abnormal features, the abnormal features are visible disease infection symptoms on the perilla leaf plant, the first abnormal image includes visible disease infection symptoms and invisible disease infection symptoms, and the pixel point at each coordinate position of the perilla leaf plant image of the reflected image is a one-dimensional reflection spectrum. The specific method for determining the disease type of the perilla leaf plant is described in detail in Example 2. Embodiment 2

[0052] like Figure 6 to Figure 7 As shown, this embodiment further discloses a method for determining the disease type of the Perilla leaf plant in step 7.

[0053] Calculate the first difference between the reflection image in the first continuous atlas and the one-dimensional reflection spectrum of the sample perilla leaf, and segment the first abnormal atlas from the first continuous atlas based on the first difference, wherein the first abnormal atlas consists of multiple first abnormal images, the second abnormal atlas consists of multiple second abnormal images, and the third abnormal atlas consists of multiple third abnormal images. Cluster the first abnormal atlas to obtain the first feature data, cluster the second abnormal atlas to obtain the second feature data, and cluster the third abnormal atlas to obtain the third feature data. Construct an image classifier, substitute the first feature data, the second feature data, and the third feature data into the image classifier to obtain the type of disease of the perilla leaf plant.

[0054] like Figure 6 As shown, the sample perilla leaf is a healthy leaf, and the one-dimensional reflectance spectrum of the sample perilla leaf is (a 1 ,...,a n ,...,a N ), a n is the reflectivity of broadband polarized light of different wavelength n, and the first difference of the pixel point with coordinate position (x, y) in the reflected image , the one-dimensional reflection spectrum of the pixel point with coordinate position (x, y) in the reflection image is (a 1xy ,...,a nxy ,...,a Nxy ), since the one-dimensional reflectance spectrum of the sample perilla leaf has slight differences at different growth stages, a constant is subtracted from the first difference degree to avoid errors in calculating the first difference degree, which leads to the expansion of the boundary and the wrong first abnormal map. The first difference threshold is reasonably set according to the visibility of the disease infection symptoms and the growth stage of the perilla leaf plant. The pixel point at each coordinate position of the perilla leaf plant image in the fluorescence image is the one-dimensional fluorescence spectrum data, and the second abnormal atlas and the third abnormal atlas are obtained according to the acquisition method of the first abnormal atlas.

[0055] In order to increase the data information of the first continuous atlas, an image segmenter is introduced for data enhancement. The image segmenter uses the one-dimensional reflectance spectrum of the sample perilla leaf as the standard, calculates the first difference between the pixel point at each coordinate position of the reflectance image in the first continuous atlas and the one-dimensional reflectance spectrum of the sample perilla leaf, sets the first difference threshold, and segments the reflectance image with each coordinate position in the reflectance image where the first difference is greater than the first difference threshold as the boundary to obtain the first abnormal image. The image segmenter focuses on the abnormal features in the identification of perilla leaf plant diseases, speeding up the detection speed of perilla leaf plant diseases. The image segmenter has the same effect on the second and third continuous atlases as on the first continuous atlas.

[0056] The data analyzer obtains the first characteristic data according to the first abnormal atlas, obtains the second characteristic data according to the second abnormal atlas, and obtains the third characteristic data according to the third abnormal atlas. Figure 7 As shown, the first difference between the pixel point at each coordinate position in the first abnormal image and the one-dimensional reflectance spectrum of the sample perilla leaf is calculated, the coordinate position where the first difference is less than the first difference threshold in the first abnormal image is excluded, the pixel points at the remaining coordinate positions in the first abnormal image are clustered to obtain one-dimensional reflectance spectrum feature data, and the one-dimensional reflectance spectrum feature data of the first abnormal image are clustered to obtain first feature data, wherein the first feature data includes at least one one-dimensional reflectance spectrum feature data. The algorithm for clustering the one-dimensional reflectance spectrum feature data is, for example, a mean clustering algorithm. The second abnormal image set is operated according to the method for obtaining the first feature data to obtain the second feature data. The third abnormal image set is image clustered to obtain the third feature data. The image clustering method is, for example, a hierarchical clustering algorithm, a spectral clustering algorithm, etc. Feature extraction is performed on the first abnormal image, so that the image classifier can learn more subtle features in the first abnormal image set, accelerate the speed of disease identification, and improve the sensitivity of disease identification.

[0057] The method for constructing an image classifier includes collecting a first data set, a second data set and a third data set, training a first classifier with the first data set, training a second classifier with the second data set, training a third classifier with the third data set, and combining the first classifier, the second classifier and the third classifier into an image classifier. The first data set includes one-dimensional reflectance spectra of sample perilla leaves and plants with various diseases of perilla leaves. The second data set includes one-dimensional fluorescence spectrum data of sample perilla leaves and plants with various diseases of perilla leaves. The third data set includes thermal radiation images of sample perilla leaves and plants with various diseases of perilla leaves. The first classifier and the second classifier may be support vector machines, neural networks, random forests, etc., and the third classifier may be convolutional neural networks, residual networks, ensemble learning methods, etc. The first data set, the second data set and the third data set are collected by crawler technology. The first classifier, the second classifier and the third classifier are trained respectively by the first data set, the second data set and the third data set to obtain appropriate classification parameters, so that the image classifier can identify diseases of perilla leaf plants.

[0058] The recognition method of the image classifier: the first feature data is identified by the first classifier to obtain a first recognition result, the second feature data is identified by the second classifier to obtain a second recognition result, and the third feature data is identified by the third classifier to obtain a third recognition result. The image classifier obtains a disease recognition result based on the first recognition result, the second recognition result and the third recognition result and outputs the disease type of the perilla leaf plant.

[0059] The first recognition result includes at least one disease type and a corresponding first probability. 4 ={u 11 ,...,u 1m ,...,u 1M}, M is the number of disease types in the first recognition result. 1m =(v 1m , p 1m ), v 1m is the mth disease in the first recognition result, p 1m is the first probability of the mth disease in the first recognition result. 1m = the number of one-dimensional reflectance spectrum characteristic data identified as the mth disease / the number of one-dimensional reflectance spectrum characteristic data in the first characteristic data. The first probability represents the possibility of the disease appearing in the first characteristic data. The greater the first probability, the greater the possibility of the disease appearing in the perilla leaf plant. The second recognition result includes at least one disease type and the corresponding second probability, and the third recognition result includes at least one disease type and the corresponding third probability.

[0060] The disease identification result U 7={u 41 ,...,u 4g ,...,u 4G},u 4g =(v 4g , p 4g ), v 4g is the gth disease in the disease identification result, p 4g is the fourth probability of the g-th disease in the disease identification result, , μ is the probability threshold. The probability threshold is set reasonably according to the visibility of the disease infection symptoms. If the visibility of the disease infection symptoms is high, the probability threshold is high, and if the visibility of the disease infection symptoms is low, the probability threshold is low. Embodiment 3

[0061] like Figure 8 As shown, an identification device for realizing the identification method of perilla leaf diseases includes: a workbench, a signal acquisition device, a first controller, a second controller, an image processor, an image segmenter, a data analyzer, and an image classifier. The workbench has multiple detection stations, and the perilla leaf plants are arranged in the sample area of ​​the workbench. The signal acquisition device includes a light generator and a light receiver, the light generator is used to generate broadband polarized light and ultraviolet excitation light, and the light receiver is used to collect reflected polarization signals, fluorescence signals, and thermal radiation signals to generate polarization data, first fluorescence data, and thermal image data. The first controller is used to move the signal acquisition device to different detection stations according to the reflection distortion area of ​​the reflected image. The second controller is used to switch the signal acquisition device to a reflected light detection state, a fluorescence detection state, and a thermal radiation detection state. The image processor is used to generate reflected images, fluorescence images, and thermal radiation images of different acquisition cycles and different detection stations, and generate a first continuous atlas, a second continuous atlas, and a third continuous atlas according to the reflected image, fluorescence image, and thermal radiation image of the same detection position. The image segmenter is used to identify the first continuous atlas, the second continuous atlas, and the third continuous atlas, and obtain the first abnormal atlas, the second abnormal atlas, and the third abnormal atlas. The data analyzer is used to obtain the first feature data, the second feature data, and the third feature data based on the first abnormal atlas, the second abnormal atlas, and the third abnormal atlas. The image classifier is used to determine the disease type of the perilla leaf plant based on the first feature data, the second feature data, and the third feature data.

[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying diseases of perilla leaves, characterized in that: The following steps are involved: Step 1: The signal acquisition device is installed on a workbench, the perilla leaf plants are arranged in the sample area of ​​the workbench, and multiple detection stations of the workbench are preset; Step 2: The signal acquisition device enters different detection stations in sequence, and the signal acquisition device collects polarized light data, first fluorescence data, and thermal imaging data of the perilla leaf plant, and generates second fluorescence data after adjusting the first fluorescence data according to the thermal imaging data; Step 3: Generate a reflection image based on the polarized light data, generate a fluorescence image based on the second fluorescence data, generate a thermal radiation image based on the thermal imaging data, and identify a reflection distortion area of ​​the reflection image in the polarized light data; Step 4: Repeat steps 2 and 3 to obtain a reflection image, a fluorescence image, and a thermal radiation image of another detection station. If at least one coordinate point of the perilla leaf plant is within the reflection distortion area in any reflection image, proceed to step 5; otherwise, proceed to step 6. Step 5: If all inspection stations have been completed, proceed to step 6, otherwise return to step 4; Step 6: After fitting the reflection distortion area, the reflection image is updated, and multiple sets of reflection images, fluorescence images, and thermal radiation images are obtained in different acquisition cycles. The reflection images, fluorescence images, and thermal radiation images of the same detection position generate the first continuous atlas, the second continuous atlas, and the third continuous atlas; Step 7: Identify the first continuous atlas, the second continuous atlas, and the third continuous atlas of the same detection position, obtain the first feature data, the second feature data, and the third feature data, and then determine the disease type of the perilla leaf plant, wherein: In step 2, the temperature deviation ΔC of each pixel is calculated based on the thermal image data. xy , the fluorescence intensity b of the pixel in the second fluorescence data xy =b xy 'exp[-E / (KΔC xy )],b xy ' is the fluorescence intensity of the pixel in the first fluorescence data, E is the fluorescence activation energy of the perilla leaf, K is the Boltzmann constant, x is the horizontal coordinate of the pixel, y is the vertical coordinate of the pixel, In step 3, three sets of polarization data are extracted from the polarized light data, and the pixel polarization degree of each pixel is calculated. The pixel points whose pixel polarization degree is greater than the polarization threshold constitute the reflection distortion area.

2. The method for identifying diseases of perilla leaves according to claim 1, characterized in that: In step 2, the signal acquisition device includes a light generator and a light receiver. At each detection station, the light generator generates broadband polarized light and ultraviolet excitation light in sequence, and the light receiver collects reflected polarization signals, fluorescence signals, and thermal radiation signals in sequence to generate polarization light data, first fluorescence data, and thermal imaging data.

3. The method for identifying diseases of perilla leaves according to claim 1, characterized in that: In step 4, the pixel point corresponding to the coordinate point of the perilla leaf plant in each reflection image is searched according to the image conversion relationship, and it is determined whether the pixel point corresponding to the coordinate point is in the reflection distortion area.

4. The method for identifying diseases of perilla leaves according to claim 1, characterized in that: In step 6, the first continuous atlas U1={A 11 ,...,A 1t ,...,A 1T }, T is the total acquisition period, A 1t is the reflection image of the acquisition period t, the second continuous atlas U2={A 21 ,...,A 2t ,...,A 2T }, A 2t is the fluorescence image of the acquisition period t, the third continuous atlas U3={A 31 ,...,A 3t ,...,A 3T }, A 3t is the thermal radiation image of the acquisition period t.

5. The method for identifying diseases of perilla leaves according to claim 1, characterized in that: In step 7, the first continuous atlas, the second continuous atlas, and the third continuous atlas of the same inspection station are substituted into the image segmenter to obtain the first abnormal atlas, the second abnormal atlas, and the third abnormal atlas. The first feature data is obtained based on the first abnormal atlas, the second feature data is obtained based on the second abnormal atlas, and the third feature data is obtained based on the third abnormal atlas.

6. The method for identifying diseases of perilla leaves according to claim 5, characterized in that: The first difference between the reflectance image in the first continuous atlas and the one-dimensional reflectance spectrum of the sample perilla leaf is calculated, and a first abnormal atlas is segmented from the first continuous atlas based on the first difference, wherein the first abnormal atlas consists of multiple first abnormal images, the second abnormal atlas consists of multiple second abnormal images, and the third abnormal atlas consists of multiple third abnormal images.

7. The method for identifying diseases of perilla leaves according to claim 6, characterized in that: The first abnormal atlas is clustered to obtain first characteristic data, the second abnormal atlas is clustered to obtain second characteristic data, and the third abnormal atlas is clustered to obtain third characteristic data.

8. An identification device for implementing the method for identifying diseases of perilla leaves according to claim 1, characterized in that: include: A workbench having a plurality of testing stations, wherein the perilla leaf plants are arranged in a sample area of ​​the workbench; A signal acquisition device, the signal acquisition device is used to collect polarized light data, first fluorescence data and thermal image data of the perilla leaf plant; A first controller, the first controller is used to move the signal acquisition device to different detection positions; A second controller, the second controller is used to switch the signal acquisition device to a reflected light detection state, a fluorescence detection state, and a thermal radiation detection state; An image processor, the image processor is used to generate reflection images, fluorescence images, and thermal radiation images of different acquisition cycles and different detection positions; An image segmenter, the image segmenter is used to identify the first continuous atlas set, the second continuous atlas set, and the third continuous atlas set; A data analyzer, the data analyzer is used to obtain the first characteristic data, the second characteristic data, and the third characteristic data; An image classifier is used to determine the types of diseases on perilla leaf plants.

Images