Winter sowing spring wheat field diagnosis method based on image recognition

Abstract

The invention relates to the technical field of crop growth state diagnosis, and provides a winter-sowing spring wheat field diagnosis method based on image recognition, and the method comprises the steps: collecting RGB images and near-infrared images of to-be-diagnosed wheat and healthy wheat leaves in the same growth period, and extracting a leaf region based on multi-space threshold segmentation; constructing a growth state quantitative calculation model based on the image monitoring data of the wheat leaves, inputting the image monitoring data acquired from the leaf areas of the wheat to be diagnosed and the reference sample into the growth state quantitative calculation model, and outputting a growth state result of the wheat to be diagnosed; the method can clearly judge whether the wheat is in a healthy state, and provides a clear decision basis for agricultural production management.

Description

Technical Field

[0001] The invention belongs to the technical field of crop growth status diagnosis, and in particular relates to a winter-sown spring wheat field diagnosis method based on image recognition. Background Art

[0002] The key growth periods of winter-sown spring wheat include: sowing period (late October to late November), overwintering period (late November to mid-March of the following year), emergence period (late March to early April), jointing and heading period (mid-to-late April to mid-to-early May), heading and flowering period (mid-to-late May) and filling and maturity period (early June to early July). Among them, the management before the jointing and heading period is relatively simple. As long as the one spray and three prevention measures are taken, large-scale diseases will generally not occur.

[0003] Wheat enters a rapid growth stage from jointing to booting, during which the demand for water and fertilizer increases significantly. It is also a period prone to various diseases and insect pests. Therefore, growth monitoring from jointing to booting is particularly important. Currently, chlorophyll meters (SPAD values), chlorophyll fluorescence meters and other equipment are commonly used in agricultural production to monitor wheat growth. However, the measurement process of these methods requires manual sampling, which is labor-intensive and easy to cause mechanical damage to crops. Single-point measurements are not representative enough to reflect the growth status of the entire field. Only a single physiological indicator can be obtained, which makes it difficult to fully reflect the growth status of the crop.

[0004] With the development of computer vision technology, crop growth diagnosis methods based on image analysis have gradually emerged, such as the vegetation index method based on RGB images. Patent CN112613338A discloses a method for estimating the nitrogen content of wheat leaf layer based on RGB image fusion features, which collects RGB images of wheat canopy and measures the nitrogen content of wheat leaf layer; first, RGB image preprocessing is performed to calculate the visible light vegetation index; the discrete wavelet transform method is used to realize multi-scale wavelet texture feature extraction in the horizontal, vertical and diagonal directions; the convolutional neural network is used to extract the deep features of RGB images; a particle swarm optimization support vector regression model based on fusion features is constructed to estimate the nitrogen content of wheat leaf layer; however, this method is easily affected by light conditions in the data collection step, the model has a large amount of calculation, and the prediction results are unstable; the intelligent diagnosis method of deep learning For example, patent CN118537731A discloses a multi-scale detection method for wheat yellow dwarf disease based on multi-spectral images of unmanned aerial vehicles, constructing a complete data set to be detected consisting of wheat yellow dwarf images; the wheat yellow dwarf images are multi-spectral images obtained by imaging different wheat varieties under complex backgrounds using spectral unmanned aerial vehicles; a dual-branch multi-scale model is constructed based on a dual-branch scale encoder and a lightweight decoder; wherein the dual-branch scale encoder includes a Transformer branch, a CNN branch and a feature fusion module for generating a multi-scale feature map; the wheat yellow dwarf disease area in the data set to be detected is detected by using the trained dual-branch multi-scale model, however, this method requires a large number of labeled samples, the model generalization ability is limited, although the recognition accuracy is good, it is difficult to promote in practice due to the high cost of the model.

[0005] Therefore, it is urgent to develop a simple, reliable and adaptable method for diagnosing wheat growth status to provide a scientific basis for agricultural production management. Summary of the invention

[0006] The purpose of the present invention is to provide a field diagnosis method for winter-sown spring wheat based on image recognition, and to construct a quantitative calculation model of the growth status based on the image monitoring data of wheat leaves. The diagnostic methods in the prior art are easily affected by the environment, have large calculation amount and high cost, etc., and the method realizes fast, reliable and economical field diagnosis, and achieves the purpose of wheat growth status diagnosis without relying on a large number of labeled samples and complex deep learning models.

[0007] In order to achieve the above-mentioned invention object, the specific technical scheme is as follows:

[0008] A field diagnosis method for winter-sown spring wheat based on image recognition, the diagnosis method is applicable to the seedling stage, jointing and booting stage, and heading and flowering stage, and the diagnosis method comprises the following steps:

[0009] Step S1, collecting RGB images and near-infrared images of wheat leaves to be diagnosed as first image data, taking healthy wheat in the same growth period as the wheat to be diagnosed as a reference sample, obtaining RGB images and near-infrared images of the reference sample as second image data.

[0010] Step S2, preprocessing the acquired first image data and second image data, converting the RGB image data into HSV and Lab space, and extracting the leaf area based on multi-space threshold segmentation.

[0011] Step S3, constructing a quantitative calculation model of growth status based on the image monitoring data of wheat leaves, wherein the image monitoring data includes: chlorophyll index, normalized vegetation index, texture value and similarity.

[0012] The mathematical expression of the growth state quantitative calculation model is:

[0013] Leaf condition index SI: Among them, w is the chlorophyll index weight value, and the initial value of the model is set to: w = 0.6; CI and CI s are the chlorophyll index of the wheat to be diagnosed and the reference sample, NI and NI s are the normalized vegetation index of the wheat to be diagnosed and the reference sample, TI is the texture value of the wheat to be diagnosed, and RI is the similarity between the wheat to be diagnosed and the reference sample.

[0014] Step S4: input the image monitoring data obtained from the leaf regions of the wheat to be diagnosed and the reference sample into a quantitative calculation model of the growth status, and output the growth status result of the wheat to be diagnosed.

[0015] The growth status results of the wheat to be diagnosed include:

[0016] When SI ≥ 0.85 and RI ≥ 0.8, the growth status of the wheat to be diagnosed is judged to be healthy;

[0017] When 0.6≤SI<0.85, if The growth status of the wheat to be diagnosed is determined to be nitrogen deficiency. Determining that the growth state of the wheat to be diagnosed is water stress;

[0018] When SI < 0.6 and When the growth state of the wheat to be diagnosed is determined to be disease infected.

[0019] Furthermore, the calculation method of the chlorophyll index CI of the wheat to be diagnosed is: Among them, G, R, and B are the average grayscale values ​​of the green, red, and blue channels, respectively.

[0020] The reference sample chlorophyll index CIs The calculation method is: Among them, G s , R s , B s They are the average grayscale values ​​of the green, red, and blue channels of the reference sample, respectively.

[0021] The calculation method of the normalized difference vegetation index NI of wheat to be diagnosed is: Among them, NIR is the near-infrared reflectance value of the wheat to be diagnosed, HR is the red light reflectance value of the wheat to be diagnosed, and NIRs is the near-infrared reflectance value of the reference sample.

[0022] The reference sample normalized vegetation index NI s The calculation method is: Among them, HR s is the red light reflectance value of the reference sample.

[0023] Furthermore, the method for calculating the texture value of the wheat to be diagnosed includes:

[0024] The RGB image is converted into a grayscale image, and the gray-level co-occurrence matrix GLCM is calculated, where the distance parameter d of the matrix is ​​1, and the direction angle θ is {0°, 45°, 90°, 135°}; the GLCM is normalized to obtain the probability matrix P(i, j).

[0025] Based on the probability matrix P(i, j), energy feature ASM, contrast feature CON, correlation feature COR and entropy feature ENT are extracted:

[0026] ASM=∑∑[P(i,j)] 2 , CON = ∑∑(ij) 2 P(i, j),

[0027] ENT=-∑P(i,j)log2∑P(i,j);

[0028] Among them, i, j are the gray values ​​of different points in the leaf area, μ x , μ y is the mean of the marginal distribution, σ x , σ y is the standard deviation of the marginal distribution.

[0029] Then, the texture value of wheat to be diagnosed is: Among them, ASMs, CONs, CORs, and ENTs are the energy characteristics, contrast characteristics, correlation characteristics, and entropy characteristics of the reference samples, respectively.

[0030] Furthermore, the calculation method of the similarity RI between the wheat to be diagnosed and the reference sample is: Where D is the Euclidean distance between the wheat to be diagnosed and the reference sample in the feature space:

[0031] σ is the adaptive scaling parameter, avg(D) is the average Euclidean distance between all samples in the training set and the reference samples.

[0032] Furthermore, the method for calculating the chlorophyll index weight value w comprises the following steps:

[0033] Step S31, collecting n groups of wheat sample data in different growth states to form a training set, wherein the wheat samples in different growth states at least include: samples in healthy state, nitrogen deficiency state, water stress state and disease infection state.

[0034] Step S32, calculate the chlorophyll index CI of each group of samples respectively n and Normalized Difference Vegetation Index NI n Relative value to reference sample:

[0035] Step S33, calculating the coefficient of variation CV of the chlorophyll index of each group of samples CI and the coefficient of variation of the normalized difference vegetation index CV NI :

[0036] Among them, std is the standard deviation function and mean is the mean function.

[0037] Step S34, chlorophyll index weight value: Verify the calculated weight value, if it satisfies: |SI 计算值 -SI 实测值 |<0.1, the calculated value is used to replace the initial setting value; otherwise, the chlorophyll index weight value w remains unchanged.

[0038] Further, the first image data and the second image data are preprocessed, and the leaf region is extracted based on multi-space threshold segmentation, including: using the Otsu algorithm for adaptive threshold segmentation, calculating the optimal threshold T, and using T as the segmentation reference of the RGB space:

[0039] T=arg max{ω0(t)×ω1(t)×[μ0(t)-μ1(t)] 2}, where ω0(t) and ω1(t) are the pixel ratios of the foreground and background regions, and μ0(t) and μ1(t) are the average grayscale values ​​of the foreground and background regions, respectively.

[0040] When performing threshold segmentation in HSV and Lab space, the hue H threshold range is set to [40°, 180°], the saturation S threshold range is set to [0.15, 1], and the brightness V threshold range is set to [0.15, 0.95] in the HSV space.

[0041] In the Lab space, the brightness L threshold range is set to [20, 100], the a channel threshold range is set to [-25, 0], and the b channel threshold range is set to [10, 50]; the segmentation results of the RGB, HSV and Lab spaces are ANDed to obtain the final leaf area segmentation mask.

[0042] Furthermore, the requirements for collecting RGB images and near-infrared images of the wheat leaves to be diagnosed and the reference samples are as follows:

[0043] 1) Under natural light conditions, the camera collects data vertically downward at a height of 40±5 cm;

[0044] 2) The light intensity should be between 30000-50000 lux;

[0045] 3) Three images were collected at each sample position and the average value was taken. The reference sample and the wheat to be diagnosed were collected under the same lighting conditions.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention constructs a comprehensive quantitative calculation model of growth status by combining multiple indicators such as chlorophyll index, normalized vegetation index, texture characteristics and similarity, which can more accurately reflect the growth status of crops compared with a single indicator; according to the different intervals of the leaf status index, it can clearly judge whether the wheat is in a healthy state, and whether it is nitrogen deficiency, water stress or disease infection in an unhealthy state, providing a clear decision-making basis for agricultural production management; the method of the present invention constructs a complete diagnostic system, and can also provide a technical reference for the diagnosis of the growth status of other crops. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of the winter-sown spring wheat field diagnosis method based on image recognition of the present invention; DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described implementation mode is a part of the present invention, not all implementation modes. Based on the implementation modes of the present invention, all other implementation modes obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] It should be noted that this method is applicable to the seedling stage, jointing and heading stage, and heading and flowering stage of winter-sown spring wheat, covering the period prone to diseases and insect pests, and is especially suitable for the jointing and heading stage. It can detect abnormal growth conditions in time, which can help farmers take corresponding prevention and control measures and reduce agricultural production risks.

[0051] like Figure 1 As shown, it is a flow chart of the winter-sown spring wheat field diagnosis method based on image recognition of the present invention, and the method comprises the following steps:

[0052] Step S1, collecting RGB images and near-infrared images of wheat leaves to be diagnosed as first image data, taking healthy wheat in the same growth period as the wheat to be diagnosed as a reference sample, obtaining RGB images and near-infrared images of the reference sample as second image data.

[0053] The requirements for collecting RGB images and near-infrared images of the wheat leaves to be diagnosed and the reference samples are as follows: the camera collects data vertically downward under natural light conditions, with a collection height of 40±5 cm to ensure the optimal balance between image resolution and field of view; the light intensity meets the requirement of 30,000-50,000 lux to avoid overexposure or underexposure affecting image quality; 3 images are collected at each sample position and the average is taken to reduce random errors, and the reference sample and the wheat to be diagnosed are collected under the same lighting conditions to eliminate interference from environmental factors.

[0054] Step S2, preprocessing the acquired first image data and second image data, converting the RGB image data into HSV and Lab space, and extracting the leaf area based on multi-space threshold segmentation.

[0055] Preprocessing includes: image quality assessment, checking image clarity, images with blur exceeding the threshold need to be re-captured, image exposure assessment to avoid overexposure or underexposure, detection and removal of possible motion blur; image standardization, image size is uniformly adjusted to 1920×1080 pixels, brightness standardization processing to eliminate the impact of uneven lighting, color balance correction to ensure color reproduction accuracy.

[0056] Extracting leaf regions based on multi-space threshold segmentation includes: using the Otsu algorithm for adaptive threshold segmentation, calculating the optimal threshold T, and using T as the segmentation benchmark in the RGB space:

[0057] T=arg max{ω0(t)×ω1(t)×[μ0(t)-μ1(t)] 2}, where ω0(t) and ω1(t) are the pixel ratios of the foreground and background regions, μ0(t) and μ1(t) are the average grayscale values ​​of the foreground and background regions, respectively; T, as the optimal threshold calculated by the Otsu algorithm, can automatically adjust the segmentation threshold according to the grayscale distribution of the image. There may be obvious differences in wheat leaf images taken under different lighting conditions and different growth environments, and a fixed threshold is difficult to adapt to such changes; by maximizing the inter-class variance (the difference between the foreground and background regions), the T value can adaptively find the best segmentation point.

[0058] When performing threshold segmentation in HSV and Lab space, the hue H threshold range is set to [40°, 180°], the saturation S threshold range is set to [0.15, 1], and the brightness V threshold range is set to [0.15, 0.95] in the HSV space.

[0059] In the Lab space, the brightness L threshold range is set to [20, 100], the a channel threshold range is set to [-25, 0], and the b channel threshold range is set to [10, 50]; the segmentation results of the RGB, HSV and Lab spaces are ANDed to obtain the final leaf area segmentation mask.

[0060] It should be noted that the AND operation of the three space segmentation results of RGB, HSV and Lab and the segmentation principle of the three spaces are as follows:

[0061] The optimal threshold T obtained by the Otsu algorithm is used in the RGB space for segmentation to obtain a binary image R(x, y); if the pixel gray value is greater than T, R(x, y) = 1, indicating the leaf area; if the pixel gray value is less than or equal to T, R(x, y) = 0, indicating the background area.

[0062] The image is segmented using three preset threshold ranges in the HSV space to obtain a binary image H(x, y), where H channel ∈ [40°, 180°], S channel ∈ [0.15, 1], and V channel ∈ [0.15, 0.95]; when the pixel value satisfies all three conditions at the same time, H(x, y) = 1, otherwise H(x, y) = 0.

[0063] In Lab space, three preset threshold ranges are used for segmentation to obtain a binary image L(x, y), where L channel ∈ [20, 100], a channel ∈ [-25, 0], and b channel ∈ [10, 50]; when the pixel value meets all three conditions at the same time, L(x, y) = 1, otherwise L(x, y) = 0.

[0064] The final leaf area mask M(x, y) is obtained by the AND operation of three binary images:

[0065] M(x, y)=R(x, y)∧H(x, y)∧L(x, y), that is, only when a pixel is judged as a leaf area in the segmentation results of the three spaces, the pixel will be marked as a leaf area in the final mask; through this multi-space joint segmentation strategy, the interference of shadows, weeds, etc. can be effectively reduced, and the accuracy and robustness of wheat leaf segmentation can be improved.

[0066] Step S3, constructing a quantitative calculation model of growth status based on the image monitoring data of wheat leaves, wherein the image monitoring data includes: chlorophyll index, normalized vegetation index, texture value and similarity.

[0067] The mathematical expression of the growth state quantitative calculation model is:

[0068] Leaf condition index SI: Among them, w is the chlorophyll index weight value, and the initial value of the model is set to: w = 0.6; CI and CI s are the chlorophyll index of the wheat to be diagnosed and the reference sample, NI and NI s are the normalized vegetation index of the wheat to be diagnosed and the reference sample, TI is the texture value of the wheat to be diagnosed, and RI is the similarity between the wheat to be diagnosed and the reference sample.

[0069] The calculation method of the chlorophyll index CI of the wheat to be diagnosed is: Among them, G, R, and B are the average grayscale values ​​of the green, red, and blue channels, respectively.

[0070] The reference sample chlorophyll index CI s The calculation method is: Among them, G s , R s , B s They are the average grayscale values ​​of the green, red, and blue channels of the reference sample, respectively.

[0071] The calculation method of the normalized difference vegetation index NI of wheat to be diagnosed is: Among them, NIR is the near-infrared reflectance value of the wheat to be diagnosed, HR is the red light reflectance value of the wheat to be diagnosed, and NIRs is the near-infrared reflectance value of the reference sample.

[0072] The reference sample normalized vegetation index NI s The calculation method is: Among them, HR s is the red light reflectance value of the reference sample.

[0073] The calculation method of the texture value of the wheat to be diagnosed includes:

[0074] Convert the RGB image to a grayscale image and calculate the gray-level co-occurrence matrix GLCM, where the distance parameter d=1 and the direction angle θ={0°, 45°, 90°, 135°} of the matrix; normalize the GLCM to obtain the probability matrix P(i, j);

[0075] Extracting energy feature ASM, contrast feature CON, correlation feature COR and entropy feature ENT based on the probability matrix P(i, j);

[0076] ASM=∑∑[P(i,j)] 2 , CON = ∑∑(ij) 2 P(i, j), ENT=-∑P(i,j)log2∑P(i,j); where i,j are the gray values ​​of different points in the leaf area, μ x , μ y is the mean of the marginal distribution, σ x , σ y is the standard deviation of the marginal distribution;

[0077] Then, the texture value of wheat to be diagnosed is: Among them, ASMs, CONs, CORs, and ENTs are the energy characteristics, contrast characteristics, correlation characteristics, and entropy characteristics of the reference samples, respectively.

[0078] The calculation method of the similarity RI between the wheat to be diagnosed and the reference sample is: Where D is the Euclidean distance between the wheat to be diagnosed and the reference sample in the feature space, σ is the adaptive scaling parameter, avg(D) is the average Euclidean distance between all samples in the training set and the reference samples.

[0079] The method for calculating the chlorophyll index weight value w comprises the following steps:

[0080] Step S31, collecting n groups of wheat sample data in different growth states to form a training set, wherein the wheat samples in different growth states include at least: samples in healthy, nitrogen-deficient, water-stressed and disease-infected states; the number of training set samples is recommended to be no less than 100 groups.

[0081] Step S32, calculate the chlorophyll index CI of each group of samples respectively n and Normalized Difference Vegetation Index NI n Relative value to reference sample:

[0082] Step S33, calculating the coefficient of variation CV of the chlorophyll index of each group of samples CI and the coefficient of variation of the normalized difference vegetation index CV NI :

[0083] Among them, std is the standard deviation function, mean is the mean function; the use of coefficient of variation can eliminate the dimension effect.

[0084] Step S34, chlorophyll index weight value: Verify the calculated weight value, if it satisfies: |SI 计算值 -SI 实测值 |<0.1, the calculated value is used to replace the initial setting value; otherwise, the chlorophyll index weight value w remains unchanged.

[0085] Step S4, inputting the image monitoring data obtained from the leaf regions of the wheat to be diagnosed and the reference sample into the growth status quantitative calculation model, and outputting the growth status result of the wheat to be diagnosed;

[0086] The growth status results of the wheat to be diagnosed include:

[0087] When SI ≥ 0.85 and RI ≥ 0.8, the growth status of the wheat to be diagnosed is judged to be healthy;

[0088] When 0.6≤SI<0.85, if The growth status of the wheat to be diagnosed is determined to be nitrogen deficiency. Determining that the growth state of the wheat to be diagnosed is water stress;

[0089] When SI < 0.6 and When the growth state of the wheat to be diagnosed is determined to be disease infected.

[0090] The SI threshold was set at 0.85 mainly considering that healthy wheat leaves have stable spectral and morphological characteristics, and the chlorophyll content and moisture content are within the normal range with fluctuations not exceeding 15%. Taking into account measurement errors and natural variations, 0.85 is taken as the lowest limit of the healthy state.

[0091] The RI threshold was set at 0.8 because the distance between healthy wheat and the reference sample in the feature space was close, and a similarity of 0.8 indicated that the Euclidean distance in the feature space was within an acceptable range, and this threshold could effectively distinguish normal physiological fluctuations from pathological changes.

[0092] The basis for determining the state of nutrient and water stress (0.6≤SI<0.85), the range of 0.6-0.85 reflects a mild to moderate physiological stress state, and wheat within this range can be restored to a healthy state through appropriate management measures. The determination of nitrogen deficiency (CI / CIs<0.8) is based on the fact that a decrease in the chlorophyll index is a direct manifestation of nitrogen deficiency. The threshold of 0.8 is based on a statistical analysis of leaf color changes under nitrogen stress, taking into account the variability of different varieties and growth stages. The determination of water stress (NI / NIs<0.8) mainly considers that the normalized vegetation index is sensitive to changes in water content. The threshold of 0.8 reflects a significant water stress state, which takes into account the need for early warning of drought stress.

[0093] The basis for determining the disease infection status (SI < 0.6 and TI / TIs < 0.7) is that in production management, it is found that an SI below 0.6 usually indicates irreversible pathological changes. This threshold can detect serious disease infections early. The TI / TIs threshold is set at 0.7. Diseases can cause significant changes in leaf texture characteristics. Texture similarity below 0.7 indicates that lesions or necrotic areas are obvious. This threshold can distinguish between disease symptoms and mechanical damage.

[0094] Taking the diagnosis of the jointing and heading stage of winter-sown spring wheat on a farm as an example, the collection conditions are: 10 am, sunny day, light intensity 42000 lux, collection height 38 cm, and the reference sample is healthy wheat of the same period.

[0095] The image processing results show that the RGB space 0tsu threshold is T=127; the leaf segmentation accuracy is 95.3%. The feature calculation results are: CI=0.62, CIs=0.85; NI=0.71, NIs=0.88; TI=0.83, RI=0.76; calculated SI=0.73; the diagnostic result obtained is: since 0.6≤SI<0.85 and CI / CIs=0.73<0.8, it is judged to be nitrogen deficiency.

[0096] After nitrogen fertilizer was applied to the plot, the leaf color improved significantly within a week and the SI value rose to 0.88, verifying the accuracy of the diagnostic results. The practical application of this method shows that it can detect abnormal wheat growth in a timely manner and provide a decision-making basis for precision agricultural management.

[0097] The following points should also be noted during use: regularly calibrate reference sample data, establish localized threshold parameters, keep the camera lens clean, and pay attention to the selection of weather conditions.

[0098] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A winter-sown spring wheat field diagnosis method based on image recognition, characterized in that: The diagnostic method comprises the following steps: Step S1, collecting an RGB image and a near-infrared image of a wheat leaf to be diagnosed as first image data, taking healthy wheat in the same growth period as the wheat to be diagnosed as a reference sample, obtaining an RGB image and a near-infrared image of the reference sample as second image data; Step S2, preprocessing the acquired first image data and second image data, converting the RGB image data into HSV and Lab space, and extracting the leaf area based on multi-space threshold segmentation; Step S3, constructing a quantitative calculation model of growth status based on image monitoring data of wheat leaves, wherein the image monitoring data includes: chlorophyll index, normalized vegetation index, texture value and similarity; The mathematical expression of the growth state quantitative calculation model is: Leaf condition index SI: Among them, w is the chlorophyll index weight value, and the initial value of the model is set to: w = 0.6; CI and CI s are the chlorophyll index of the wheat to be diagnosed and the reference sample, NI and NI s are the normalized vegetation index of the wheat to be diagnosed and the reference sample, TI is the texture value of the wheat to be diagnosed, and RI is the similarity between the wheat to be diagnosed and the reference sample; Step S4, inputting the image monitoring data obtained from the leaf regions of the wheat to be diagnosed and the reference sample into the growth status quantitative calculation model, and outputting the growth status result of the wheat to be diagnosed; The growth status results of the wheat to be diagnosed include: When SI ≥ 0.85 and RI ≥ 0.8, the growth status of the wheat to be diagnosed is judged to be healthy; When 0.6≤SI<0.85, if The growth status of the wheat to be diagnosed is determined to be nitrogen deficiency. Determining that the growth state of the wheat to be diagnosed is water stress; When SI < 0.6 and When the growth state of the wheat to be diagnosed is determined to be disease infected.

2. The winter-sown spring wheat field diagnosis method based on image recognition according to claim 1, characterized in that: The calculation method of the chlorophyll index CI of the wheat to be diagnosed is: Among them, G, R, and B are the average grayscale values ​​of the green, red, and blue channels respectively; The reference sample chlorophyll index CI s The calculation method is: Among them, G s , R s , B s are the average grayscale values ​​of the green, red, and blue channels of the reference sample, respectively; The calculation method of the normalized difference vegetation index NI of wheat to be diagnosed is: Among them, NIR is the near infrared reflectance value of the wheat to be diagnosed, HR is the red light reflectance value of the wheat to be diagnosed, and NIRs is the near infrared reflectance value of the reference sample; The reference sample normalized vegetation index NI s The calculation method is: Among them, HR s is the red light reflectance value of the reference sample.

3. The winter-sown spring wheat field diagnosis method based on image recognition according to claim 2 is characterized in that: The calculation method of the texture value of the wheat to be diagnosed includes: Convert the RGB image to a grayscale image and calculate the gray-level co-occurrence matrix GLCM, where the distance parameter d=1 and the direction angle θ={0°, 45°, 90°, 135°} of the matrix; normalize the GLCM to obtain the probability matrix P(i, j); Extracting energy feature ASM, contrast feature CON, correlation feature COR and entropy feature ENT based on the probability matrix P(i, j); ASM=∑∑[P(i,j)] 2 ,CON=∑∑(i-j) 2 P(i,j), ENT=-∑P(i,j)log2∑P(i,j); where i,j are the gray values ​​of different points in the leaf area, μ x ,μ y is the mean of the marginal distribution, σ x ,σ y is the standard deviation of the marginal distribution; Then, the texture value of wheat to be diagnosed is: Among them, ASMs, CONs, CORs, and ENTs are the energy characteristics, contrast characteristics, correlation characteristics, and entropy characteristics of the reference samples, respectively.

4. The winter-sown spring wheat field diagnosis method based on image recognition according to claim 3 is characterized in that: The calculation method of the similarity RI between the wheat to be diagnosed and the reference sample is: Where D is the Euclidean distance between the wheat to be diagnosed and the reference sample in the feature space, ; σ is the adaptive scaling parameter, avg(D) is the average Euclidean distance between all samples in the training set and the reference samples.

5. The winter-sown spring wheat field diagnosis method based on image recognition according to claim 4 is characterized in that: The method for calculating the chlorophyll index weight value w comprises the following steps: Step S31, collecting n groups of wheat sample data in different growth states to form a training set, wherein the wheat samples in different growth states at least include: samples in healthy, nitrogen-deficient, water-stressed and disease-infected states; Step S32, calculate the chlorophyll index CI of each group of samples respectively n and Normalized Difference Vegetation Index NI n Relative value to reference sample: Step S33, calculating the coefficient of variation CV of the chlorophyll index of each group of samples CI and the coefficient of variation of the normalized difference vegetation index CV NI : Among them, std is the standard deviation function, mean is the mean function; Step S34, chlorophyll index weight value: Verify the calculated weight value, if it satisfies: |SI 计算值 -SI 实测值 |<0.1, the calculated value is used to replace the initial setting value; otherwise, the chlorophyll index weight value w remains unchanged.

6. The winter-sown spring wheat field diagnosis method based on image recognition according to claim 5, characterized in that: Preprocessing the acquired first image data and second image data, extracting the leaf area based on multi-space threshold segmentation includes: using Otsu algorithm for adaptive threshold segmentation, calculating the optimal threshold T, and using T as the segmentation benchmark of RGB space: T=arg max{ω0(t)×ω1(t)×[μ0(t)-μ1(t)] 2 }, where ω0(t) and ω1(t) are the pixel ratios of the foreground and background regions, μ0(t) and μ1(t) are the average grayscale values ​​of the foreground and background regions, respectively; When performing threshold segmentation in HSV and Lab space, the hue H threshold range is set to [40°, 180°], the saturation S threshold range is set to [0.15, 1], and the brightness V threshold range is set to [0.15, 0.95] in the HSV space; In the Lab space, the brightness L threshold range is set to [20, 100], the a channel threshold range is set to [-25, 0], and the b channel threshold range is set to [10, 50]. The segmentation results of the RGB, HSV, and Lab spaces are ANDed to obtain the final leaf region segmentation mask.

7. The winter-sown spring wheat field diagnosis method based on image recognition according to any one of claims 1 to 6, characterized in that: The requirements for collecting RGB images and near-infrared images of the wheat leaves to be diagnosed and the reference samples are as follows: the camera collects data vertically downward under natural light conditions, the collection height is 40±5 cm, and the light intensity meets the requirements of 30,000-50,000 lux; 3 images are collected at each sample position and the average value is taken, and the reference sample and the wheat to be diagnosed are collected under the same lighting conditions.

Images